Table of contents

This article reviews the state-of-the-art in remote bonding of fiber Bragg grating sensors, primarily for Lamb wave measurements in structures. The presence of damage in a structure modifies Lamb waves through reflection and scattering, as well as the potential conversion between Lamb modes. While FBG sensors have been applied to capture ultrasonic waves for the past 30 years, the mounting configuration called remote bonding has recently come to attention as an alternative method to capture guided waves from structures. In this case, the FBG is not located at the bond, but instead at a remote location along the optical fiber. The Lamb waves are converted into propagating acoustic waves along the fiber, which are measured by the FBG. This article presents a discussion of the primary benefits of remote bonding, including the higher sensitivity to low-amplitude Lamb waves, the fact that the FBG can situated in a less harsh environment than the sensing region, and the insensitivity to quasi-static strain. The properties of ultrasonic modes in optical fibers and their conversion from Lamb waves is first reviewed. Strategies to detect these waves with FBGs and the associated instrumentation is also presented. Recent examples from the literature utilizing remote bonded FBGs are then presented. Finally, acoustic couplers to transfer the ultrasonic modes from one or more optical fibers to another are also reviewed.

Optical flow methods have been developed over the past two decades for application to particle image velocimetry (PIV) images with the goal of acquiring higher resolution measurements of the velocity field than conventional cross-correlation (CC)-based techniques. Numerous optical flow velocimetry (OFV) algorithms have been devised to solve the ill-posed optical flow problem, with various physics-inspired strategies to tailor them to fluid flows. While OFV can be applied to continuous scalar fields, it has demonstrated the most success on images of tracer particles, i.e. traditional planar PIV images. Compared to state-of-the-art CC algorithms, OFV methods have demonstrated an order of magnitude increase in spatial resolution and up to a factor of two improvement in overall accuracy when evaluated on synthetic data, at the cost of increased computational time. The requirements for particle seeding density, inter-frame displacement, and image quality are also more stringent for OFV methods compared to CC. OFV has been applied sparingly in experiments to date, but appears to offer the same advantages demonstrated on synthetic data. At this stage, OFV seems best suited to planar velocity measurements, although extensions to stereoscopic measurements have been demonstrated.

This paper presents a comprehensive review of recent advancements in bearing health monitoring and remaining useful life (RUL) prediction. It highlights key innovations in anomaly detection, health indicator construction, degradation modeling, and RUL estimation, examining developments across statistical, machine learning, and deep learning approaches while analyzing their strengths, limitations, and application contexts. Special emphasis is placed on the role of deep learning in capturing complex degradation patterns from multi-dimensional time series data and improving predictive accuracy in dynamic industrial settings. Additionally, this review explores multi-source data fusion techniques, which enhance anomaly detection robustness by integrating information from diverse sensor modalities. By identifying critical challenges and suggesting future research directions, this study aims to advance the development of robust and adaptive prediction models for intelligent maintenance in industrial applications.

Large volumes of light sources and their characteristics control devices are challenges to out-of-laboratory applications of quantum precision measurement devices, such as spin-exchange relaxation-free (SERF) inertial and extremely weak magnetic field measurement devices. High-Q whispering gallery mode (WGM) resonators have excellent optical field confinement capability, extremely narrow linewidth, and small mode volume, which can easily realize compact low noise and high-frequency stability light sources, having a great potential to realize more compact quantum precision measurement devices. This will broaden the use of quantum precision measurement devices in areas such as transportable SERF gyroscopes, wearable magnetometers, etc. In this review, the fundamentals and characteristics of WGM optical cavities are firstly introduced, as well as its fabrication techniques and built-up materials. Then, the laser stabilizing applications of WGM optical cavities including wavelength tuning, self-injection locking, and thermal stabilization are discussed. Finally, the application prospects of quantum precision measurement are discussed.

Advanced three-dimensional (3D) tracking methods are essential for studying particle dynamics across a wide range of complex systems, including multiphase flows, environmental and atmospheric sciences, colloidal science, biological and medical research, and industrial manufacturing processes. This review provides a comprehensive summary of 3D particle tracking and flow diagnostics using digital holography (DH). We begin by introducing the principles of DH, accompanied by a detailed discussion on numerical reconstruction. The review then explores various hardware setups used in DH, including inline, off-axis, and dual or multiple-view configurations, outlining their advantages and limitations. We also delve into different hologram processing methods, categorized into traditional multi-step, inverse, and machine learning (ML)-based approaches, providing in-depth insights into their applications for 3D particle tracking and flow diagnostics across multiple studies. The review concludes with a discussion on future prospects, emphasizing the significant role of ML in enabling accurate DH-based particle tracking and flow diagnostic techniques across diverse fields, such as manufacturing, environmental monitoring, and biological sciences.

This study introduces a compact suspended microstrip low-pass filter for millimeter-wave applications, leveraging an I/O coupling structure to achieve a wide stopband and a high roll-off rate (ROR). The filter's I/O coupling lines create additional transmission zeros (TZs) outside the passband, allowing precise control over these zeros' placement and significantly enhancing stopband performance. With six TZs in the stopband, four can be flexibly tuned without affecting the ROR and passband width, enabling adaptable stopband bandwidth. The filter demonstrates a −3 dB cutoff at 42 GHz, an 81 GHz stopband (with 20 dB suppression), and a ROR of 5.29 dB GHz⁻¹ representing an eightfold improvement over conventional filters in this frequency range. Additionally, the filter maintains exceptional performance in the passband, with shallow insertion loss (IL $\lt$ 0.2 dB) and high return loss (RL $\gt$ 25 dB), and the overall size of the filter is 1.35 × 2.54 mm, all within a compact design ideal for high-performance millimeter-wave applications.

Wire-arc additive manufacturing (WAAM) is acknowledged as a highly effective and economical 3D printing method for large component production and repair. However, defects like porosity, oxidization, and distortion may occur due to the high deposition rate and improper manufacturing process, leading to a vital concern about the structural quality. Acoustic emission (AE) signals generated with the WAAM process carry a wealth of manufacturing information, but the interpretation of AE signals often requires a deep understanding of both the physics of the AM process and the characteristics of the AE signal. This interdisciplinary knowledge gap can hinder the development of accurate and reliable signal analysis models. To reveal the manufacturing physics-related AE features from the complicated broadband AE signals, a spectral decomposition and characterization method is proposed and investigated for WAAM process monitoring. The AE hits are first identified with the average filtering and envelope peak finding so that the intense noise can be suppressed and AE hits can be highlighted. Secondly, the wavelet entropy and spectral centroid are extracted for each AE hit to characterize the printing status with the AE waveform. Then, the AE features varying with the manually controlled printing status are investigated and fused with experimental study, thus improving the identification accuracy. The experimental investigation on four coupons manufactured with shielding gas flow control shows that the spectral features are sensitive to the WAAM process, thereby providing a promising candidate for real-time monitoring and evaluation of WAAM.

This study proposes a non-contact force-measuring sensor design that utilizes a polymer optical fiber. The sensor design uses two separate fibers (i.e. sensor input and sensor output fibers). The sensor input fiber is helically wound around a supportive circular pipe, with aim of creating bend losses at every bend loop. The sensor output fiber gains advantages by coupling the variable optical power from different bend loops of the sensor input fiber. The variable amount of optical power coupling is directly associated with the applied force on the spring, which serves as an intermediary in transferring force to the sensor. Further, for variable ranges of forces, three different springs are used, which sustain variable force ranges (0–20, 0–40 and 0–60 N), respectively. The proposed sensor design addresses the critical need for force sensors capable of effectively measuring force across a wide dynamic range. The results of this study support the sensor design, which has a wide range of force-measuring capabilities, from low to high, with a close and repeatable response. The low-range force measurement (0–20 N) indicates the sensor's high sensitivity of 0.567 µW N⁻¹. Further, the sensor is also capable of a large force measuring range (0–60 N) with sensitivity of 0.189 µW N⁻¹.

The objectives in traffic sign detection and recognition scenario are predominantly small, which frequently result in missed and erroneous detection due to their limited information content and complex environment. To address these problems, this paper proposes a new network architecture cross-dimensional and dual-domain feature fusion-you only look once (CDFF-YOLO) which is integrated of various modules. For the purpose of overcoming the difficulty in extracting information from small objects, the Multi-dimension Spatial information Fusion module in the network are used to extract feature sequences at different dimensions by superposition. So as to address the issue of the loss of detail information of small objects, embed the Multi-branch Perceptual Attention module into the C2f module to capture feature information and enhance the global-local feature information exchange. In order to solve the issue of uneven illumination and occlusion in the detection scene. The DFF module is employed to transform and fuse the extracted small object feature information from the Space-to-depth convolution at the frequency and spatial domains, thereby enhancing the network's capability to reconstruct and fuse feature information in dual-domains. The experimental data on the TT100K dataset demonstrate that the enhanced algorithm exhibits an increase of 3.7% in mAP@50, 4.8% in mAP@50:95, and a 4.5% and 3.7% rise in the average precision and average recall for small objects, respectively. Additionally, the frames per second remains 157. The improved algorithm also performs well on the CCTSDB dataset. It is evident that the CDFF-YOLO algorithm has the capacity to markedly enhance the detection efficacy of traffic signs, while maintaining optimal detection speed.

Multiple refractions between different media lead to camera refractive distortions, thereby making it difficult to ensure underwater measurement accuracy. Consequently, precise modeling and calibration of underwater multilayer imaging systems is essential to address these distortions. In this paper, employing a perspective projection scale factor along the light propagation direction, a universal forward multilayer refractive model is presented. On this basis, the refractive model incorporates a general rotation matrix, which not only enriches application diversity but also enhances solution precision. Allowing for direct underwater camera calibration without prior calibration in air, the model simplifies the calibration process and enhances its suitability for practical applications. Through underwater calibration experiments, the refractive model enhances solution efficiency, improves identification precision and distributes evenly and mediately in terms of reprojection error. In addition, it increases the number of identifiable parameters, demonstrating its robustness and superiority in complex parameter identification scenarios. Finally, underwater measurement experiments indicate that the absolute accuracy calibrated and measured by the refractive model is better than 0.25 mm. Overall, combining all the benefits of forward and backward compensation methods, the forward refractive model offers versatility in handling multiple refractions without limited refractive layers.

As an important part of the hydraulic system, hydraulic oil pollution will seriously affect the performance and the lifetime of the hydraulic component. Therefore, the pollution of hydraulic oil is crucial to the lifetime of the hydraulic component. In this study, an assessment model with internal leakage flow as a parameter is proposed. A Life assessment hydraulic test system is built. The test is conducted using hydraulic fluids containing test dust. It evaluates the effect of the contaminated fluid on the degradation of the internal leakage performance and reduction of lifetime. The results show that the internal leakage flow of the hydraulic directional valve, classified under ISO 4406 cleanliness level 21/19/17, increased from 6 ml to 68 ml following a test lasting 120 h. It verifies the lifetime based on the internal leakage flow is described by a logarithmic function. The hydraulic directional valve has a pressure loss of 0.8 MPa higher in port A than in port B at an operating pressure of 15 MPa. This paper provides a reference for further study on the effect of wear due to internal leakage failure on lifetime and improvement of service life of hydraulic directional valves.

Acquiring high-precision and robust calibration methods for determining the intrinsic and extrinsic parameters of cameras is essential to ensuring the accuracy of visual measurements and three-dimensional reconstruction. This paper addresses the challenge of reduced calibration accuracy in camera systems caused by eccentricity errors in the extraction of circular target feature centers. We propose a precise calibration method that corrects these eccentricity errors associated with circular targets. The method begins by solving for the homography matrix of each calibration image using the coordinates of both image points and target points. Subsequently, the homography matrix is used to generate the frontal view of the circular target, enabling the precise extraction and back-mapping of the circular target's center based on this frontal view. Finally, using the corrected coordinates of the circle center and the calibration method proposed by Zhang in the paper 'A Flexible New Technique for Camera Calibration', we accurately determine the intrinsic and extrinsic parameters of the camera. Simulation experiments show the proposed calibration method aligns perfectly with ground truth without noise. In practical tests, it outperformed the checkerboard and traditional circular target methods, reducing reprojection errors by 0.0295 and 0.0176 pixels, respectively. Additionally, the cross-ratio experiment showed a 0.076% reduction in error for corrected circle center points. These results highlight significant improvements in calibration accuracy, meeting high-precision needs in visual sensing.

When using transmission methods to measure the morphology of cylindrical lenses, the refraction of light at the front and rear surfaces of cylindrical lenses often makes it difficult to determine the light path. To address this challenge, this paper proposes a phase deflection method based on transmission to measure the surface morphology of cylindrical lenses. This method involves calculating the displacement of points with the same phase before and after the cylindrical lens is placed in the light path, and then deriving the corresponding thickness of the cylindrical lens at these points from the displacement. The feasibility of this method has been verified through simulations. Experimental measurements of the cylindrical lens morphology were compared with the optical coherence tomography system, yielding a standard error of 5.4 μm and a root mean square error of 6.2 μm, demonstrating the effectiveness of the proposed method. Compared to existing methods, this approach significantly reduces the impact of refraction caused by the rear surface of the cylindrical lens after light transmission. Additionally, it eliminates the need for a camera lens, thereby removing the cumbersome step of camera calibration. Furthermore, interference fringes can be made very dense, enabling precise measurement of small lenses.

The refractive index is an important physical characteristic that reflects material information. In this paper a refractive index measurement method based on surface plasmon resonance (SPR) with optical fiber excitation structure and phase-detection using a dual-frequency laser interferometer is presented. A four-layer fiber SPR model is built based on Kretschmann structure. Theoretical analyses indicate that the phase difference variation between p and s polarization components of the reflected light is approximately linear with the refractive index in a certain range, and thus the calculation formulae are derived. The presented optical measurement system meets the principle of the common optical path to get good anti-interference ability and stability. Signal processing adopts phase detection, which avoids the influence of light intensity fluctuation and achieves high resolution. The refractive index measurement experiments show that the results are consistent with the theoretical analysis. Comparison results with the Abbe refractometer show a good agreement. The presented method is a new attempt to combine fiber SPR with laser heterodyne interference phase measurement.

The diaphragm pump is a vital component in the online oil chromatography device for oil–gas separation, and its performance plays a crucial role in determining the accuracy and reliability of gas detection. However, over time, due to continuous use and external environmental factors, the performance of the diaphragm pump gradually declines. This degradation results in an insufficient vacuum during the degassing process, which adversely affects the accuracy of gas detection. To tackle this issue, this paper combines simulation and experimental approaches to thoroughly investigate the causes of diaphragm pump performance degradation and its theoretical lifespan. It also examines the impact of insufficient vacuum on the detection accuracy of various gases. In addition, by analyzing the current variation patterns of the diaphragm pump during operation, a multilayer perceptron (MLP) algorithm is applied to create a fault diagnosis and data error correction model for the diaphragm pump. Experimental results demonstrate that when the vacuum level surpasses 70%, the MLP model is capable of keeping the correction error for seven types of fault gases below 10%, significantly enhancing detection accuracy. This study offers valuable insights for the reliability assessment and performance optimization of online oil chromatography devices, helping to reduce maintenance costs and improve the safe operation of power transformers.

Laser trackers or total stations are most widely used in particle accelerator measurement with high measurement accuracy but low efficiency. Close range photogrammetry has great characteristics of both high accuracy and efficiency, other than that, it enables non-contact observation, which is ideal for high-radiation accelerator measurement environments. In order to obtain the absolute position of the measurement area using close range photogrammetry, the position and orientation of the camera and the scale of the photo must be accurately determined at the moment of observation. A new three-dimensional visual measurement equipment is developed in this paper, which can obtain image, distance, and angle data at the same time. Unified adjustment of all three types of observations can more accurately determine position, orientation and scale datum. We used simulated data for experimental validation, comparing the new algorithm with another two algorithms: the traditional fixed point constraint and centroid datum constraint. The experiment results show that it can significantly improve positioning accuracy by adding distance and angle observations. The three-dimensional coordinate accuracy can reach up to 0.073 mm, while the accuracy of the other two algorithms can only be achieved 0.137 and 0.153 mm, the improvement is approximately 52.4%. Especially in the forward direction of the tunnel, the positioning accuracy is significantly improved, with an improvement of about 67.8 %.

In fringe projection profilometry, a method for solving 3D information by combining Gray code fringe images with phase-shifting fringe images has been widely used. In practice, there may still be phase mismatch between wrapped phase and complementary Gray code arrays. Based on the above problems, this paper proposes an improved complementary Gray code combined with phase-shifting profilometry (PSP) based on phase adjustment. First, the projected images are calibrated in advance, that is, the initial phase (image order) of the phase-shifting fringe images are adjusted so that the wrapped phase and the Gray code arrays match each other, and this image order is called the best order. Then, in the following practical measurements, the phase-shifting fringe images are substituted into the phase-shifting equation in accordance with this best order to obtain the wrapped phase. Finally, phase unwrapping is performed using the complementary Gray code phase unwrapping method to obtain the final unwrapped phase and 3D information from the phase-height mapping method. Compared to the traditional complementary Gray code combined with PSP, the proposed method can effectively improve the measurement quality without adding new projection images. Experiments show that the proposed method can improve the root mean square error (RMSE) and mean absolute error (MAE) accuracy by 89.63% and 89.76%, respectively, compared with the traditional complementary Gray code combined with the phase-shifting method (which has phase mismatch at this time), which improves the RMSE and MAE accuracy by −10.11% and −8.53%, respectively, and improves the efficiency compared with multi-frequency temporal phase unwrapping.

In this work, an octagonal shaped CSRR has been modified to form a nested structure for achieving a better field-material interaction. This modification results in an increase in the electrical length of the resonator providing higher sensitivity while accurately determining the complex permittivity of unknown samples. The resonator has been designed using high frequency structure simulator software with its structural dimensions chosen to operate at 2.4 GHz of the ISM band. It has been fabricated on a low-cost FR4 substrate where the nested CSRR structure is etched on the ground plane. A lumped equivalent circuit model of the proposed design is further developed using advanced design system software for the validation of the operating principle of the sensor. Some commonly available liquids covering a wider range of dielectric constants have been used to develop an equation between the permittivity parameters of the liquid and the transmission coefficients of the resonator. The sensor has effectively predicted both the real and imaginary parts of permittivity of toluene and aniline used as unknown samples with accuracy as high as 97%. The average and normalized sensitivity of the proposed sensor have been found to be 78.34 MHz/' and 3.273% respectively. A comparison of this work with other recently reported works clearly indicates the superiority of this work and its potential to be used as a dielectric sensor.

In view of the problems that robotic arms find it difficult to effectively identify and grasp workpieces with different textures and hardness in complex industrial environments, and the low path planning accuracy of robotic arms in practical application scenarios, this paper proposes a composite sensor based on piezoresistive effect and triboelectric effect. The composite sensor simultaneously generates piezoresistive and triboelectric signals. By comparing the output signals of the sensor, it is possible to achieve high-precision dynamic monitoring of the flexible grasping and joint movement processes of the robotic arm. The base material of the piezoresistive sensor and the negative electrode material of the triboelectric sensor are both porous MXene/PDMS structures. The piezoresistive sensor enhances its conductivity by spin-coating CNT slurry on the surface of the base material and uses PVA hydrogel as the electrode. The triboelectric sensor uses copper as the positive electrode material. Experiments show that the developed sensor has a measurement range (0.015–70 kPa) and good repeatability. The experiment verifies that the composite sensor can be applied to the high-precision detection of robotic arm's flexible gripping and joint movements.

Deep learning-based soft sensors often encounter the challenge of black-box problems, which lack interpretability and fail to provide intuitive mapping relationships. In complex industrial processes, physical sensors are susceptible to wear and tear, causing a divergence between test data distributions and training data, thus undermining sensing performance stability. To address these challenges, this paper proposes a hybrid-driven and self-interpretable soft sensor with symbolic representation, called the Kolmogorov–Arnold conditional autoencoder (KACAE). This framework employs symbolic representation to intuitively express the mapping relationships between variables, providing the model's self-interpretability. It also incorporates domain knowledge to determine specific activation function types and simplify the network structure, achieving an effective balance between performance and computational complexity. Furthermore, an adaptive weighted mechanism is introduced to eliminate correlations between hidden features, enhancing model stability. Finally, the effectiveness, self-interpretability, and stability of the KACAE are validated through two industrial case studies.

Pressure-sensitive paint (PSP) is an optical pressure-measurement technique that uses oxygen quenching. In unsteady PSP measurements, PSPs necessitate not only pressure sensitivity and response speed but also higher luminescence intensity. In this study, spherical silica nanoparticles are mixed into the paint to achieve a brighter luminescence. We prototype PSPs using particle size and mixing ratio as parameters and evaluate their performance. The results exhibit the luminescence intensity changes due to the low refractive index of silica; the smaller the particle size and the higher the particle ratio, the stronger the luminescence intensity. Silica-based PSPs are approximately 1.5 times brighter than titania-based PSP; which have frequently been used in previous studies. Additionally, a comprehensive performance investigation indicates that the silica-based PSP with the brightest formulation has a pressure sensitivity of 0.72 %/kPa with a response speed of 11.0 µs. These results prove that the higher luminescence is compatible with the high sensitivity and fast response using low-refractive index nanoparticles.

As a quintessential electrostatic excitation micro electro mechanical system device, the capacitive micromachined ultrasonic transducers (CMUTs) is extensively utilized across medical, industrial, military and numerous other fields. To comprehensively analyze the fundamental properties of CMUTs, it is imperative to investigate their precise mechanical behavior under electrostatic forces. When the fringing field effect of the CMUT is ignored, this manuscript introduces the electromechanical coupling model of the partially center electrode-covered CMUTs (PCEC-CMUTs) to analyze its fundamental properties under nonlinear electrostatic forces for the first time. Utilizing the Ritz method, a novel segmentation function is employed to express the electrostatic deflections of the PCEC-CMUTs. Through the solution results of this model, fundamental performance parameters such as collapse voltage, electrostatic deflection and resonant frequency can be rapidly obtained. Subsequently, the finite element model (FEM) with COMSOL 6.1a is used to validate the theoretical model. The relative errors of electrostatic deflections and resonant frequency between the theoretical model and the FEM are controlled less than 5% in the region less than 95% of the collapse voltage, which proves the correctness of the theoretical model. These theoretical analyses are instrumental for the design and optimization of electrostatically actuated micro-devices such as CMUTs, and provide a theoretical foundation for establishing immersion equivalent circuits for PCEC-CMUT cells and arrays.

Data series of attitudes determined using vector measurements are widely available in many satellite missions. These data series are reprocessed to provide filtered attitude estimates with improved accuracies. A moving-window polynomial fitting method is employed. First, the attitude data within a moving window are transformed to small-magnitude relative attitudes with respect to a chosen reference frame. Second, these relative attitudes, often unit-norm quaternions, are reparameterized as Gibbs vectors. Third, the parameters of the empirically chosen polynomials representing the kinematics of the relative Gibbs vectors within the moving window are estimated with least-squares method. Fourth, filtered attitudes can be readily constructed with the Gibbs vectors calculated with the polynomial model at any time instant not necessarily those of the data points. Furthermore, with the Gibbs vectors and their derivatives calculated using the polynomial model, the angular rates can be readily extracted also at any time instant, as byproducts. Experiments with simulation and real data are conducted; and the results show the performance of the proposed method in terms of both attitude filtering and angular rates extraction.

Among the various defects that inevitably occur on expressways, cracks are the most common and significant indicators of highway pavement damage. Timely and accurate crack recognition is urgently required for highway maintenance. Highway pavement crack recognition depends primarily on human visual inspection and expressway maintenance vehicles. These approaches are very time-consuming, labor-intensive, and difficult to implement in civil engineering. Cost-effective fixed pan/tilt/zoom (PTZ) camera–based crack recognition was investigated in our previous work. However, for pavement cracks captured at long camera distances, the limited resolution of the PTZ vision generates low-resolution crack images. In addition, the cracks exhibit weak features, such that the crack pixels density and distribution are significantly affected by the background noise, making it challenging to recognize these cracks. Aiming to solve these problems, a high-order kernel-based modified bicubic interpolation is proposed to typically reveal and characterize discrete pixel variations, obtain high-quality super-resolution crack images, and improve the recognition performance of cracks. Extensive experiments with respect to the crack datasets captured by PTZ cameras on G4/highways in China are conducted to verify the performance of the proposed method. Two measurement parameters, namely Just Noticeable Blur (JNB) and Structural Similarity Index, confirm the high quality of the super-resolution crack images. Experimental comparisons demonstrate that super-resolution crack image-based crack recognition achieves out-performance, such that the mAP, precision (P), recall (R), and $F1$-score are increased to $95.3\%, 97.3\%, 96.1\%$, and 97.4%, respectively. This method proves the feasibility of high-efficiency crack recognition using modified bicubic interpolation for fixed PTZ vision-based expressways maintenance engineering.

Permanent magnet motors may suffer from imperceptible localized demagnetization due to surface damage during operation. This paper proposes a lightweight cross-scale decoupling feature fusion network (LCDFFN) for permanent magnets defect detection, which introduces separable convolution in the backbone and incorporates the adaptive multi-scale feature fusion with interactive attention mechanism to enhance small target detection. The adaptive attention module (AAM) and feature enhancement module based on dilated convolution are introduced in the feature fusion network, improving multi-scale object detection. Finally, the decoupled feature prediction network outputs the defect identification feature map. In addition, we also propose a novel fuzzy intersection over the union loss function. LCDFFN achieves a mAP@0.5 of 97.8%, with 108.4M parameters and a detection speed of 113.62 FPS. The proposed method significantly improves defect image detection for permanent magnets and is highly practical for industrial production.

This paper presents a bearing fault diagnosis model based on a simulated data-driven time-domain and frequency-domain feature fusion using a one-dimensional convolutional neural network (1D CNN). The proposed simulation data-driven time–frequency feature fusion 1D CNN multiscale attention (TFF-MSA) model solves the problem of insufficient generalization ability due to the scarcity of experimental data through the simulation data of the dynamic model of bearings, and accurately classifies the bearing faults in a strong noise environment. Traditional 1D CNN-based bearing fault diagnosis models often focused on either time-domain signals or frequency-domain spectra. The proposed model incorporates a parallel 1D CNN structure that simultaneously extracts features from both the time and frequency domains. These multi-domain features are then fused to capture comprehensive information from the bearing vibration signals. Specifically, to enhance the extraction of time-domain features, we introduce the pyramid attention module to mine multiscale features. Meanwhile, in the frequency-domain feature extraction, a specific fault feature frequency weighting method, allowing the proposed model to extract specific fault-related frequencies from the signals, is also proposed. Experimental results demonstrate that the proposed model which is only driven by simulated signals from dynamic model achieves an average diagnostic accuracy of 94%.

In this study, mesoporous MgO nanoparticles were successfully synthesized using carbon quantum dots as templates and an outstanding cataluminescence sensor system was established based on these nanoparticles. In a particular situation, the constructed sensor demonstrated remarkable linear range was 0.3106–310.6 ppm (R² = 0.9990) with a low limit of detection of 0.0994 ppm and a signal-to-noise ratio (S/N) of 3. The mesoporous MgO nanoparticle sensor was utilized to analyze the isobutyraldehyde content in real samples. Outstanding recoveries were achieved within range of 83.4%–111%, with the relative standard deviations (n = 7) ranging from 2.0% to 8.2%. Moreover, the 35 d test results showed that the sensor has good stability. Density functional theory and gas chromatography-mass spectrometry were used to research the mechanism of the designed cataluminescence sensor. This work presents a universal methodology for constructing isobutyraldehyde gas sensors with high sensitivity and selectivity, and mesoporous MgO nanoparticles sensors that expand the applications available in the industry and environment analysis.

NdFeB magnets, composed of neodymium, iron, and boron, are essential components in various technological applications owing to their superior magnetic properties. In the production of rare earth permanent magnets, surface defects in NdFeB have a significant impact their application, leading to substantial economic losses. Given the high precision and large volume required for NdFeB surface defect detection (32 000 pieces per day), accurately detecting and sorting sintered NdFeB raw materials online is a challenging task. This paper proposes a fast and accurate surface defect detection model for NdFeB, which enables real-time sorting based on defect detection results. The proposed model utilizes an attention module to enhance detection accuracy and enable real-time control. Compared to the original model, the proposed model improves detection accuracy by nearly 6% and achieves a detection speed of 29 FPS, thus meeting real-time performance requirements. This study offers new insights into the application of computer vision in the development of intelligent factories.

To demonstrate the potential of an ultrahigh-speed camera with frame rates beyond a million images per second for high dynamic material analysis, cylindrical concrete specimens are tested under compression at strain rates up to 164.3 s⁻¹ in a split Hopkinson pressure bar facility. The evaluation includes the determination of the actual longitudinal wave propagation velocity and the induced longitudinal deformations. The results obtained from image sequence analysis are validated with those obtained by the well-established strain gauge measuring technique that estimate the specimen deformation in the same facility. Data analysis for both measuring techniques is based on the one-dimensional wave theory. Image quality challenges affecting the material analysis are overcome by applying a Bézier fitting filter to the image sequences. Using the image-sequence-based approach to select stress-wave-induced motion points, different wave characteristics are acquired. Subsequently, selecting the points in compression during the first stress wave transit yields longitudinal wave propagation velocities up to 2995  m s⁻¹ with a standard deviation of 265 m s⁻¹. From the validation results, first, an observed bilateral complementarity of both techniques enables a comprehensive analysis of the specimen deformation. Second, individual discrepancies between the material properties obtained from both techniques up to ultimate compressive strength for deformation, velocity, strain, and stress of 20.8 %, 14.4 %, 8.3 %, and 10.0 %, respectively, are achieved.

Terrestrial mobile gravimetry using temperature-stabilized inertial measurement units (IMUs) is emerging as an efficient technique for acquiring high-accuracy gravity data in dynamic environments. This study explores the integration of the iCORUS-2 IMU system with GNSS along the Cebeci-Elmadağ route in Ankara, Türkiye. The iTempStab technology was incorporated into the iCORUS-2 system to stabilize temperature variations to enhance the accuracy of gravity measurements. Notably, this research represents the results of vehicle-based terrestrial mobile gravimetry using a thermally stabilized IMU. An extended Kalman filter (EKF) was applied to integrate IMU and GNSS data. The advantage of terrestrial mobile gravimetry over airborne methods is the ability to make static updates for vertical gravity, zero gravity gradient and zero velocity measurements, which can be introduced into the EKF as measurement updates. The method was tested by analysing the statistical impact of these updates at varying intervals: without any gravity measurement updates, and by applying updates at approximately 2 km, 3 km, 4 km, 5 km, 6 km, and 7 km intervals. Results indicated that without updates, a root mean square error (rmse) value between external terrestrial gravity data and mobile gravimetry estimates is 1.362 mGal. However, with updates at approximately 2 km intervals, the rmse value significantly decreased to 0.948 mGal, representing a %30.4 reduction in rmse and achieving sub-mGal accuracy. The inclusion of temperature stabilization and frequent gravity measurement updates demonstrated a marked improvement in data accuracy and consistency. These advancements not only validate the potential of mobile gravimetry for geophysical surveys but also broaden its applications in fields requiring precise gravity data acquisition.

This study presents a decision-making model for autonomous vehicles that integrates deep reinforcement learning (DRL) with a risk correction mechanism. This integration aims to enhance dynamic environmental perception and risk assessment in complex traffic scenarios. Initially, the model constructs driving information for the ego vehicle and surrounding vehicles. Subsequently, it introduces a polarized self-attention (PSA) and bidirectional long short-term memory (BiLSTM) combined safety mechanism (PSA-BiLSTM) to improve attention to potential hazardous vehicles in complex high-speed scenarios. Furthermore, a risk correction module has been designed to evaluate and adjust decision-making actions, ensuring driving safety. The model leverages real-time driving information of the target vehicle and its surroundings to generate vehicle decision commands at each sampling point using DRL. Simulation results demonstrate that this approach significantly improves safety and learning speed compared to pure reinforcement learning methods. Additionally, it exhibits higher driving efficiency and lower accident rates compared to imitation learning and traditional rule-based decision models in complex high-speed scenarios.

Current tactile perception of surfaces with identical materials and similar textures relies on additional equipment to maintain a constant velocity for sensors, which constrains the practical application of tactile sensing devices in daily life. A finger-shaped tactile sensor leveraging the high sensitivity of Fiber Bragg Grating (FBG) in tactile sensing is proposed in this paper. It is able to distinguish different textures based on the variation amplitude of the central wavelength in FBG. The deployment position of FBG was optimized using COMSOL Multiphysics finite element simulation software, analyzing the influence of texture depth on the output signal, and an experimental platform was constructed to collect 11 texture signals from three categories. The experimental results show that different texture features are reflected by the amplitude of FBG signal, using the SVM classification model to classify four kinds of regular ridge spacing textures and three kinds of fabric texture samples collected at random velocity, and the recognition accuracy is 99.17% and 98.89%, respectively. This method, compared to other similar texture tactile experiments, has better classification ability and advantages in adapting to different texture perception speeds, providing valuable reference for intelligent perception of robot fingertip touch.

In this paper, a differential-unwrapping-integral-regression-extraction (DUIRE) algorithm is proposed. Using the least squares fitting and the empirical mode decomposition (EMD) to replace the polynomial fitting, the cumulative error caused by noise is significantly reduced. While the maximum measurable strain is improved as compared to that of the traditional algorithm, the baseline drift, which occurs in the DUI algorithm, is resolved. In addition, the strain resolution is also improved due to the noise reduction in EMD, thus enhancing the dynamic range and signal fidelity. The dynamic range based on the DUIRE algorithm can reach 97.14 dB@1 kHz. The dynamic range improvement effect is more noticeable when the input signal-to-noise ratio is low. In a fiber Bragg grating based DAS system with 9.80 km sensing distance, only the proposed DUIRE algorithm can correctly restore the applied multi-frequency vibration signal.

This paper presents a data-driven calibration method for triaxial magnetic sensors that eliminates the need to solve sensor error model parameters. The calibration was achieved solely through the input and output of the sensor in a known magnetic field. Using both simulation and experimental data, this study employed a multilayer perceptron (MLP) algorithm as a data-driven calibration model. In static and dynamic calibration experiments conducted in a laboratory setting, the calibration performance of MLP was compared with that of traditional calibration algorithms. The experimental results demonstrate that in a static experiment under a 90 μT magnetic field, the sensor error after calibration using the data-driven method was reduced by 30.37% compared to the traditional method. In the dynamic experiments, the proposed method reduced the errors by 60.58% compared with the traditional method. The method proposed in this paper offers a new approach for improving the calibration accuracy and effective measurement range of triaxial magnetic sensors, and holds potential application value for other types of triaxial sensors.

In the hot-melt extrusion (HME) process, process parameters are critical in defining the quality of extruded products. Abnormal processing parameters may degrade product performance and lead to quality inconsistencies. Therefore, real-time monitoring of the HME process is crucial to ensuring product quality and consistency. However, most existing detection methods are constrained in their applicability owing to material-specific limitations. When process parameters vary slightly, the detection signals may exhibit minimal variations, thereby hindering the ability to differentiate between normal and abnormal conditions. To address these challenges, this paper proposes a novel two-stage fault detection and tracking method for the HME process based on data-driven techniques. A comprehensive material detection strategy is implemented by collecting and analyzing five types of data: near-infrared, Raman, ultrasound, melt pressure, and current. A global maximum variance-kernel principal component analysis anomaly detection model is constructed to identify four distinct abnormal conditions effectively. T² and squared prediction error statistical contribution maps are calculated to perform dimensionality reduction and extract features from the data. A conditional time reverse mapping-support vector machine anomaly localization model is developed that leverages the extracted features to establish the relationship between the five data types and specific abnormal conditions, thereby accurately pinpointing the root causes of anomalies. Experiments conducted using HME process equipment validate the proposed method. The results demonstrate that the two-stage fault detection and tracking approach effectively identifies various faults and achieves better recognition accuracy. This method provides a reliable solution for real-time monitoring and quality control in the HME process.

Non-contact measurement of structural vibrations from video camera has been an emerging method for vibration test and structural health monitoring. However, its accuracy is very sensitive to camera's ego motion, which may be induced by environmental winds, ground motion or unmanned-aerial-vehicle perturbation. In case of highly perturbed ego motion, the observed structural vibration from video is contaminated and even dominated by camera's rigid body motion. Thus, to mitigate the effects of camera's ego motion on vision-based measurement, a camera motion compensation method is proposed within the Lucas–Kanade framework. The realization of the proposed approach is mainly two-fold. At first, the camera's rigid body motion, possibly being large, is estimated with the help of the projection transformation of a reference template. Next, the structural vibration is recovered in conjunction with the estimated camera's ego motion. Numerical simulations and real-world experiment are conducted to verify the efficiency and accuracy of the proposed method for vibration measurement under highly perturbed camera ego motion.

Today, more and more silicon-based chips apply new flip-chip packages to accommodate the high density and power consumption requirements of electronic products. In this case, defect detection in the flip-chip package is essential. However, silicon-based materials are opaque to visible wavelengths, making it difficult to identify defects within the chip or wafer. By utilizing the property that near-infrared (NIR) light can pass through silicon, NIR confocal imaging is a promising non-destructive method to detect defects inside silicon chips. In this study, we report an integrated NIR confocal scanning microscope system consisting of a NIR microscope and a NIR confocal scanning system, which uses an InGaAs focal plane array as a detector instead of the conventional combination of a small pinhole and a photo-multiplier tube or an avalanche photodiode. The proposed confocal scanner can detect devices of different sizes in the field of view, obtain high-contrast and three-dimensional images of the internal structure, and can be effectively employed as an in-line inspection system for detecting surface and internal defects during chip manufacturing.

Stainless steel sheets with bright surfaces have been widely used in household appliances, decorative materials, highway and railway vehicles, and other fields. In order to reduce welding deformation while maintaining overall aesthetics, the commonly used welding method is laser welding. In laser welding, various welding defects are the main causes of weld failure. At present, the main detection method for defects in such slender welds is laser structured light visual scanning detection. However, due to the fact that the reflection of high gloss stainless steel material is mainly mirror reflection, there will be a substantial amount of signal loss in structured light visual inspection. This article proposes a characterization and detection method for the surface morphology and defects of laser welding of high reflective materials. The method uses latitude and longitude RGB light sources for illumination, which can form distinct and different forms of rainbow bars on various types of weld defects, providing powerful conditions for defect recognition. Unlike laser structured light vision, this method fully utilizes the specular reflection characteristics of the material surface and has a faster detection speed. This article uses a ball model to reveal the decoupled correspondence between material surface color and gradient direction and magnitude under latitude-distributed and longitude-distributed RGB light source illumination, explaining the mechanism and reliability of the detection method, and providing conditions for further quantification of defects. Finally, this article combines U-Net network and its various improved versions to achieve high-precision welding defect detection, with an average accuracy of 89.88% and a recall rate of 92%.

Simultaneously measuring the surface morphology and color of a sample is crucial for full-color 3D printing and digitization of cultural artifacts. However, traditional methods are susceptible to the variations in the measurement angle and illumination condition, making it difficult to balance the color fidelity and spatial accuracy. Here, we propose a laser triangulation method that is implemented with a fixed focus, dark-field illumination, and data-driven spectral classification. With a fixed focus, this method measures the height of surface by analyzing the shape of reflected light spot. The dark-field illumination diminishes the background contributed by the specular reflection of white light, enabling an accurate spectral measurement. The data-driven spectral classification algorithm establishes color dataset for each sample, and ensures a precise color recognition through principal component analysis dimensionality reduction and K-nearest neighbor classification. In our experiment, this method achieved an axial resolution of 5.1 μm. The color classification accuracy was as high as 99.79% on ColorChecker cards and 96.03% on Pantone color match cards. Three samples were reconstructed and evaluated to validate this method, the mean absolute error of the morphology measurements did not exceed 25 μm, and the color difference (CIE2000) does not exceed 1.0, confirming its ability to accurately measure the surface morphology and color simultaneously.

Spatial and temporal characteristics of a table-top scale laser produced plasma formed on the surface of a tungsten target are reported. The emitted radiation was spectrally filtered to a narrow band of ca. 1 nm full width half maximum (FWHM) centred on a wavelength of 13.6 nm using a combination of Zr thin film transmission filters and a Mo–Si multilayer mirror (MLM). The temporal profile of the 13.6 nm flux was measured for different laser fluences while the spatial profile of the emission was sampled in one region using a back illuminated charge coupled device (CCD), both done with the aid of a flat MLM. The emitting plasma was imaged at 13.6 nm by replacing the flat MLM with a curved mirror which returned an extreme ultraviolet (EUV) source size of up to ca. 130 µms. The peak flux was estimated to be ca. 10¹⁴ photons nm⁻¹ sr⁻¹. For comparison purposes the W target was replaced by a solid Sn target which produced, on average, almost double the EUV flux at 13.6 nm, albeit with greater shot to shot jitter.

Large-size cabin assembly is a challenging and labor-intensive task, while visual pose estimation of the cabin could enhance its efficiency and precision. The existing frameworks are mainly based on ellipse detection and projection calculation, leading to a limitation of accuracy and generalization. In this paper, we introduce a novel framework for pose tracking of large-size cabins based on iterative optimization of manipulation points to reduce the impact caused by 6DoF pose coupling, where four manipulation points iteratively transform the docking surface contour and align it with captured image. The method is applicable to arbitrary shapes of cabin docking surface and also extended to multiple perspectives. Both actual cabin assembly experiments and simulation experiments are performed, the results demonstrate that the proposed method is more accurate and robust than conventional methods.

To address the challenges of feature extraction and low detection accuracy for long-range three-dimensional (3D) objects in rainy environments, this paper proposes an improved probabilistic and geometric depth (PGD) machine learning-based detection method using multi-sensory measurements. In this new method, the deformable convolutions replace the regular convolutions in the PGD network to enhance the feature layer adaptability to the target shapes. A receptive field enlarging network is also designed to improve the detection performance for distant objects. Furthermore, an optimized multi-level feature pyramid is developed to improve the detection accuracy across varying scales. Lastly, a decision-level fusion aggregates the PGD detection head and the branch network detection head using a deep fusion module to produce the 3D object information and recognize different targets in a long distance. The proposed method is validated using the KITTI, nuScenes public datasets, and real-world testing datasets, where the camera and millimeter-wave radar are used to collect these datasets. The analysis results indicate 4%, 5.4%, and 12% increase in detection accuracy over the original PGD algorithm across these three datasets, demonstrating superior feasibility and accuracy of the proposed method for long-range object detection.

Processing interferometric images is a key task in optical interferometry for measurement purposes. It is well known that there are two main kinds of interferograms: those with open fringes and those with closed fringes. Often, processing a single interferogram with closed fringes can be challenging; therefore, applying phase-shifting is almost obligatory. However, it requires several images and a high precision mechanical device to achieve accurate results. On the other hand, interferograms with open fringes are more straightforward to process. However, precise knowledge of the carrier frequency of the fringes is essential for accurate processing. An alternative solution to the challenges posed above is the use of digital moiré interferometry, which consists of superimposing a reference interferogram and another obtained with the object to be measured. In this work we propose a digital moiré interferometry method that uses the Fourier transform and Butterworth-type filters. The feasibility and versatility of the proposed method is experimentally demonstrated using a Mach–Zehnder-type interferometer to analyze gas flow in a flame.

Because of the Global Navigation Satellite System (GNSS) signal occlusion, inertial measurement unit drift and other factors on positioning, location deviations of multi-temporal mobile laser scanning (MLS) point clouds collected in the same region are always exist. In order to improve the quality of multi-temporal MLS point clouds, it is necessary to correct the location deviations by point cloud registration. This work presents a non-rigid automatic registration method of multi-temporal MLS laser point clouds based on the characteristics of short road markings. Specifically, the central points at both end edges of short road markings were extracted as control points. The correspondences between control points in different point clouds were obtained by KD-tree and optimized by polygon similarity and Otsu methods. Then, based on the GNSS time and coordinate difference of true correspondences, the mathematical model of non-rigid registration adjustment was constructed by combining with gross error detection and polynomial fitting. Finally, multi-temporal MLS point clouds were registered according to the GNSS time and adjustment results. Validation results demonstrate that the registration accuracy reaches up to 2.8 cm. The proposed method provides a new way for high-precision fusion and change detection of multi-temporal MLS point clouds in road scenes.

Existing three-dimensional digital image correlation (3D-DIC) for surface 3D shape and deformation measurement requires the users to input key calculation parameters (e.g. subset size) to proceed with stereo and temporal matching. However, the lack of clear guidelines for optimal parameter selection often leads to ambiguity and uncertainty in the final measurements. To eliminate the ambiguity and realize full-automatic, user-independent, accurate and precise 3D-DIC measurements, we present a simple yet effective Smart DIC-3D. By fully considering local speckle quality and deformation, Smart DIC-3D automatically selects the optimal subset size for each calculation point in both stereo and temporal matching. Additionally, a fully automated initial value estimation method, combining speeded-up robust features with a reliability-guided displacement tracking strategy, ensures automatic reliable initial value estimation for both matching processes. Both numerical experiments with simulated stereo speckle images and practical applications including complex shape reconstruction and non-uniform deformation measurement were conducted to verify the effectiveness and accuracy of Smart DIC-3D. The experimental results show that Smart DIC-3D has lower random and under-matched systematic errors than regular 3D-DIC, enabling high-fidelity 3D shape reconstruction and deformation measurement independent of the practitioners' input.

Holographic microscopy has emerged as a low-cost and highly compact technique for 3D imaging of microscopic particles in suspension. However, its broad application is largely limited by the inclusion of multiple steps in extracting the particles from the hologram image, which can be computationally expensive and often involves human intervention. We introduce HoloDINO, a transformative model that leverages instance segmentation for streamlined, end-to-end particle detection and contour extraction. By pre-training a data-intensive transformer model on synthetic particle contours and fine-tuning it on experimental data, our approach demonstrates robust performance across synthetic holograms of varying particle concentration, morphology, and optical properties, as well as experimental holograms of dental aerosols and water spray droplets. HoloDINO surpasses conventional methods, which typically involve multiple steps—such as reconstruction, autofocusing, and segmentation—by consolidating these into a single, efficient process that delivers precise morphological data for each particle in one forward pass. This advancement not only facilitates real-time applications but also significantly enhances the generalization capabilities across diverse settings, paving the way for broader adoption of holography in particle analysis.

Subway tunnels deteriorate over time, posing safety risks that necessitate regular inspections. Traditional methods, like manual surveys and total station measurements, are inefficient and provide sparse data. LiDAR technology overcomes these limitations by capturing dense 3D point clouds efficiently. However, its effectiveness is influenced by factors such as incidence angles, scanning distance, and point cloud density, all dependent on station placement. This study develops a model for straight and curved tunnels to analyze how station positioning affects point cloud quality. Results indicate that placing LiDAR stations lower and closer to tunnel walls maximizes coverage, with a maximum effective scanning length of 23.48 m per station for a 2.75 m radius tunnel. An alternating station setup further enhances efficiency and accuracy, minimizing station replacements. Validation using Nanjing subway tunnels confirms the effectiveness of the proposed approach in improving data collection and structural assessments, providing practical guidance for LiDAR-based inspections.

The dynamic crack propagation characteristics of brittle materials are a crucial focus in impact dynamics. The randomness of crack growth under impact loading poses significant challenges to accurately monitoring crack behavior during experiments. In this study, two novel crack refinement measurement techniques based on optical frequency domain reflectometry technology were developed to provide more comprehensive strain distribution data under impact loading and to capture detailed crack propagation behavior. Additionally, a trending algorithm was introduced to process large datasets and accurately pinpoint crack locations. The study begins by presenting the two innovative deployment methods and the fabrication process of custom specimens. After conducting impact tests on the specimens, the resulting strain distribution data were compared with actual crack propagation. The results demonstrated precise crack location identification down to the millimeter level, and the crack extension path was accurately mapped using strain profiles from fiber-optic measurements. These findings validate the effectiveness of the proposed approach, which shows great potential for accurately monitoring, locating, and quantifying cracks using distributed optical fibers.

Ships, as thin-walled structures, experience various structural responses at critical components and may even fail under sustained time-varying wave forces. Understanding the structural behaviors of key components during operation is essential for accurate ship safety assessments. The lack of an effective strain monitoring system that integrates both high spatial resolution and dynamic monitoring capabilities affects the timely detection of these structural responses. This paper presents a high spatial resolution dynamic strain monitoring system using optical frequency domain reflectometry, achieving a system spatial resolution of 0.11 mm, a sensing spatial resolution of 20 mm, and a sampling frequency of 10 Hz. A box girder model of a river-sea-going ship is employed to simulate structural responses under wave-induced conditions. Simulation results validate the system's accuracy. By analyzing strain distribution and variation, this study explores the dynamic collapse behavior of structures under wave forces.

Infrared imagery surpasses the limitations of visible light images and finds widespread applications in fields such as military reconnaissance and security surveillance. Recent studies on infrared target detection aim to preserve local features and global representations to the greatest extent. However, compared to visible light images, infrared images exhibit inherent challenges such as insufficient texture information and coarse boundaries, which introduce new difficulties to this research. To address these issues, this paper introduces additional information cues from the perspective of enriching feature map information. Specifically, we propose a multidomain feature fusion object detector (MFFOD), whose backbone feature extraction network consists of a convolutional branch and a fast Fourier transform branch. This hybrid domain representation enables the extraction of both domain-specific information and global high-frequency and low-frequency information with minimal computational overhead. Furthermore, in the intermediate layers of the network, we have carefully designed a feature injection module that enables comprehensive interaction between channel features and spatial features within a single feature map. Experimental results demonstrate that MFFOD achieves average detection accuracies of 88.97%, 90.32%, and 99.25% on three significant infrared scene datasets, outperforming existing target detection methods. We hope that this general detection algorithm will provide a robust reference for future infrared target detection research.

Low-light visual perception problems, such as night simultaneous localization and mapping or structure-from-motion have attracted increasing attention, and the performance of keypoint detection and local feature description play a crucial role. Many traditional algorithms and machine learning methods have been widely used to detect and describe local features. However, the performance of the existing techniques in the face of highly low-light scenes will be drastically degraded, resulting in subsequent practical application needs that cannot be met. Therefore, an efficient self-supervised deep learning model DarkMatcher is proposed, which can directly detect and describe features from images in extremely dark environments in an end-to-end way. This model consists of a backbone based on new dynamic deformable convolutional blocks and a novel DarkMatcher module that combines multiple attention mechanisms to realize cross-scale feature information interaction. The former enhances the feature extraction capability of the model for low-light environment images. The latter effectively strengthens the matching ability of extremely dark environments and weak texture areas, and further improves the feature matching accuracy in low-light scenes. In addition, transfer learning and real-time training strategies are used to enhance the generalization and feature representation capabilities of the model. Many experimental results indicate that DarkMatcher possesses the best matching performance and robustness for feature points in extremely dark environment, with an average matching accuracy of 71.24% and an average execution time of 51 ms for each pair of images. Besides, the visual pose estimation experiment has also obtained good results as validation.

Owing to the unique imaging formulation principle of acoustic cameras, certain dimensions of the target may be lost in acoustic images. Some studies have been conducted to retrieve these missing details from a single pseudo front depth that contains 3D information about the target. Nevertheless, the use of a single pseudo front depth may face ambiguity problems in recovering the 3D information of acoustic targets. Therefore, this paper presents a novel semi-supervised generative adversarial network (SGAN) for estimating the depth of two adjacent acoustic images at know relative sonar poses, which generates high-accuracy reconstruction results, while reducing the dependence on the labeled ground truth of acoustic targets. Specifically, we design a semi-supervised depth estimation framework that utilizes image warping for semi-supervised training, resulting in denser depth maps. Various datasets containing underwater targets and terrain were generated to simulate real underwater scenarios. Experiments with simulator data and real images verify the feasibility of using SGAN for depth estimation in 3D target reconstruction.

Surface defect detection in industrial manufacturing ensures product quality and prevents malfunctions. To address issues such as multi-scale damage, low contrast, and small defects on the surfaces of industrial components, we propose an efficient multi-scale feature enhancement network for improving the detection performance of industrial surface defects. First, a multi-scale extraction module is proposed to extract defect features at multiple levels to ensure sufficient semantic information for multi-scale damage and enhance the feature extraction ability of defects with different scales. Dual-orientation attention is then introduced into the detection network to establish a connection between spatial and channel dimensional information, which enables the network to focus on defect regions and filter out background noise. This alleviates the problems of low contrast and small defects. The experimental results confirm that the proposed network demonstrates superior detection performance compared to other detection algorithms across five surface defect datasets. Additionally, the parameters are reduced by 7.9%, the floating-point operations decrease by 6.7%, and the model size is reduced by 5.2%. These improvements collectively provide an efficient solution for industrial surface defect detection.

Challenges like scale variations, shape diversity, complex backgrounds, and sample imbalance in remote sensing images make some targets difficult to detect. Consequently, a model called RNAF-You Only Look Once (YOLO) is proposed, combining a composite convolution module with an Intersection over Union (IoU)-based weighted loss function for remote sensing object detection. A composite convolution module, RepFocalNet, is designed and incorporated into the backbone network to replace the original C2f layer, enhancing multi-scale modeling and feature extraction capabilities in complex backgrounds. Adaptive spatial correlation pyramid attention is introduced after the ninth layer, enhancing sensitivity to subtle features and improving small object detection. Furthermore, Focal Inner Soft IoU is designed to replace the original loss function. Ablation experiments were conducted on the DIOR dataset to verify the effectiveness of each module. Following this, the proposed method was compared with several leading methods to further evaluate its performance. Compared to the YOLOv8 model, RNAF-YOLO improved mAP@50 by 1.5% and increased recall by 3%. Additionally, the classification accuracy for the bridge and vehicle categories increased by 4.2% and 8.9%. Compared to other methods, RNAF-YOLO demonstrates superior performance across multiple classification accuracy metrics. As a consequence, the proposed method demonstrates superior performance in remote sensing object detection, effectively highlighting difficult-to-detect targets.

Needle roller bearings without cages are extensively utilized in the mechanical transmission systems of various equipment due to their high load-carrying capacity and compact design. However, due to the limited performance of the current needle loading equipment, abnormalities such as missing needles, skewed needles and toppling needles. If abnormal bearings are installed in the equipment above, it will make the equipment appear noisy, vibration is obvious, the smoothness of operation is reduced. To address this issue, this paper proposes a systematic detection method based on double-threshold segmentation for identifying mixed needle assembly anomalies. Initially, starting from the principles of ray detection, we analyze the characteristics of ray imaging for different needle assembly anomalies in needle roller bearings. Image samples of single and mixed needle assembly anomalies are collected using a ray detection platform. Subsequently, a dual-threshold segmentation method is introduced for mixed needle assembly abnormality images by combining different regions of the needle-bearing ray image and the grey value characteristics of the anomalies for threshold selection. A step-by-step detection method based on dual-threshold segmentation is then proposed. This method utilizes a mask to extract the non-needle assembly region in the high-threshold segmented image, achieving identification and quantification of skewed and toppling needles based on area threshold information of the abnormal needle roller contour. Additionally, a ring mask is employed to extract the needle roller gap in the low-threshold segmented image, allowing for the screening of the needle gap contour based on the area threshold, thus identifying and quantifying missing needle anomalies. Finally, experimental verification is conducted on various needle assembly anomaly images, demonstrating that the proposed method can effectively identify and quantify different needle assembly anomalies in needle roller bearings, offering valuable application insights for quality inspection during needle assembly processes.

In order to solve the problems of scale difference, inconspicuous spatial feature expression, redundancy, and overlapping of spectral feature information in hyperspectral and multispectral image fusion, a fusion algorithm based on multi-scale space-spectrum hybrid attention network is designed in this paper. The algorithm enhances the capability of shallow feature extraction through the multi-scale mechanism, uses the space-spectrum hybrid attention mechanism to mine the correlation between deep space and spectrum, designs a cross-spectrum fusion module to realize the effective reconstruction of spatial and spectral information, and uses the loss function to constrain the difference between the fused image and the reference image in terms of color, detail edge, and spatial structure. Experiments were conducted on PaviaUniversity, IndianPines, and hyperspectral image of Natural Scenes2004 datasets, and compared to algorithms such as ResTFNet, spatial spectral reconstruction network, MoGDCN, DBSR, and Fusformer. The proposed algorithm, MSSSHANet, performs better in spectral curve smoothness, spectral difference value, root mean squared error, peak signal to noise ratio, Erreur Relative Globale Adimensionnelle de Synthèse, and Spectral Angle Mapper value, which plays an active role in improving fusion quality. However, the generalization ability of the proposed algorithm in special remote sensing tasks needs to be verified, and future research will focus on data preprocessing optimization and the development of time-spatial-spectral integration fusion technology.

This study aimed to evaluate the effect of bad contact electrodes on the quality of electrical impedance tomography (EIT) data and the reliability of lung ventilation imaging. Thirty healthy volunteers were examined using EIT under normal relaxed tidal breathing in a supine position. The five electrode–skin contact conditions (C1–C5) for electrode No. 2 were defined by systematically varying the contact area. Specifically, C1 represented full contact, C2 covered 25% of the electrode surface, C3 covered 50%, C4 covered 75%, and C5 involved completely covering the electrode with an insulating material to simulate high impedance. Three scenarios (S_A, S_B, S_C) were established: S_A maintained impedance stability, S_B simulated variations in contact, and S_C introduced dynamic changes in impedance during a single breath cycle. EIT parameter-based deviation scores were used to assess spatial and temporal ventilation distributions across varying electrode contact conditions and the Bland–Altman plot was used to assess image quality. Deviation scores in S_A consistently hovered around 0.0, except for C5, which scored 8.5 ± 4.0. In contrast, S_B displayed significantly higher deviation scores, with C2 recording the lowest at 4.0 ± 4.0. The absolute differences in tidal impedance variation between S_A and S_B remained minimal across all conditions. However, significant differences emerged between S_A and S_C from C2 onwards, where a notable shift in left–right lung ventilation distribution was observed. Changes in electrode contact impedance significantly impact the quality of EIT data. While absolute impedance within a stable range has minimal effect, fluctuations within a breath cycle compromise image quality.

Optical flow reflects motion information in videos, serving as a crucial visual cue. However, high computational demands make dense optical flow estimation challenging to deploy on mobile platforms. We address this by introducing EdgeFlow, an optical flow network designed explicitly for edge GPU. To enable efficient feature extraction, we propose a novel low memory access cost net. Furthermore, we design a convolutional gated core for optical flow update, capable of obtaining high-quality features at low computational cost. During training and inference, we decouple the upsampling procedure. Inference speed is increased by only upsampling the final update outputs, reducing unnecessary computations. Our model demonstrates significant efficiency, utilizing only 52% of the parameters required by the baseline model. In generalization tests conducted on the Sintel dataset, EdgeFlow outperforms the baseline by achieving an 13% reduction in error rate. Additionally, the processing speed of EdgeFlow deployed on the Jetson Xavier NX platform is 7.5 times faster than the baseline. Finally, after using TensorRT to accelerate the model (with three recurrent updates), the inference speed is 98.67 FPS at 640 × 480 resolution. Source code is available at https://github.com/sklibra/EdgeFlow/tree/master.

In industry, pointer meters are still widely used in traditional industries to monitor the condition of equipment. However, most of the meter reading methods can only read specific gauges. For pointer meters with different ranges, we propose a range-adaptive method for automatic meter reading called Auto-PCI, which uses YOLOv5s to detect the scale values on the instrument panel, and PP-OCRv3 to read these values and determine the range of the target meter. Skewed meter images are corrected by perspective correction, combined with the use of U-Net to accurately extract the pointer and scale areas on the meter for accurate readings. Due to environmental factors, PP-OCRv3 may misread or fail to recognize meter scale values, so we propose a range estimation algorithm to improve the accuracy (Acc) and stability of meter range recognition. Compared with existing methods, our method significantly improves the Acc of readings with a 33.3% reduction in error.

Electrical impedance tomography (EIT) is an imaging modality based on applying electrical currents to the target object. Traditionally, both the magnitude and phase of electric potential measured on the target surface are required to reconstruct the admittivity, i.e. conductivity and permittivity, inside the target. However, measuring the phase of potential requires a more advanced EIT system than that used in amplitude data-based electrical resistance tomography. In this article, we propose a novel EIT method, amplitude-based multi-frequency EIT (AMFEIT), which reconstructs admittivity using only amplitude data. Therefore, AMFEIT can be used with devices that do not provide the phase data. The performance of AMFEIT is compared to that of traditional single frequency EIT using numerical and experimental tests. In simulations, the effects of certain modelling errors and choice of frequencies on the reconstruction quality are studied. In the experimental setup, several targets inside a water tank are imaged. Both the numerical and experimental tests support the feasibility of AMFEIT for reconstructing admittivity distributions. When the frequencies are properly chosen, the AMFEIT yields reconstructions that are of the same quality as those based on conventional EIT that uses phase data.

Subsurface cracks may remain undetected until significant harm occurs, which often lead to their neglect in conventional nondestructive evaluations. In addition, the scattering of the elastic waves may affect the detection results. This study investigates the impact of the presence of aggregates on the identification and quantification of subsurface cracks using Rayleigh wave propagation. To achieve this purpose, concrete mesostructure models that integrate absorbing boundaries using the stiffness reduction method and subsurface cracks are constructed. The expected arrival times of the waves are derived and compared with the B-scans to examine the propagation path of the mode-converted body waves. Furthermore, we introduce a novel signal extraction technique using the energy spectrum based on wavelet coefficients and a knowledge-based rule to assess the depth of subsurface cracks. Finally, attempts are made to demonstrate and validate the far-field signal enhancement phenomenon of subsurface cracks using mode conversion theories and the proposed area ratio index. The results show that the theoretical curves and B-scan maps are well aligned, suggesting that the analysis of the waveform propagation path is accurate. The proposed method is quite robust for both homogeneous and heterogeneous material configurations. The crack parameters can be obtained using the fitted curve obtained by determining the peak frequency of the transmitted signal.

Precipitable water vapor (PWV) is one of the key factors in weather disaster preparedness and water forecasting. Due to the nonlinearity and nonstationarity of the PWV sequence, the existing models cannot achieve stable prediction accuracy. Therefore, this paper proposes a novel decomposition–reconstruction-prediction hybrid prediction model, named improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN)-permutation entropy (PE)-autoregressive integrated moving average with exogenous (ARIMAX)-one-dimensional convolution neural network-bidirectional long-short-term memory (1D CNN-BiLSTM), for predicting PWV. Firstly, the PWV sequence derived from the Global Navigation Satellite System (GNSS) is decomposed using the ICEEMDAN. Secondly, the PE of each decomposed modal is calculated and the PWV sequence is reconstructed into low- and high-frequency components. Then, considering spatio–temporal information, geographic information and meteorological data (longitude, latitude, altitude, day of year, hour of day, surface air pressure and temperature) aided modeling, the low- and high-frequency components are predicted using ARIMAX and 1D CNN-BiLSTM, respectively, and superimposed to obtain the predicted value and prediction intervals. Finally, the proposed model is validated for performance using the PWV derived from six GNSS stations. Compared with other models, the results show that ICEEMDAN-PE-ARIMAX-1D CNN-BiLSTM significantly improves PWV prediction performance, the mean root mean square errors for the six stations are 0.3765 mm, 0.7517 mm and 1.4696 mm for the 1 h, 12 h, and 24 h forecasts, respectively.

Airflow resistivity (AFR) is one of the most significant parameters of any given fibrous porous material for predicting the corresponding acoustical performance. In this experimental research, the AFR is first measured via the ISO 9053 method and then the results are compared with indirect measurements via an impedance tube (IT) and transmission loss (TL) tube following ASTM E1050 and ASTM E2611, respectively. The indirect measurements are carried out up to the duct cut-off frequencies, i.e. 300–4500 Hz, to demonstrate that AFR can be evaluated within an error of $\unicode{x2A7D}$2.5% even in the mid-frequency ($\sim$1–2 kHz) region, superseding the limits of the ultra-low-frequency (2–100 Hz) region as mentioned in the previous literature. From the present experimental studies, it is also revealed that the AFR calculated via IT measurement is somewhat ambiguous since there are periodic peaks and valleys due to the resonance of the air cavity used in the measurement. In contrast, the AFR calculated via the TL tube is even, particularly in the mid-frequency ($\sim$1–2 kHz) regime, making it more reliable for AFR estimations.

To meet the requirements of smart factories and intelligent industrial settings, it is of great significance to develop a reading recognition method for pointer meters. As existing methods lack the robustness necessary for complex environments and are overly specialized for specific meter types. In this paper, a robust and general method for pointer meter reading recognition is introduced. Our method demonstrates adaptability to a wide range of pointer meter types and maintains effectiveness under extreme conditions. Initially, a Tiny-YOLO object detection network is utilized to distinguish meters from the background. Subsequently, a multi-head key point detection network with a fusion strategy based on Swin Transformer is employed to identify four key points. The unique multi-head structure facilitates the key point detection of various image quality. The percentage indication is calculated using the angular method based on the key points. The numerical indication and the units come from the refined optical character recognition network by a large language model. Several extreme experiments have been conducted to demonstrate the robustness of our methods, with an average quoted error of less than 0.24%.

Gas–liquid two-phase stratified flow is a common occurrence in various industries, demanding real-time monitoring and precise measurement for safety and operational efficiency. Current methods primarily rely on sensors or limited imaging technologies, leaving room for improvement. In this paper, we introduce the EIGEN-Net framework, specifically designed for real-time monitoring of gas–liquid stratified flow within pipelines, utilizing electrical impedance tomography (EIT). EIT, a noninvasive imaging method, shows potential despite challenges from gas affecting voltage stability in image reconstruction. EIGEN-Net capitalizes on measurement matrices and deep learning to enhance measurement accuracy. It extracts crucial physical properties characterizing liquid fractions and employs acquired voltage data to pre-reconstruct a tensor. By fusing these tensors with prior information, it significantly improves imaging outcomes, effectively addressing data insufficiency issues. The incorporation of adaptive information into deep learning networks eliminates the necessity for data classification. Moreover, this approach accommodates scenarios of continuous flow variation, enhancing imaging resolution. EIGEN-Net's versatility extends to other two-phase flow scenarios and sensor measurement matrices, holding promise for applications in various industrial inspection domains.

EIGEN-Net:Deep Learning Fusion Network Guided by Leading Eigenvalues of Measurement Matrix

Fault diagnosis in rotating machinery, especially in aviation, is an active research area. RF sensors have the potential to provide on-wing information for fault diagnosis for gas turbine engines. A dual-polarized, distributed RF sensor system was tested in an axial compressor to determine if the measurement system could detect and predict rotor offset and whirl. A magnetic bearing system was used to dynamically control the rotor position. The RF polarimetry data were used as the input data to simple, multilayer neural networks. Regression analysis was performed to predict the rotor shaft centerline position and rotor whirl orbits' major and minor axis lengths. Typical neural network error was within 5 μm for whirl orbits up to 175 μm. The quality of the neural network prediction as a function of the number of features was studied. The results suggest that a final configuration may only require a single transmit-receive antenna pair and one RF tone.

Zero-shot learning (ZSL) endeavors to extend knowledge to novel classes by capitalizing on the semantic overlap across different categories. Contemporary ZSL approaches concentrate on isolating image-specific local features pertinent to attributes and using these to align class semantic vectors. However, existing methods often overlook the integration of global and local representation, a synthesis that could significantly enhance zero-shot recognition accuracy. This paper introduces an innovative ZSL technique, termed ZSL based on the fusion of global representation and local representation (ZGLR). Our approach incorporates a Transformer encoder constructed upon attribute prototypes to extract attribute-level features, which are then mapped to local representation to align with class semantic vectors. Concurrently, we introduce a discriminative semantic-visual mapping network that embeds class semantic vectors into the visual domain, thereby aligning global representation. During training, both global and local representation are optimized in tandem, while in the testing phase, the outcomes from both representational levels are consolidated to bolster classification precision. Practical results from three benchmark ZSL datasets demonstrate the superiority of our put-forward ZGLR solution.

Prediction of product quality characteristics plays an important role in the timely identification of quality conditions and in triggering an alarm for abnormal products. In modern manufacturing, the large number of parameters collected by sensors and complex interactions between operational parameters have led to complexity and difficulty in quality monitoring during the production process. To minimize noise interference, reduce modeling complexity, and improve prediction accuracy and interpretability, this study proposes a time-series causal discovery and quality prediction framework for multistage manufacturing processes. Initially, a hierarchical Peter–Clark momentary conditional independence algorithm with multiple time-lag detection accuracy algorithm was proposed. It is designed to identify optimal time lags, establish causal relationships between process parameters and quality characteristics, and efficiently extract the root process parameters. Furthermore, a temporal pattern attention–long short-term memory model is employed to predict the quality characteristics for multivariate time series data, and is aided by the obtained causal structure. Finally, data obtained through a simulation and a case study involving a multistage continuous production chemical process are utilized to verify the performance and superiority of the proposed method.

Condition recognition in coal flotation is a critical component in achieving flotation process control and optimization. To develop reliable models for condition recognition, two key challenges must be addressed: first, the effective extraction and integration of froth visual feature information from multiple flotation cells; and second, the fusion of froth visual features with process parameters to create a comprehensive representation of flotation conditions. To tackle these challenges, a time series classification model based on multi-source heterogeneous data was proposed for coal flotation condition recognition. First, a bubble size feature representation method combining the tsfresh and maximum relevance minimum redundancy algorithms was proposed. This method accurately describes bubble size distribution features through multi-scale feature extraction and mutual information-based feature selection. Next, a neural network feature evaluation model based on L1 regularization was presented, which quantifies the importance of visual features through the sparse expression of network weights, facilitates feature selection, and applies the resulting feature weights to the subsequent weighting process in classification models. Finally, a residual temporal network that integrates residual feature extraction and temporal modeling was proposed. This model employs one-dimensional convolution and cascaded residual structures for deep feature extraction, while temporal dependencies of condition changes are captured by combining long short-term memory and attention mechanisms. Validation on a flotation froth dataset collected from field sites in Anhui demonstrates that the proposed model achieves a condition classification accuracy of 85.08%, significantly improving the performance of flotation condition recognition compared to existing models.

The detection of defects on automotive coated surfaces is of paramount importance to ensure the quality of automotive appearance. However, due to the challenges posed by the lack of sufficient samples of product-specific coated defects, the uneven number of species, and the difficulty of detecting defects in small targets due to the complex background of the body surface. The detection of defects on automotive coated surfaces still relies on experienced manual labor to a significant extent. To address these issues, Poisson fusion is employed to enhance the data for generating defect images, thereby mitigating the limitations of insufficient coated defect samples and an imbalanced distribution of defect species. Based on this, a new YOLOv8-SC defect detection model is proposed, in which the C2f-Star is used to replace the original C2f structure. This aims to improve the model's ability to extract defects of varying sizes in complex backgrounds, while reducing the number of model parameters and the amount of computation. The content-guided attention fusion (CGAFusion) module, situated before each detection head of the model, has been designed to enhance the model's capacity to extract defects of varying sizes and complexity. This is achieved through the adaptive fusion of low-level and high-level features. The proposed model's superiority in terms of detection metrics has been demonstrated through the use of example validation. The experiments were conducted on the enterprise dataset CPD-DET and the public dataset NEU-DET. The results demonstrated that the defective image generation and data enhancement method could significantly improve detection performance and have good generalization. The proposed YOLOv8-SC reduces the model parameters by 12.2% compared to the normal model. Additionally, the mean average precision (mAP) at 50% and 50%–95% thresholds are enhanced by 4.7% and 2.3%, respectively. Furthermore, the accuracy of the model exhibits a discernible positive growth trend with an increase in the sample set size. The design of an automotive coated surface defect detection system is presented at last. When deployed in industrial settings, this system can facilitate the intelligent detection of coated defects.

Accurate identification of prohibited items in x-ray security images is essential for ensuring public safety. However, current methodologies struggle to simultaneously address irregular deformation, multi-scale features, and background occlusion of prohibited items, leading to inadequate detection accuracy. To address these challenges, we propose an adaptive efficient focusing network (AEFNet) designed to target regions, thereby enhancing the automatic detection of prohibited items. Specifically, to accommodate the irregular deformation of target regions, we introduce the DACSP module, which dynamically adjusts sampling positions to enhance the network's adaptability and focus on occluded targets. To address detail loss and managing multi-scale features, we propose a multi-scale focus feature module and a focusing diffusion pyramid network (FDPN), which enable the fusion of semantic and perceptual features, improving the use of contextual information at different detection scales. Additionally, detail-enhanced convolution improves the efficacy of feature utilization at different scales, while facilitating a lightweight network design. Finally, we employ the PIoUv2 function to optimize localization loss, resulting in significant performance enhancement. Experimental results show that AEFNet performs effectively across various x-ray security image datasets (PIDray, CLCXray, OPIXray) achieving 74.7%, 61.3%, and 89.2% mAP respectively, and AEFNet also demonstrates strong generalization capabilities on the PASCAL VOC dataset in non-prohibited item detection scenarios.

In recent years, the rapid advancement of unmanned aerial vehicle (UAV) technology has amplified the need for robust object detection algorithms capable of handling aerial perspectives in remote sensing applications. Traditional detection algorithms are often difficult to simultaneously cope with two challenges in UAV application scenarios, i.e. small target detection and lightweighting. To address these issues, we propose lightweight transformer-based detector, a lightweight object detector designed specifically for UAV-based object detection in remote sensing scenarios. To solve the problem of difficult feature extraction for small targets, we propose three modules to optimize feature fusion, namely Cross Stage Partial-Omni Kernel, dynamic sampling (DySample), and Lightweight Group Convolution Channel Shuffle. In order to improve the real-time operation of the model, we adopt a dual knowledge distillation strategy that combines feature distillation and logical distillation, which maintains the original accuracy while the model is lightweight. Experimental results on the VisDrone 2019 and TinyPerson datasets show that Our model achieves state-of-the-art performance while decreasing the network parameters and computation by nearly 50%.

Aiming to solve the urgent problem of sag measurement in constructing ultra-high voltage (UHV) power transmission lines non-line-of-sight conditions, we have developed an internet of things-based quality inspection robot for UHV overhead transmission line conductors to assist in construction. This robot not only improves the accuracy of sag measurement but also possesses the function of detecting and locating line defects. Utilizing sensors, it monitors sag, environment, and operating conditions in real-time and uploads the results to a cloud platform. The proposed lightweight YOLOv5s-SGSW algorithm, deployed on NVIDIA Jetson TX2, detects transmission line defects with 81.9% precision at 35.62 FPS. The integration of autonomous positioning algorithms and defect detection locates the defects on the transmission line. The relative errors are 0.25% for the X-coordinate and 2.07% for the Y-coordinate of the defect localization. A new tension-based sag compensation method is proposed to compensate for the sag error caused by the robot's weight, with experimental results showing an error of 1.3%. This indicates that the robot can meet the expected design goals and satisfy the practical engineering requirements.

Due to the spatial resolution limitation of the magnetic Barkhausen Noise (MBN) sensor, the mechanism by which MBN signals are affected by the elastoplastic evolution process at the grain scale remains unclear. In this study, a high-resolution MBN sensor was used to investigate the distribution of MBN signals near grain boundary in grain-oriented silicon steel under controlled residual plastic strain and applied stress, simulating a cold work hardening process. It was observed that MBN signals near the grain boundary varied with the introduction of applied stress and residual plastic strain. A multi-factor mechanism was analyzed and proposed, incorporating the effects of microstructure, applied stress/strain, and residual stress at the grain scale. These findings provide valuable insights for the subsequent improvement of the micromagnetic theory at the grain scale.

High-energy laser impact tests on identical materials and irradiation parameters often show deviating results between different testing institutes. This particularly affects the specular component of the reflection and the temperatures. Even the supposedly easy-to-determine perforation time sometimes shows large variations between different test institutes. For a reliable and reproducible determination of test results, a standardization of the experimental setups is necessary. Test results using two different reference materials are presented. These tests demonstrate that standardizing experimental setups and evaluation methods significantly reduces the deviations of the experimental results. The presented multi-institutional approach aims at gaining data of laser reflection, temperature, and perforation time of laser impacted samples, and at developing an optimized experimental setup based on different existing laboratory equipment and experimental approaches.

Potholes present a significant safety risk on non-motorized vehicle lanes, especially under low-visibility conditions. Effective pothole detection on non-motorized vehicle lanes is crucial to improve public transportation safety. This study proposes an integrated algorithm that harnesses smartphone sensors to enhance pothole detection accuracy. The algorithm begins with data processing, incorporating techniques such as the quaternion algorithm, synthetic minority over-sampling technique, and wavelet-domain denoising. This preprocessing addresses challenges such as significant smartphone placement uncertainty, limited pothole data, and intense noise signals, all of which severely affect the prediction accuracy of machine learning models. The processed data is subsequently used to train machine learning models for pothole detection, including artificial neural networks (ANNs), bootstrap forest, and Naïve Bayes. The accuracy and precision of the models are evaluated and compared. The results show that the accuracy of pothole detection with the integrated algorithm improved to 92%–97%, surpassing the 70%–90% accuracy reported in previous studies. Using the ANN prediction model, the integrated algorithm achieved the highest overall accuracy of 97.02%, with an F-measure of 95.15%. Additionally, the Naïve Bayes model effectively addresses the class imbalance in pothole detection, achieving the highest precision (97.93%). These results confirm the effectiveness and improved accuracy of the proposed integrated pothole detection algorithm.

Machine learning (ML) models will soon be found in devices of critical structures or even in high-risk applications. Arranging the Artificial Intelligence Act of European Union, requirements and a methodology for an approval of measurement devices by a regulatory authority or a state institution need to be developed. In this study, using the example of loudness classification from electroencephalography measurement data, phenomenological methods were developed for the evaluation of ML models in the context of an approval process. A schedule of six sequential steps is proposed that can be used as a flowchart in such a process. After identifying the relevant features (step 1) performance measures and benchmarks were defined (step 2). In the following performance evaluation (step 3) the influence of the input data on the model's output was determined. The range and characteristics of the input data at which the device operates correctly were identified and quantitative descriptors were determined for the full range of operating conditions specified in standards or regulations for approval (step 4). Perturbation-based techniques were developed and adapted to define corner cases and to create new test data that identify the areas of the input data where performance falls below a certain threshold (step 5). The newly found test data were finally applied in step 6, and it could be shown that the data can reveal weak points in the models. All processes can and must be used without knowledge of the models, as an independent test of a finished device is always carried out during a type approval. Thus, this study contributes to the still very nascent field of type approval development for measuring devices, software and methods containing machine or deep learning.

Digital volume correlation (DVC) is widely used for the analysis of three-dimensional displacement and strain fields based on computed tomography (CT) scans. However, the applicability of DVC methods is limited in regard to geomaterials: the speckles are directly correlated with the microstructure of the material, and the speckle structure cannot be artificially altered, resulting in generally poor speckle quality. Additionally, most geomaterials exhibit elastoplastic properties and undergo complex large deformations under external loading, sometimes leading to strain localization phenomena. These factors contribute to inaccuracies in the displacement field obtained through DVC, and at present, there is a shortage of correction methods and accuracy assessment techniques for the displacement field. If the accuracy of the DVC displacement field is sufficiently high, the gray residue of the two volume images before and after deformation should be minimal, and the use of this characteristic to develop a correction method for the displacement field is feasible. The proposed self-correcting strategy of the displacement field is based on image matching for local DVC method, which is based on the experimental measurement error. We demonstrated the effectiveness of the proposed method by performing CT triaxial tests on granite residual soil. Without adding other DVC parameters, the gray residue showed that the proposed method can effectively improve the accuracy of the displacement field. Additionally, the accuracy evaluation method can reasonably estimate the accuracy of the displacement field. The proposed method can effectively improve the accuracy of the DVC three-dimensional displacement field for the state of speckles with poor quality and complex large deformation.

For individual calibration and testing events, no third-party assessments are conducted on the validity of measuring instrument calibrations, personnel qualifications, or the appropriateness of measurement conditions in current calibration and testing practices, as is done during accreditation review audits. Since performing such assessments manually can be labor-intensive for human reviewers and prone to errors, we have developed a system for the automatic validation of digitized data, working alongside the current accreditation system. This system is designed for the documented records of the calibration of a reference photovoltaic device. The system incorporates three key functions: data packaging, automatic validation, and auditable log by the means of the blockchain system. Specifically, all data related to the calibration of an item is packaged by an application, the extracted data from the package is validated by programs, and the validation logs are stored securely without tampering. This system can support quality compliance in calibration and testing processes.

In most existing intelligent fault diagnosis methods, noise is considered harmful and may decrease diagnosis accuracy. In contrast to these methods, this study proposes a novel fault diagnosis method with extra noise injection, termed an adaptive-noise-injected convolutional neural network (CNN). Noise is intentionally injected into a CNN model's softmax layer to improve fault diagnosis accuracy. The injected noise is used in the iteration of the CNN model and adaptively adjusted according to the change in model loss. Bearing datasets from Case Western Reserve University and Paderborn University were used to validate the effectiveness of the proposed method. The robustness of the proposed method was illustrated by injecting Gaussian and uniform noise. By comparing the ablation study results with those of the state-of-the-art methods, and t-test results before and after noise injection, the effectiveness of noise injection in enhancing diagnosis accuracy was demonstrated. The proposed method performed well on small samples and in complex working conditions, and possesses good generalizability and the ability to deal with real-world datasets.

Background magnetic field compensation can effectively improve the signalto-noise ratio of biomagnetic field measurements. But for complex environmental disturbances, creating a stable near-zero magnetic field environment is a challenging topic. Therefore, this paper constructs a background magnetic field compensation system based on the Improved Linear Expanded State Observer (ILESO), which effectively reduces the magnetic field fluctuations. Firstly, the system model is obtained based on the input and output characteristics of the system. Then, an improved linear extended state observer is designed to observe and compensate the environmental disturbance, and the adaptive rate is established to balance the estimation capability of the observer and the measurement noise. Finally, the experimental results validate the effectiveness of the proposed method. For 1 Hz disturbances, the peak-to-peak magnetic field value was reduced from 3.71 nT to 0.33 nT. Under the influence of external environmental magnetic field disturbances with an amplitude of 3417 nT within the 0–15 Hz frequency band, the internal magnetic field within the magnetic shielding room remains consistently stable within 1 nT. These results have a significant reference for obtaining an ultra-stable, near-zero environment.

Bearing fault diagnosis under variable speed conditions is essential due to the complexities introduced by speed fluctuations. The accurate detection of multi-harmonic faults is critical for ensuring reliability in intricate operating environments. From the perspective of the beneficial effects of noise, in this study we propose a novel damping-regulated generalized stochastic resonance (GSR) array method designed for multi-harmonic fault diagnosis under variable speed conditions. First, we employ computed order tracking to transform non-stationary time-domain signals into stationary signals in the angular domain. A damping-regulated GSR oscillator is then introduced within this domain, forming the basis of our GSR array. By analyzing the system stationary response, we reveal the diagnostic performance in theory to assess the array's capacity for enhancing multi-harmonic fault characteristics. Through simulations and experimental validation, our method demonstrates superior diagnostic accuracy, particularly in variable speed scenarios. It excels in preserving and enhancing weak multi-harmonic fault characteristics while offering significant advantages in high diagnostic robustness. These findings provide significant potential for practical applications in fault diagnostics across various engineering systems.

Weak signal amplification is extremely important in biological systems (such as brain, nerve cells, gene regulatory networks), which relates to signal coding and processing. This study incorporates intra-layer coupling within a simple Y-shaped unidirectional chain as a means of exploring the effect of small changes in the network structure on the propagation of weak signals, where the nodes in the chain are treated as bistable oscillators. It is found that when the intra-layer coupling is below a critical threshold and the inter-layer coupling is at a medium range, the transmission of signals in the initial layer is significantly enhanced. This study also reveals the amplification mechanism of weak signals in the Y-shaped unidirectional chain. The successful amplification of weak signals in the Y-shaped unidirectional chain is related to the synchronization state of the feed-forward oscillators and the amplitude of oscillators. The signal amplification is also determined by the frequency of the input signals. Furthermore, this paper has contrasted numerical simulations with analytical calculations in this simplified topology. This research contributes to understanding the relationship between network structure and weak signal transmission.

The presence of complex electromagnetic noise significantly impacts the accuracy of magnetic targets signal detection, necessitating the development of an effective denoising method to enhance detection precision. Nevertheless, traditional denoising methods faces problems such as difficulty in selecting wavelet basis, difficulty in specifying the decomposition level, and unreasonable selection of thresholds. This study introduces improved wavelet threshold denoising based on peak-to-sum ratio and composite evaluation index T, named as (PSR-T-IWTD). PSR-T-IWTD integrates the improved wavelet basis selection method, improved wavelet decomposition level selection method, improved threshold selection method, and improved threshold function design method. Calculate the composite evaluation index T and select the wavelet basis with the smallest T as the optimal wavelet basis. The optimal number of decomposition level is determined by the PSR of the wavelet detail coefficients. An improved threshold selection method and threshold function are introduced to further enhance the performance of wavelet threshold denoising (WTD). Finally, the magnetic field denoising test of the ship model was designed and compared with Butterworth low-pass filter (BLPF), optimal wavelet selection wavelet adaptive threshold denoising (OWSWATD) and improved WTD based on T (T-IWTD) to verify the effectiveness of PSR-T-IWTD. The test results show that PSR-T-IWTD has lower computational complexity. Meanwhile, PSR-T-IWTD improves the signal-to-noise ratio by 10.2%, 6.8% and 8.3% compared to BLPF, OWSWATD and T-IWTD, respectively.

Many research studies have been performed to extract non-stationary signal properties, such as the instantaneous frequency (IF) and envelope, to diagnose machinery faults under time-varying speed conditions. However, accurate IF extraction is difficult from such complex fluctuating signals to identify faults efficiently. Therefore, this paper proposes the statistical complexity measure (SCM) as a feature that evaluates the randomness and complexity without any additional requirement, providing an outstanding result in multi-component fault detection under speed fluctuations. Primarily, the influence of SCM is proved by ranking the features with the help of the random forest algorithm. The fault classification of individual signal characteristics confirms its ranking order as well. Then, it is targeted toward the efficacy of the signal fusion approach in multi-component fault diagnosis. Focusing on the essence of SCM, it is merged with the other three signal properties based on the ranking technique to investigate which one of the feature compositions is more advantageous. The machine learning classifier, SVM-RBF, achieves 94.7%, which proves the potential effects of SCM features in industrial rotational machinery components under variable speed conditions.

In recent years, with the rapid advancement of autonomous driving technology, the requirements for environmental perception tasks have become increasingly important. Four-dimensional (4D) millimeter-wave radar, an economical and reliable technology, has begun to attract attention. Additionally, LiDAR can measure accurately and is not easily interfered with, so it is also very popular. To overcome the limitations of single-sensor perception, this paper proposes the BSM-NET method, a multi-bandwidth, multi-scale, multi-modal fusion approach for 4D radar and LiDAR. This paper uses image technology to clean point cloud data, reducing errors and noise. BSM-NET consists of two key modules: multi-bandwidth fusion (MBF) and multi-scale fusion (MSF). MBF enhances data quality by capturing point cloud density and addressing issues such as gap filling and noise reduction. MSF improves accuracy and robustness through high-precision calculations. For better integration, we use the radar and LiDAR based multi-modal fusion (RLMF), which enables different types of data to learn from each other, allowing us to effectively fuse 80-line LiDAR and 4D radar data. Experimental results demonstrate that BSM-NET significantly outperforms the current state-of-the-art algorithms on Dual-Radar dataset. Compared to M2-Fusion, our fusion method achieves notable performance improvements of 2.04% and 2.21% in medium-difficulty 3D object detection and bird's eye view detection for the vehicle category, respectively.

Rolling bearings are essential in rotating machinery, and accurate remaining useful life (RUL) predictions are necessary for effective maintenance and optimal performance. Poor trend of health indicators (HI) and operating condition variations can reduce the reliability and accuracy of RUL predictions.To address these challenges, we propose a deep transfer network based on dual-task learning for predicting the remaining life of rolling bearings (DTLDL-RUL). This framework integrates health status assessment and RUL prediction, leveraging task commonalities and differences to create a robust model, then transfers it to improve generalization in the target domain. Specifically, we first mine spatiotemporal features from vibration signals to generate and classify HI, labeling them for subsequent tasks. Next, we use the shared feature extractors and private residual networks to capture common and specific features of each task, merge them with the multi-gate control networks, and adaptively adjust task weights in the loss function to enhance model adaptability. Finally, the trained model is transferred to the target domain using domain adaptation to extract domain invariant features and consider target-specific features, enhancing generalization. Furthermore, to enhance prediction accuracy, we incorporate physical models as constraints in the loss function, combining data-driven and physical principles to improve model interpretability. Experiments conducted on the 2012 PHM and XJTU-SY datasets demonstrate that the proposed method achieves high accuracy and generalization in RUL predictions.

Early bearing fault diagnosis is essential for stable operation and low-cost maintenance of rotation motors. A rolling bearing fault diagnosis method for rotation motors is proposed, leveraging a novel interpretable short-time kurtosis error (SKE) indicator and the improved snow ablation optimizer (ISAO). Initially, the Cauchy mutation mechanism is employed to bolster the global search capability of ISAO, thereby optimizing parameter settings. Subsequently, ISAO is utilized to adaptively configure key parameters within the adaptive feature mode decomposition (AFMD), reducing reliance on prior knowledge. The SKE indicator is then developed to enhance the efficacy of AFMD in extracting fault features. The proposed AFMD is validated using bearing single and composite fault data from two experimental platforms. The experimental results under various fault conditions show that AFMD can adaptively perform fault feature extraction and robust to noise interference. Meanwhile, the proposed method performs better than existing methods, such as traditional FMD and dung beetle optimizer-based variational mode decomposition. This study advances state-of-the-art bearing fault diagnosis and offers a more interpretable and adaptive rotation motor condition monitoring solution.

Accurate and reliable baseline modeling of gas turbines is crucial for effective gas path fault diagnosis. This study proposes a novel baseline modeling approach based on physics-guided multi-model method utilizing the engine condition monitoring measurements. The physics-guided multi-model architecture, comprising a baseline model and compensation models, employs environmental compensation to refine the output parameters. The baseline model is constructed using data from a single environmental condition. Compensation models are built using residual data between baseline model outputs and engine gas path measurements from other environmental conditions. This approach was validated using turboshaft ground test data across different seasons. Baseline models including output power (Pdn), high-pressure compressor outlet pressure (Ps3), and power turbine inlet temperature (Tt45) were developed using both single and multi-model methods. Back propagation (BP) networks and recurrent neural network were used to validate the physics-guided multi-model architectures. Comparing the prediction accuracies of single and multi-models reveals that the multi-model structure offers superior accuracy. The compensation-based multi-model method reduced the mean absolute error (MAE) by 57.3% relative to the uncompensated BP and by 44.6% relative to the multi-input BP, achieving final MAE values of 0.967% for Pdn, 0.78% for Ps3, and 2.3% for Tt45. At the same time, the multi-model approach enhances interpretability by separately predicting the effects of control and environmental parameters on engine monitoring parameters.

The failure of gearboxes, a critical component of mechanical power transmission systems, can significantly disrupt process cycle times and decrease production line throughput. Predicting failures in gear transmission systems is notably challenging due to their complex geometry and the interaction of simultaneous faults, which complicates fault isolation. Typically, multiple sensors are deployed at various locations to isolate and analyse faults in gearboxes. However, not all sensor data yield reliable results, making it crucial to select the most effective sensors. This study employs a principal component analysis-based approach to select the best sensor for fault detection. The results show an 18% increase in fault detection accuracy when using the most effective sensor. To validate the proposed approach, experiments were conducted under four gear conditions, considering different speeds, loads, fault severities, and fault types.

Wafer defect classification is a key component in the wafer manufacturing process. Under stable operating conditions and sufficient test data, an effective wafer defect classification model can help engineers quickly and accurately judge and solve problems in the production process. However, the complexity of the production process leads to serious imbalance between various types of defects, which greatly reduces the performance of traditional defect classification method. This paper proposes a Jacobi regularized generative adversarial network (JRGAN) for sample imbalanced wafer image defect generation. The JRGAN architecture includes a generator, a discriminator, a Jacobi regularization term, and an auxiliary classifier. The model takes random noise and sample labels as input, and integrates the Jacobi regularization term into the generator to minimize the statistical difference between the generated image and the real image. The regularization term in the discriminator improves the robustness of the network training process. This paper uses the MIR-WM811K and MixedWM38 datasets collected from real factories to verify the effectiveness of the JRGAN model proposed in this paper on the residual neural network (ResNet). Experimental results show that the proposed method can improve the quality of generated samples and improve the accuracy of wafer defect classification. The defect classification accuracy in the MIR-WM811K and MixedWM38 datasets is 97.14% and 97.38%, which is 2.21% and 0.29% higher than that of the original datasets.

Aiming to address the issue of the complex and harsh working environment of rotating machinery, the features of vibration signals associated with structural faults are often obscured by noise, resulting in low accuracy in fault diagnosis. This paper proposes a method for feature enhancement and diagnosis of rotating machinery structural faults, which combines the low-pass Teager energy operator intrinsic time-scale decomposition (LTEO-ITD) recurrence plot (RP) with the ResNet18 network. Firstly, the low-frequency components of the vibration signal are extracted and enhanced using the LTEO. The method effectively suppresses noise interference and enhances fault features. Then, the fault features are extracted using ITD. The component that contains the highest number of fault features is selected based on kurtosis analysis, followed by the generation of the corresponding RP. Finally, the data is input into the ResNet18 network for diagnostic verification. The effectiveness and feasibility of the proposed method are verified through vibration signals from the rotating machinery experimental platform and the comprehensive rotating machinery experimental platform. The proposed method achieves a diagnostic accuracy of 100% on both datasets. The comparative validation was conducted using five distinct image encoding methods. The experimental results show that the proposed method effectively extracts fault features of structural faults, thereby enhancing the accuracy of fault diagnosis.

The existing radar-based human fall detection methods mostly consider relatively simple fall actions, overlooking the complexity of actual human falls involving different postures and speeds. Additionally, the dynamic changes in household items such as electric fans and curtains in real indoor environments can affect the performance of fall detection. To address these problems, we propose a human fall detection approach based on adaptive local region-duration features using multi-input multi-output radar. We first utilize multiple target detection and tracking to obtain target trajectory information. Then, we select local region and extract multidimensional dynamic sequences of targets based on their trajectory information. Finally, we adaptively intercept corresponding information sequences through suspicious fall event detection and extract adaptive local region-duration features for the classification of fall and non-fall events. The experimental results show that the proposed approach is capable of detecting various complex fall actions with different durations. Moreover, the proposed approach achieves an average accuracy of 98.2% and 98.0% in the absence and presence of interference, respectively.

In the monitoring of rotating machinery area, intelligent fault diagnosis based on signal analysis has been widely applied. However, due to modulation of the hardware transmission path and interference from environmental noise, the quality of collected vibration signals is prone to degradation. Convolutional neural networks (CNNs) are currently the most widely used models for fault diagnosis. However, their lack of dedicated denoising structures makes them less robust against noise. Therefore, this paper proposes an end-to-end denoising CNN fault diagnosis model. Firstly, a discrete-wavelet attention layer (DAL) and convolutional layers are alternately employed to extract signal features in the wavelet domain. Secondly, according the periodic self-similarity of vibration signals, the Gramian noise reduction (GNR) method is utilized to enhance fault features in the signal. Subsequently, GNR and DAL are integrated into the model to simultaneously extract features from the original signal and the vibration signal enhanced by GNR, thereby enhancing the model fault diagnosis performance in noisy environments. Finally, various levels of noise are added to Case Western Reserve University and Dalian University of Technology data, and compared with other advanced methods, to verify the effectiveness and universality of the proposed method.

Remaining useful life (RUL) prediction is an essential technique in the prognostics and health management of aero-engines, crucial for ensuring reliability and safety. Recently, data-driven methods have achieved notable progress in aero-engine RUL prediction. However, they often neglect the impact of dimensional information and the coupling information between dimensions, failing to meet the requirements of long-term prediction tasks for aero-engines. This paper presents a prognostic model for aero-engine RUL prediction, utilizing a temporal convolutional network (TCN) enhanced by dual-dimensional fusion attention to tackle these challenges. Initially, the feature attention module is employed to weight the data from various engine sensors, thereby emphasizing key features. Subsequently, the TCN learns the temporal dependencies from the weighted input data. Additionally, a dual-dimension fusion attention module is designed to account for interactions between different dimensional data of the engine. This module extracts features from various sensors and time steps through parallel sub-structures and captures the coupling information between dimensions using an attention fusion unit, achieving inter-dimensional correlation of aero-engine data. Finally, the proposed model's effectiveness was verified using the widely recognized C-MAPSS dataset. The results demonstrate that it outperforms state-of-the-art methods in aero-engine RUL prediction accuracy, confirming its superiority.

The alpha-stable distribution is capable of modeling non-Gaussian impulsive noise, which widely exists in practical applications. Research on weak signal reception with alpha-stable-distributed noise is not only of great significance but also faces substantial challenges. In light of this, an innovative weak signal reception algorithm integrating a soft-limiter with the adaptive stochastic resonance (ASR) technique is proposed. This approach capitalizes on the initial noise suppression capability of the nonlinear soft-limiter, coupled with the unique benefit of the ASR technique in fostering a cooperative resonance effect through automatically adjusting stochastic resonance system parameters, even amidst highly dynamic signal and noise conditions. Through an in-depth quantitative study of the stochastic resonance system mechanism, detailed steps for ASR system implementation are provided. The realization of the ASR system can make full use of the random noise energy transferring to ordered signals, thus achieving the goal of using random noise to enhance weak, useful signals. By employing the ASR system as a weak signal pre-enhancement unit, the performance of weak signal reception in alpha-stable-distributed noise environments can be significantly improved. Finally, extensive numerical simulations are provided to verify the efficacy of the proposed algorithm.

Class imbalance remains a persistent challenge for intelligent fault diagnosis (IFD) algorithms in practical applications. Expanding imbalanced data sets using finite element model (FEM) data offers a promising solution. However, significant distributional discrepancies between FEM data and actual measured signals pose challenges for effective information fusion. In this study, we propose a novel FEM-assisted fault diagnosis paradigm to address the class imbalance issue by combining domain adversarial (DA) strategies with an adaptive fault frequency attention module (AFFAM). First, we apply the variational mode decomposition algorithm to partition the frequency spectrum, facilitating model optimization through the incorporation of prior knowledge, thus laying a foundation for subsequent multi-dimensional feature extraction. We then employ the proposed AFFAM to extract multi-dimensional features from the input, enabling the model to integrate information from different granularity levels and bridging the information gap between simulated and real-world data. Finally, data integration is achieved through a data analysis strategy. The DA-AFFAM framework is validated using a customized bearing test bench. Results indicate that the classification accuracy for multi-class imbalance problems reaches 92.88%, even with a class imbalance ratio of 200:1.

The fast kurtogram is a very useful tool in the field of fault diagnosis, but it also contains two drawbacks. On the one hand, its kurtosis indicator is susceptible to random impulse interference, leading to frequency band mis-selection. On the other hand, the frequency band division method determined by the tree filter bank may cause under or over decomposition problems. Therefore, in this paper, a novel fault diagnosis theory is constructed combining the adaptive recombination empirical wavelet transform (AREWT) with the envelope spectral energy ratio (ESER). Adaptive decomposition of frequency bands is first performed using AREWT. Afterwards, the ESER is proposed as a statistical indicator to choose the optimal demodulation frequency band which overcomes the effects of non-Gaussian noise interference and improves diagnostic accuracy. To further highlight the fault elements of the selected components, an adaptive sparse coding shrinkage algorithm is introduced for sparsely denoising sensitive components. Correspondingly, clear fault feature frequency components can be extracted from the envelope spectrum. Finally, the practicability and superiority of the proposed AREWT-ESER approach are fully validated through numerical simulation signals and case studies.

The accurate analysis of close and crossed frequency components is a challenge for the existing time–frequency analysis methods due to energy interference in the adjacent and crossing parts. In addition, the energy error at the intersection is caused because the energy value is superimposed repeatedly. To solve these problems, a new adaptive demodulation extraction transform (ADET) is proposed in this paper. An instantaneous frequency (IF) trajectory estimation method and an amplitude extraction (AE) matrix are designed to accurately identify the adjacent and crossing frequencies. The IF trajectory is calculated using the particle swarm optimization algorithm and the generalized demodulation technique. The AE matrix extracts the energy value from the time–frequency domain by a position function, which converts the IF trajectory to corresponding time–frequency information. Moreover, an iteration procedure is designed to avoid the superposition of energy at the crossing point. The effectiveness of ADET is confirmed through simulated and experimental signals.

Power line extraction from three-dimensional (3D) point clouds is a key step for power inspection. However, the power line is intertwined and covered with trees and other objects in the urban area, which brings a tremendous challenge to its extraction. This article proposed a local-to-global method to extract power lines from mobile Light Detection and Ranging point clouds based on multiple characteristics. Firstly, initial point clouds clusters are selected through data preprocessing. Secondly, given the linearity and independence of the power line, we innovatively design a local line-plane separation modeling to divide the point clouds into high-confidence power line points, non-power line points, and unconfirmed points. Then, a novel global energy function is designed for the graph cut model according to the long linear and short radius characteristics of the power line. The proposed method has been tested on Toronto 3D and WuHan Avenue datasets. The average Precision, Recall, and F1score can reach up to 0.92, 0.73, and 0.79 respectively, which is superior to state-of-the-art approaches. The result demonstrates the efficacy and robustness of our proposed approach in extracting power lines within urban environments using a mobile laser scanning system.

In the field of power systems, the variational mode decomposition (VMD) algorithm is widely utilized in the data preprocessing stage of prediction models for wind power, photovoltaic power, load, and other related areas. However, if the two key parameters of VMD—mode number and penalty factor—are not properly set, the decomposition results may suffer from mode mixing and mode repetition. Moreover, the boundary effect issue in VMD causes numerical oscillations at the boundaries of the decomposition results, thereby leading to numerical distortion. To address these problems, this paper proposes a novel time-frequency information-based variational mode decomposition (TFI-VMD) algorithm. Firstly, TFI-VMD clusters the time-frequency information of the data using an improved density peak clustering algorithm, with the number of cluster centers serving as the mode numbers in VMD. Then, by minimizing the mode mixing energy, the optimal value for the penalty factor is determined. Finally, TFI-VMD uses decomposition results free from boundary effects to train the broad learning neural network, thereby correcting the decomposition data that contain boundary effects. Case study results indicate that the proposed algorithm can effectively identify the parameters in the original VMD and address the boundary effect issues in the decomposition results.

To enhance the cross-domain diagnostic ability of the model, domain adaptation method is adopted. When using traditional domain adaption methods to extract domain invariant characteristics of axial flow fan faults, the characteristics of the source and target domains will be close to each other, thereby the distribution of trained source domain characteristics will be changed. When the fault characteristics of the source domain gather at the classification boundary, the trained model will incorrectly classify some target samples. In addition, single source domain adaptation can lead to poor model generalization ability. To resolve the above issues, a multi-source domain adaption intelligent fault diagnosis method based on asymmetric adversarial training is proposed. In this method, the asymmetric adversarial training method is used to realize unidirectional movement of fault characteristics from the target domain to the source domain; triplet-center loss is used to expand the inter-class distance and shorten the intra-class distance of fault characteristics in the source domain; domain invariant characteristics are extracted from different source and target domains, and they are inputted to their respective fault classifiers, then aligning the outputs of each classifier using cosine similarity. To improve the cross-domain diagnostic ability of the model, a strategy of aligning weights is adopted. The industrial actual data verification results indicate that this method is effective in solving relevant practical industrial problems.

In the industrial production process, safety accidents caused by the leakage of hazardous gases from pressure vessels are common. The assessment of leakage orifice detection is crucial. Traditional detection methods have low sensitivity. This study proposes a network model called gas leakage detection net (GLD-Net) for gas leakage detection, which utilizes an array of acoustic sensors for data collection of gas leakage signals. The network is based on the Yolov5n model as the baseline, and the C3 module of the backbone network structure incorporates the swift parameter-free attention block (SPAB) structure from the fast parameter self-attention network mechanism. The C3-faster-SPAB module is designed to achieve lightweight feature extraction. In terms of the neck network, GLD draws inspiration from the efficient reparameterized generalized-feature pyramid network (FPN) concept and designs the vision transformer with BI-level routing attention block (Biformer-Block), which is suitable for capturing narrow-band features of gas leaks. It integrates the lightweight biformer cross stage partial stage (Bi-CSPStage)module to enhance the expressive power of the FPN. The network replaces traditional convolution operations with depthwise separable convolution and reduces the scale of detection in the neck network. Experimental results show that after adjusting the depth of the network, the parameter size and computational complexity of GLD-Net are reduced by 96.7% and 96.1%, respectively. It exhibits higher average accuracy than other lightweight model algorithms on datasets with different noise levels, and achieves a frame rate of up to 500 FPS. It possesses fast and accurate detection capability, providing a reference basis for real-time leakage detection in industrial production and manufacturing.

Accurate wind speed and direction measurement is crucial in many environmental applications, but achieving reliable results in the presence of mixed noise, such as impulse and Gaussian noise, remains a significant challenge. Traditional methods often fail to maintain accuracy under these conditions. This paper proposes a co-prime array wind measurement method based on fractional low-order eigenspace beam forming (FL-ESB) algorithm. The proposed method adopts a co-prime arc ultrasonic sensor structure consisting of two sets of uniform arc subarrays. This structure can increase the array aperture and improve the measurement accuracy. Combined with the idea of fractional low order covariance, the FL-ESB algorithm is proposed to realize the effective suppression of impulse and Gaussian mixed noise. The estimated variance and wind parameters' Cramer Rao Bound are obtained through theoretical analysis and simulation. Simulation results show that under a signal-to-noise ratio of −4 dB, the root mean square error of wind speed is less than 0.6 m s⁻¹, and the root mean square error of wind direction angle is less than 2^∘. These results confirm that the proposed method effectively measures wind speed and direction in mixed noise environments.In conclusion, the FL-ESB-based co-prime array method provides a robust solution for accurate wind measurement, offering improved noise suppression and accuracy compared to traditional methods.

Rail damage can pose a tremendous hazard for high-speed trains, making damage diagnosis critical in the field of engineering. Currently, deep learning enables an end-to-end approach for rail damage diagnosis. However, the training and test data in real applications are often out of distribution, and the test data may even represent fault categories that were previously unseen. To address this situation, an unseen damage diagnosis framework (UDDF) that effectively embeds the mechanism damage features from the simulation signals of all possible damage categories has been proposed. In particular, the mechanism-embedded generative adversarial networks in the UDDF utilize a hierarchical embedding technique to ensure the stability of the mechanism embedding process. In addition, a k-means clustering discriminator uses an unsupervised method to guarantee the minimum intra-category sample spacing of the generated unseen categories. After the generation of all types of damage categories, the generated and existing original data are included as a new dataset for the training of a diagnostic model. The trained diagnostic model can perform classification tasks without acquiring all types of damage signals in real situations. Finally, the effectiveness of our proposed diagnostic framework is validated through comparative and ablation studies on a dataset that contains finite element simulation and experimental data of ultrasonic guided wave signals with damage at different locations and depths of rails.

As an essential part of the train, the health status of the bogie bearing is vital to its safety. In a complex working environment, the fault signal of the bearing is highly dimensional and nonlinear, which makes high-precision fault diagnosis difficult. Therefore, a fault diagnosis method based on improved complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and adaptive weighted distance manifold learning (AWDU) is to address this issue. Firstly, an intrinsic modal component screening index is presented, which takes into account the advantages of the cross-correlation coefficient, mean square error, and kurtosis coefficient. It is embedded in CEEMDAN to improve further the algorithm's signal decomposition and noise reduction ability. Secondly, the AWDU technique is proposed to reduce the dimensionality of raw data, which can accurately calculate the distance and adjacency relationship between high-dimensional data and better guide low-dimensional embedding. Eventually, a range of experiments was executed across two different benchmark cases, and the results indicated that the average accuracy of this method in each experiment was not less than 99.64%, and the performance was satisfactory in the comparison of similar methods, which verified the effectiveness, stability, and generalization of the proposed method.

Research shows that multi-system synergetic stochastic resonance (SR) can obtain preferable performance for weak signal detection than a single SR system. However, how to design or select suitable multi-system synergetic mode remains a challenge. Therefore, an underdamped Duffing bistable SR system embedded with overdamped feedback control is constructed to boost the poor detection effect of weak signals by a single resonant system. Firstly, SR phenomenon in the underdamped Duffing bistable system embedded with overdamped feedback control is analyzed. Through the information exchange between the two subsystems and the series-parallel relationship of multiple integrals, it is feasible to enhance the noise utilization as well as to obtain a larger output signal-to-noise ratio. Secondly, the effects of feedback and damping coefficients on the resonant system are investigated, and an adaptive multi-parameter optimization SR algorithm is put forward in combination with gray wolf optimization algorithm to detect weak signals in a highly noisy background. Finally, vibration data of two faulty bearings is applied in the analysis of the investigated algorithm, and the results of comparison show that the spectral peak corresponding to the characteristic frequency of the output signal derived from the investigated method is more prominent, which effectively promotes the detection performance of bearing faults.

High-speed railway turnouts are susceptible to failures due to the intricate structure. The interaction forces between wheels and rails reflect the health condition of turnouts. Hence, force signal analysis emerges as a robust technique for the turnout health assessment. A crucial aspect of this analytical process is the accurate identification of signal segments within the measured force data that are originated specifically from turnouts. This identification is vital as the locations of force signals measured often misalign with the recorded positions of the turnout. Given the sparsity of the force impacts induced by the frogs of turnouts, this paper proposes a Box-Cox enabled oscillatory sparse decomposition (BC-OSD) method. Given the sparse and low oscillatory nature of signal segments produced by frogs, the BC-OSD proves effective in isolating frog-generated features from complex force signals. BC-OSD employs Oscillatory Sparse Decomposition to partition the signal into two segments based on their oscillatory properties. The integration of the Box-Cox Sparse Measures enables the algorithm to adaptively adjust its performance of subbands selection based on the actual signal characteristics. The decomposition outcome relies upon the chosen parameters; thus, a systematic discussion on the parameter selection for processing wheel–rail force signals is provided. The efficacy of the proposed method is validated through the analysis of measured wheel–rail force signals.

The prevalent issues of small samples and heavy noise in industrial settings have severely limited the application effectiveness of intelligent diagnostic methods. To address these challenges, an efficient feature-focused lightweight capsule network is proposed. Firstly, a plug-and-play lightweight Ghost multiscale convolution block is designed to integrate multiscale convolution with gating mechanisms through a multi-branch architecture, enabling multi-scale feature fusion and enhancing the model's noise robustness. Secondly, an efficient feature-focused routing mechanism is proposed to optimize information transfer between capsules through multi-step feature reconstruction and feature focusing strategies, which substantially enhances generalization capabilities in small-sample scenarios. Finally, a dual-norm frequency-domain regularization loss function is designed to leverage the complementary advantages of first-order and second-order norms, enhancing both feature extraction capability and reconstruction quality. Extensive experiments on two distinct datasets demonstrate that EFLightCaps achieves superior diagnostic performance and computational efficiency in scenarios with small samples and heavy noise.

For outdoor scenarios, millimeter-wave radar has attracted considerable attention due to its advantage of being less affected by lighting and weather conditions. In this paper, we propose a road dynamic target recognition method using a low-cost millimeter wave radar. We first track targets to estimate the global trajectory of each target, and then extract effective local trajectories from them to remove the interference of non-attention targets. Subsequently, we cut the effective local trajectories to estimate the speed and approximate size of the target. Finally, we utilize the support vector machine by exploiting the estimated speed, target size and some other features to classify the targets into people, bicycles and cars. Moreover, a real-time low-cost system is designed to implement the proposed method. Experiment results show that the proposed method is robust to the installation height and angles, and can achieve an average accuracy of 96.19% for triple classification.

Domain adaptation methods have demonstrated considerable potential in enhancing the performance of fault diagnosis models across diverse operating conditions. While single-target domain adaptation has been extensively explored, its efficacy is reduced in scenarios involving multi-target and incremental-target domains. However, such scenarios are common in real-world projects, necessitating the exploration of more sophisticated, multi-domain, and incremental domain adaptation methods. In response, we propose an iterative-cluster domain adaptation (ICDA) method, which is tailored to address incremental multi-target domain adaptation challenges. On the basis of single-target domain adaptation, ICDA enhances domain alignment through cyclic iteration and clustering algorithms. Notably, it has been demonstrated to be effective in a range of domain adaptation scenarios, including single-target, multi-target, and incremental-target adaptation, even in the unlabeled target domains. The efficacy and superiority of ICDA over baseline models has been substantiated through validation on the CWRU and WINDEY datasets.

Efficient and accurate surface defect detection is a crucial task for industrial production, where traditional computer vision methods often fail to reliably identify subtle manufacturing defects. This paper proposes a novel hybrid supervised learning-based dual-task guided network (DTGNet) model that combines the strengths of supervised and unsupervised learning techniques to address the limitations of traditional approaches. By integrating partial convolution, scalable convolution and content-guided attention fusion (CGAFusion), our model improves recognition of complex defect features and achieves state-of-the-art performance on benchmark datasets. Experimental validation on the KSDD2 dataset empirically demonstrates DTGNet's superior performance, with Lite DTGNet achieving a 93.51% accuracy. This innovative approach not only advances industrial quality control technologies but also provides a flexible, data-efficient framework.

Since deep learning has been introduced into the field of intelligent fault diagnosis it has made significant accomplishments with large amounts of data. However, in practical industrial settings there is a general lack of labeled data and operating conditions are not stable, therefore existing trained models ignore these problems and diagnostic accuracy and generalization are severely degraded. Therefore, this study proposes a multiple-prototype, domain adversarial network for fault diagnosis of rotating machinery, especially bearings, to address the issues of domain distribution shift and a shortage of labeled samples from the target domain. Firstly, the proposed method adopts a residual network with 12 layers (ResNet12) as the encoder to extract distinct features from the target domain and source domain data. Then, a domain-invariant representation module, using the domain adversarial method, is introduced to bridge the discrepancy between a source and target pair domain. Furthermore, a class typicality weight module is designed to calculate weights by constructing multiple prototypes of the source and target domains to improve the generalization of the model. Extensive experiments are executed on two datasets with variable working conditions to test and verify the feasibility and superiority of the proposed method.

The two-stage Wiener process (WP) model has become a common method to describe the phased deterioration of bearings over time. However, this model ignores the correlation of feature data distribution structure and change points (CPs) between the two stages, as well as the limitations of maximum-likelihood estimation methods for WP model parameter estimation. Therefore, this paper proposes a remaining useful life prediction approach that integrates feature distribution CP identification and a martingale process. First, a two-step feature screening method adopting trend consistency and composite score is proposed to construct a health indicator, which accounts for the trend consistency of the same feature on different bearings and can avoid redundancy while containing sufficient degradation information. Next, a t-neighborhood granular mean-shift clustering method is proposed, which makes the divisibility of the feature distribution more obvious and can identify CPs sensitively, flexibly and stably. Finally, a martingale method is introduced so that the parameter estimation of the two-stage WP model depends on the entire degradation path, which overcomes the limitations of WP model parameter estimation and enables the model to better characterize the bearing degradation process.

Modal identification and extraction of monitoring signals are of great significance for aero-engine condition monitoring and health management. Empirical mode decomposition (EMD) method and variational mode decomposition (VMD) method have wide applications in time–frequency signal analysis, but EMD method is very sensitive to noise and sampling frequency. The accuracy of the VMD method relies on the input parameter settings. The dynamic mode decomposition (DMD) method is considered as a high-dimensional data-driven method which can predict the short-term behavior. However, the one-dimensional data decomposed by the DMD method only contains one mode. In this paper, the DMD method based on Hankel reconstruction (HDMD) is applied to the signal sequence to investigate its capability and potential as a signal processing method. First, multi-frequency, similar-frequency, and variable-frequency-variable-amplitude signals are constructed and resolved by various methods. It is shown that the HDMD method is more robust to noise, has a higher frequency resolution, and is capable of uncovering potential frequency trends compared to the EMD method and the VMD method. In addition, this paper applies the HDMD method to identify the measured signals of the aero-engine. It is shown that the HDMD method can accurately decompose the frequency modes of the core components. Compared with the EMD and VMD methods, the HDMD method can reveal the pattern of the engine frequency when it varies within a small range, which is more applicable to the analysis of the actual signals.

The non-Gaussian and other complex forms of data distribution for the speed and load of different bearings are due to various nonlinear factors. When processing data, marginal distribution only focuses on the distribution differences on a single layer and cannot fully capture the joint distribution differences between multiple layers. To solve this problem, this paper proposes a bearing multi-condition fault diagnosis method based on joint adversarial adaptation network. First, joint distribution adaptation is designed to transfer data, establish connections between distribution constraints between layers, form a unified joint distribution, and calculate the joint distribution difference between each layer of the network. Then, the adversarial training method is adopted in the forward propagation process, the target domain data enters the domain classifier through the feature extractor, where it is confused with the source domain. Finally, a dropout layer is added to the feature extractor part to modify the neurons of the hidden layer. The weight update no longer depends on the joint action of fixed relationship nodes, which prevents the fault feature from working only under specific nodes. To have a better generalization performance of the model, different bearing speeds and loads are used for experiments. The experimental results show that the proposed method improves the accuracy by 14.27% and 8.07% respectively compared with other methods.

To achieve rapid and precise identification of water supply pipeline leakage faults, this study introduces a diagnostic framework that integrates multisource multiscale attention entropy (MMATE), an enhanced least-squares twin support vector machine (ELSTSVM), and an adaptive boosting (AdaBoost) algorithm. The MMATE-ELSTSVM-AdaBoost architecture follows a three-stage workflow. First, an improved wavelet threshold denoising (IWTD) technique featuring a novel threshold-processing function is developed to suppress signal noise. Second, MATE values are extracted from the denoised signals, and a multidimensional feature vector is constructed by integrating MATE characteristics across multisource sensors. Third, the ELSTSVM's objective function is enhanced through a reformulated distance metric, augmented with regularization and compactness terms to optimize model robustness and generalizability. The refined MMATE feature vectors are then fed into the ELSTSVM-AdaBoost ensemble for fault classification. Experimental evaluations demonstrate that the proposed framework significantly outperforms conventional single-feature approaches, standalone models, and state-of-the-art SVM variants, achieving a peak classification accuracy of 99.10% in leakage detection tasks.

With the fast growth of embedded and handheld intelligent devices, the demand for lightweight fault diagnosis (FD) models of wind turbine gearboxes (WTG) is increasing in actual industrial scenarios. The existing lightweight FD approaches suffer from poor diagnostic performance and low robustness in complex operating environments. In this paper, a FD approach is proposed based on a new lightweight FD network (LFDNet) for high diagnostic performance of WTG. In LFDNet, the stem module, stage module and downsample module are respectively redesigned to reduce the model complexity and maintain superior and robust diagnostic capabilities, which enable it to diagnose the faults of WTG precisely, even with noisy interference. In the validation experiments on a self-made dataset and an open-source dataset, the diagnostic accuracy achieves 99.41% and 99.69%, respectively, which is higher than multiple lightweight deep learning methods. In addition, the anti-noise experiment also shows that the proposed approach has good noise robustness.

Mechanical condition monitoring of power circuit breakers (CBs) is vital for predictive maintenance, enabling early detection of mechanical issues. However, traditional offline testing methods require the CB to be taken offline, which is inefficient and incurs high maintenance costs. This paper introduces a novel online monitoring scheme for detecting CB contact timing and asynchronism using an enhanced dual threshold method-phase space spectral radius (DTM-PSSR). Vibration signals are initially processed using the DTM for variational mode decomposition. The signal component with the highest correlation to the original signal is selected for PSSR reconstruction. The reconstructed energy spectrum identifies the energy peak, used to calculate asynchronism and contact timing. This method employs a single accelerometer to directly capture relevant parameters without the need to take the CB offline, providing a practical, non-invasive solution for online CB monitoring and streamlining the maintenance process.

Accurate detection of the size of pipeline leaks is the key to minimizing economic and resource infrastructure loss. However, extracting critical information about different leak sizes from complex pipeline signals is a major challenge. To address this difficult problem, this paper proposed a water pipe leakage detection method based on a Dual-domain feature fusion attention network (DFFAN). It employs time series imaging to fuse images generated under different transform domains of the original signal to achieve high-accuracy detection of leak size. Specifically, by fusing the Gramian Angular Field (GAF) matrix with the Continuous Wavelet Transform (CWT) image of the original signal, DFFAN can comprehensively capture the temporal, amplitude, and time-frequency information in the signal. The designed attention fusion module can optimally allocate weights to features according to their importance and further enhance the leakage features through weighted fusion. The effectiveness of the method is evaluated using the collected leakage data. The experimental results show that the proposed method can effectively recognize the four leakage states in two material pipes. It has better detection performance compared with advanced pipeline leakage detection methods.

How to achieve accurate predictions in a few-shot scenario with limited data is currently a pressing issue. It requires a thorough excavation of domain differences to extract embedded information, improve model generalization, and clarify the decision-making processes, thereby enhancing model credibility and feasibility. To achieve this goal, this study combines the advantages of knowledge-driven and data-driven model, and proposes a knowledge and data tandem-driven algorithm based on deep feature decoupling network (DFDN) to solve the few-shot prediction problem. This method offers both a mechanistic explanation of the employed model and the ability to address few-shot prediction problems. First, the cross-domain metrics of the few-shot data are obtained by an auto-encoding DFDN, and a multi-stage degradation model is constructed. Particle filtering (PF) algorithms and sampling importance resampling techniques are then used to identify the model parameters. The remaining useful life of the bearing is derived through the forward inference of PF and compared to the results of the conventional model for validation. Experiments indicate the superiority and effectiveness of the algorithm approach in few-shot prediction.

The non-stationary operating conditions of wind turbines, along with adverse environmental factors, frequently cause the rotation shaft to move abnormally, resulting in periodic impulse interference and noise. There are significant challenges for the accurate diagnosis of wind turbine bearing faults. Box-cox sparse measures (BCSM) deconvolution is a validity method for machinery fault diagnosis because it is significantly for reducing noise and eliminating the interference of the system transmission path. The improper selection of the initial filter can result in the failure of BCSM deconvolution (BCSMD) to effectively extract fault characteristics of wind turbine bearings subjected to strong periodic interference. To overcome this limitation, a simulation-guided BCSMD method is proposed. In the initial step, a finite element method simulation is conducted to identify the potential carrier frequency derived from the first flexural frequency in modal analysis. Subsequently, this carrier frequency is employed as the central frequency, along with a predetermined filter length, to ascertain the appropriate filter. Finally, the simulation designed filter is used as the initial filter of BCSMD to filter the raw signal. Compared with the original BCSMD method, its main contribution lies in the organic integration of a finite element simulation with filter design, thereby effectively preventing the iteratively derived filter from converging to non-fault frequency bands. Experimental results indicate that simulation-guided BCSMD can precisely identify fault characteristics even in the presence of periodic impulse interference.

The vigorous development of data-driven methods has promoted the application of intelligent fault diagnosis technology in various manufacturing industries. However, it is difficult for the model to obtain satisfactory diagnosis results and generalization performance with small samples under variable working conditions. To solve these problems, a new prototype discriminant network based on domain contrast learning is proposed, which has self-supervised few-shot cross-domain fault diagnosis capability. First, sample pairs are constructed based on differences in data domain distribution. The domain-invariant features between classes are extracted by increasing the distance between classes and reducing the differences within classes using unsupervised training. Then, a prototype discriminant network is used to accurately diagnose under few-shot and variable working conditions. To realize accurate diagnosis in two typical rotating machinery diagnosis cases of bearings and gearboxes, the performance of the proposed framework is verified, and higher diagnostic accuracy and generalization performance are achieved compared to existing methods.

Weak magnetic testing (WMT) is suitable for evaluating the girth weld stress due to its advantages in non-contact detection. However, the stress-induced magnetic signal collected by WMT is prone to interference from external noise signals. Therefore, weak magnetic excitation is proposed to enhance the strength of stress-induced magnetic signals. In this paper, the enhancement mechanism of the weak magnetic excitation on the stress-induced magnetic signal is revealed from the perspective of magnetic domain energy. Then, a tensile experiment is carried out to analyze the stress-induced magnetic signal variation law of cover welding, filler welding, and root welding layers under different excitation magnetic fields. Finally, the stress-magnetic coupling model of weak magnetic excitation at the pipe three-dimensional stress state is established, which is validated by a full-scale pressurized pipe test. The results illustrate that the excitation magnetic field will modify the strength and variation features of stress-induced magnetic signals of weld joints. The strengthening effect of the excitation magnetic field is to first increase and then decrease. There is an optimal magnetic field that can make the stress-induced magnetic signal of different welding layers change synchronously. In addition, the optimal magnetic field value at a three-dimensional stress state is greater than that under single axial stress.

Support matrix machine, as an effective classification method, is widely used in single task fault diagnosis. However, for the entire mechanical equipment system, the state information between different components is coupled with each other, and it is difficult to fully express the completion state information of a task by only constructing a diagnostic model for that task. In view of this, this paper proposes a multi-task collaborative enhancement matrix machine (MTCEMM) method. First, a dimension enhancement term is defined, in which nonlinear offsets are used to help the model capture nonlinear relationships in the data, thus improving the information fusion power of the multiple models. Then, the classification hyperplane for different tasks is constructed simultaneously, which completes information sharing among multiple tasks during the construction process, thereby achieving collaborative enhancement of multi-task sample information. Finally, the MTCEMM method is applied to multi-task fault diagnosis of mechanical system, and five evaluation indicators are also selected to display the classification results.

The laying trajectory of saline–alkali land subsurface pipes for salt discharge plays a crucial role in assessing the efficiency of saline–alkali land reclamation. Due to the complex soil conditions of saline–alkali land and the physical properties of the subsurface pipes, directly reconstructing the trajectory using an inertial measurement unit (IMU) often fails to achieve sufficient accuracy. To address this issue, this paper proposes a trajectory reconstruction method for saline–alkali land subsurface pipes based on an improved extended Kalman filter (EKF) and adaptive gradient optimization. This method integrates accelerometer and gyroscope data using the proposed dynamic weighted threshold EKF (DWTEKF). The dynamic weighted threshold control module adjusts the noise covariance matrix to reduce the impact of inaccurate initial state information on subsequent attitude estimation. Meanwhile, this paper proposes an adaptive gradient descent algorithm with suppression factors and dynamic step sizes (AGDSD). This algorithm adaptively adjusts the step size to optimize the gradient direction estimate and incorporates a suppression factor through discriminant testing to reduce the influence of pipe oscillations, thus improving the robustness of attitude estimation and trajectory reconstruction. Experimental results show that the DWTEKF algorithm exhibits excellent performance in noise suppression and signal fidelity. Compared to traditional adaptive gradient descent algorithm (AGDA) and standard gradient descent algorithm (GDA), the AGDSD algorithm improves the average attitude error by 44.02% and 57.31%, respectively. Furthermore, the proposed trajectory reconstruction method significantly enhances trajectory fitting accuracy, successfully validating the effectiveness of the method.

To handle the influence of historical data on state estimation, this study proposes a correntropy-weighted robust Kalman filter (CWRKF). The proposed method investigates the weight using the correntropy based on the current residual, thus suppressing the impact of historical data on system state estimation. In addition, an adaptive kernel bandwidth is defined for correntropy to avoid excessive use of the current residual and accuracy degradation. Further, the correntropy-weighted system noise covariance estimation is obtained and embedded into the Kalman filter. The proposed CWRKF is verified by simulations, and the results demonstrate the superiority of the proposed CWRKF in state estimation under time-varying system noise covariance conditions over the existing approaches.

Planetary gearboxes generally operate under variable speed conditions in response to actual industrial production requirements, which is prone to cause single or compound faults. The planetary gearbox fault signal exhibits time-varying, weak and complex characteristics under varying conditions, and common diagnostic approaches are difficult to effectively extract and reveal fault-related features. In addition, the computed order tracking requires a speed signal as an auxiliary calculation. Unfortunately, due to economic and installation space limitations, encoders and tachometers are not always available, which poses a huge challenge for extracting variable speed planetary gearbox faults. To tackle these challenges, a novel tacholess order tracking method based on improved adaptive chirp mode decomposition (IACMD) and adaptive maximum second-order cyclostationarity blind deconvolution (CYCBD) is proposed in this contribution for diagnosing planetary gearbox single and composite faults. First, the IACMD algorithm is utilized to adaptively decompose the fault signal, which constructs a composite index (CI)-based signal mode selection and recombination scheme to perform optimal modal decomposition. Second, the instantaneous dominant meshing multiply trend line with obvious amplitude advantage is extracted from the time–frequency representation of the original fault signal, and then converted to the reference shaft rotation frequency to perform angle domain resampling on the sensitive model with the highest CI value. Third, the adaptive CYCBD is adopted to deconvolute the angular domain sensitive modal to heighten the weak fault signatures. Finally, the envelope spectrum of the improved signal is utilized to identify prominent fault characteristic orders and ascertain failure types. Both numerical simulations and practical engineering data for different planetary gearbox failure cases have been thoroughly examined to demonstrate the correctness and feasibility of the proposed approach. Moreover, comparison with some existing technologies demonstrates the superiority of the proposed approach.

Hydroelectric power plants (HPPs) are critical for sustainable energy generation, but their maintenance and operational stability are often compromised by structural vibrations, particularly in key components like the discharge ring units. Predicting these vibrations in advance is essential to prevent damage, enhance operational efficiency, and extend the lifespan of HPP components. This paper presents two advanced deep learning models designed to predict future vibrations in the discharge ring of river-type HPPs. By combining multiple deep learning architectures, the proposed models process complex sensor data to accurately predict vibration patterns. The models employ the hybrid compositions of deep learning models specifically optimized for time-series prediction of mechanical stresses. In this study, vibration patterns of five distinct HPP turbine units (TUs) are modeled with a hybrid approach and comprehensive analyses are provided for each TU. Validation of the developed models with real-world operating data from HPPs reveals the proposed models' accuracy, resilience, and potential for predicting future vibration signals. The proposed models achieve significant improvement in predictive accuracy over traditional methods, providing a reliable tool for early detection of vibration-induced risks in hydroelectric power infrastructure. The proposed models achieved minimum error rates with mean absolute error (MAE) of 0.025, mean squared error (MSE) of 0.006, R² of 0.999 and root mean squared error (RMSE) of 0.080 for convolutional neural network + bidirectional long-short term memory (CNN + BiLSTM) and MAE of 0.038, MSE of 0.008, R² of 0.994 and RMSE of 0.089 for CNN + gated recurrent unit. This study contributes to advancing predictive maintenance in HPPs and offers a scalable solution for enhancing the safety and resilience of renewable energy facilities.

Accurate prediction of the remaining useful life (RUL) of rolling bearings is a challenging task due to the complexity of time series data, inherent uncertainty in predictions, and domain shift between source and target datasets. To address these challenges, a parallel deep learning framework is proposed that integrates transfer learning and uncertainty quantification for RUL prediction. The framework combines AlexNet with the convolutional block attention module and Transformer with a gated convolutional unit to effectively extract degradation features from time series data. Additionally, Bayesian optimization is employed for hyperparameter tuning, reducing the need for manual adjustments, while a domain adaptation module addresses distribution differences between the source and target domains. Variational inference methods extend the model to a Bayesian deep neural network, providing confidence intervals for uncertainty quantification in RUL predictions. Experimental results on two rolling bearing datasets demonstrate that the proposed CANN-GT-BDA model outperforms state-of-the-art models in terms of predictive accuracy, offering a reliable and uncertainty-aware solution for industrial predictive maintenance.

Characterization of the dual-impact phenomenon is an effective way to evaluate the extent of damage in a cylindrical bearing, and thus the bearing's remaining life. Therefore, it is of great importance to conduct in-depth research and achieve through understanding on the impact behaviors during the defect development process throughout the life cycle of the bearing. In this paper, a nonlinear dynamic model considering defect evolution is established for cylindrical bearings. The model integrates non-penetrating and penetrating damage models to simulate the continuous evolution of inner and outer raceway damages, overcoming the limitations of traditional single-damage modes. Its automatic switching mechanism ensures uninterrupted simulation across multiple damage modes, enabling dynamic damage propagation analysis. The size and severity of raceway defects during the bearing's life cycle can be accurately inferred by analyzing the timing of vibration responses with defect estimation methods. This enables precise assessment of the bearing's health status. To validate this, fault bearings with varying damage types and sizes were manufactured and tested. The vibration signal analysis results have shown a good simulation-experiment agreement. The developed model provides a good candidate for digital twin modeling for the roller bearing condition monitoring.

This paper proposes a switched control strategy for heavy-duty vehicles, aiming to improve ride comfort and roll safety in unstructured road conditions. The realization of the comprehensive control target depends on the active suspension system: on the one hand, it can suppress the vibration caused by road excitation; on the other hand, it can dynamically adjust the output vertical force of the actuator to achieve the purpose of anti-roll. Focusing on the important influence of complex road conditions with road curvature and inclination on stability of heavy-duty vehicles, a vertical-lateral coupling nine-degree-of-freedom model is established. Then, a preview-based Model Predictive Control switching controller is designed to make full use of road information, which allows the controller to be solved in advance, thereby improving the real-time performance. In addition, variable weight coefficients are used to realize the switched control of ride comfort and roll safety based on the preview road information. Finally, the simulation results givened by Matlab/Simulink show that over the one without control, the proposed controller improves the ride comfort by about 53.2% and the roll safety by about 45.2%.

Large steel structures inherently have errors such as clamping and continuous thermal deformation, which complicate traditional robotic multi-layer multi-pass (MLMP) welding of medium-thick plates to ensure quality, efficiency, and universality. To address these conditions, this paper proposes an adaptive process and path planning method for MLMP welding of V-grooves utilizing line laser displacement sensors. First, dimensions and profile of the groove are obtained, feature points are extracted, and the robotic path and welding process for the root welding are planned based on the V-groove's feature information. Then, reference process parameters is used for filling and cover welding. After each pass/layer, feature points of the V-groove will be extracted, and the previous pass/layer formation will be analyzed. And then algorithm proposed in this paper will be used to adjust the process and robotic path to ensure that, before cover welding, the average groove depth will remain within a specific range. Finally, MLMP experiments have been conducted on two V-grooves of different sizes. The results indicate that this method effectively fills the groove. Maintain the average remaining depth of the groove between 1 and 2.5 mm before cover welding, meeting the requirements of different sizes of V-grooves. The entire surface and cross-section of the weld seam are free of defects, and the fusion between the weld seam and the sidewall exceeds 1.5 mm, meeting the requirements for industrial use.

Data based fault diagnosis technologies are important measures to improve the operation safety, stability, and reliability of manufacturing processes, which are key entry points and innovation powers to promote the intelligent manufacturing as well as improve the operation efficiency. Class-balanced datasets are often used for modeling and fault diagnosis by the traditional data based methods. However, in practical engineering applications, manufacturing processes often produce multiple classifications of imbalance data, bringing great challenges to the promotions and applications of the classical data based methods. To this end, an intelligent fault diagnosis framework is proposed for manufacturing processes, of which the issues on the intraclass and interclass imbalance have been specially focused. Specifically, considering the non-independently identically distribution characteristics among different fault data and the low recognition rates of minority fault samples, a new cost sensitive convolutional neural network is constructed as a base classifier by coordinating the cross entropy loss function with a specific cost sensitive index. Subsequently, a federated learning based aggregation algorithm is designed to optimize the participation weights of local classifiers with the purpose of cooperating with multiple classifiers to improve the model generalization performance. Finally, the validity of the proposed framework is demonstrated by a typical hot rolling process with different forms of imbalance fault data. The simulation results show that superior diagnosis performance can be achieved compared with some comparative algorithms in each scenario.

The underwater environment is characterized by its inherent complexity and dynamics, leading to substantial interference with the precision of underwater measurement data. To enhance the precision of underwater data measurements, underwater robotic platforms necessitate improved motion and stability characteristics. As amphibian mammals, beavers possess excellent amphibious abilities and a wide range of environmental adaptability. Based on the observation of biological morphology and hind limb fin structure of beavers, this study analyses their swimming mechanism and designs a beaver-like robot. This study introduces an efficient control algorithm designed for a beaver-like robot platform. The algorithm integrates reinforcement learning with conservative Q-learning, model-based policy optimization and deep Q-network methods to facilitate offline training of the robot. A training weight allocation system is employed to enhance adaptability across diverse swimming conditions in the complex underwater setting. Simulating the robot's underwater environment, the algorithm has demonstrated effective training in both speed and stability. The pitch angle is successfully stabilized between −0.245 and 0.305 rad, while the robot's speed reaches up to 0.38 m·s⁻¹.

It is still a challenge to evaluate the reliability of displacement control systems (DCS), which are widely used in extreme scenarios with multi-component degradation, external shock impacts, measurement noise and disturbances, and have closed-loop self-compensating capability. This paper proposes a performance-based reliability model for DCS, which considers the challenges above to ensure the safe application of DCS within a certain fault-tolerant range. Specifically, the coupled relationship between degradation and shock is characterized by the energy-related additional degradation amount, and the effects of sensor measurement error and external environmental disturbance on the system performance are taken into consideration. A nonlinear hybrid model of DCS is established by integrating the cumulative degradation amount as an extended state variable; then, based on the control performance, a reliability index is defined and a general performance-based reliability evaluation method is proposed. The method is validated through a case study on a hydraulic servo control system, demonstrating its ability to accurately describe the performance changes of DCS under various degradation interactions, to provide compensation capability analysis and to more accurate reliability evaluation.

The pre-identification of impulse road surfaces, such as speed bumps, potholes, and manhole covers, can significantly enhance the performance of active suspension systems. However, existing current identification methods often fail to balance between low cost, high reliability, and precise speed. This study proposes an impulse road surface identification method based on a reinforcement learning (RL) network. First, a real dataset of impulse road surfaces was collected, with suspension dynamic responses serving as inputs and random road levels along with impulse road surface features as outputs. This data was used to develop a random forest extreme gradient boosting (RF-XGBoost) network to recognize road surface information from vehicle dynamics. Next, a semantic segmentation network was employed to segment road surfaces during intelligent vehicle travel, using road surface information identified by the random forest network as the reward function for RL. The RL policy was pre-trained using the actual collected data. To validate the effectiveness of the proposed control strategy, a simulation environment was constructed in Prescan, where speed bumps, potholes, and manhole covers of varying heights and sizes were randomly arranged. The RL-based road surface identification algorithm was implemented in Simulink, and a co-simulation was conducted with the CarSim vehicle model. The enhanced RF-XGBoost network effectively distinguishes between random and impulse road surfaces, achieving average recognition accuracies of 95.7% for random surfaces and 98.2% for impulse surfaces. During initial training iterations, the RL network exhibited lower accuracy; however, as training progressed, the accuracy for unfamiliar road surface features reached an average of 96.5%, and overall recognition accuracy improved by an average of 9%. The simulation results demonstrate that the proposed RL identification method effectively acquires information about the road ahead, shows robustness and generalization, and provides crucial disturbance data for subsequent active suspension control.

An audit trail check is expected to be incorporated as a new task for the verification of a measuring instrument in legal metrology. To prepare for this incorporation, an idea of an audit trail check using runtime verification (RV) and abstract formal specifications were proposed. The idea was to allow an RV tool to judge whether an audit trail satisfied the check formula, and the specification was a template for the check formula. In the present study, we demonstrate this idea. The target of our demonstration is a weighing instrument, which is only defined in a design document and not implemented. We adopt a first-order temporal logic to describe the check formulas and use MONPOLY as an RV tool. For the demonstration, we generate an audit trail of the instrument based on the behavior specified in the design document, derive check formulas from the abstract formal specifications, and then run MONPOLY. This study presents not only the demonstration, but also both our audit trail design for generating an audit trail and the derivation of check formulas. They will be helpful in the implementation of audit trail checks using RV for various measuring instruments.

To address the issue of fruit stacking and obstructing target fruits during daily fruit sorting, this paper presents a novel object detection algorithm that leverages efficient multi-scale grouping and enhanced multi-head self-attention. The proposed target detection algorithm is based on real-time detection transformer (RT-DETR) as the baseline to simplify optimization challenges and enhance robustness. Additionally, we introduce efficient multi-scale attention to preserve channel information, optimize the multi-head self-attention, and adopt cascade grouping to reduce computational redundancy. Furthermore, we use a new loss function (Inner-MPDIoU) combined with a bounding box similarity comparison metric (MPDIoU) and inner idea to enhance the accuracy of detecting moving occluded targets. Experimental results demonstrate that the optimized RT-DETR algorithm achieves an average accuracy of 96.3% in detecting moving stacked fruit models with a detection speed of up to 67 FPS. This confirms the effectiveness of our algorithm in matching and recognizing blocked fruit targets, surpassing common algorithms for recognizing obstructed targets.

As a green and renewable energy source, wind energy experienced considerable growth, offshore wind power technology is increasingly valued by nations worldwide. Planetary gearboxes are critical but failure-prone components in wind turbines, which present a significant risk to the operational stability of the entire system. To address the issue of noise interference in the vibration signals of gearboxes under extreme conditions, this paper proposes the VMTransformer model, which integrates, variational mode decomposition (VMD) and the multichannel transformer for the fault diagnosis of planetary gear in wind turbine generators. The noise in the vibration signal is decomposed and reconstructed in the frequency domain using the VMD, and the significant frequency information is retained for signal noise reduction. The results of the decomposition are subjected to multi-scale feature fusion over multiple channels, which can maximize the frequency domain information provided by the VMD to further refine and enhance the model's ability to extract important features in the signal. Efficient channel attention is incorporated into the transformer model to suppress redundant information, enhancing the extraction of relevant information from each channel, and improving both the model's stability. The experimental results demonstrate that the model proposed in this paper obtains an accuracy of 98.62%, which is a significant improvement over other models, and the model also performs well in noisy environment.

When rolling bearings work under variable speed conditions and experience compound faults, their vibration signal components become more complex, some of which coexist in a narrow frequency band, and are easy to interfere with each other and coupled together. In some cases, the tachometer cannot or is difficult to install, resulting in insufficient speed information, further increasing the challenge of fault diagnosis. Aiming at the above problem of compound fault diagnosis in variable speed without tachometer, a compound fault diagnosis method of rolling bearings in variable speed based on adaptive chirp decomposition and generalized demodulation is proposed to solve the problem of determining fault components in time–frequency ridges extracted by ridge search algorithm. Firstly, the fast path search algorithm and ridge extraction method are used to fit the rotation frequency equation, and then the instantaneous frequency equation and phase function are obtained by combining the bearing fault characteristic coefficients. Then the signal components and the optimized instantaneous frequency ridge are obtained by adaptive chirp mode decomposition. Finally, the signal is stabilized by generalized demodulation. The method can optimize the initial instantaneous frequency of the vibration signal, accurately extract the fault characteristic components, and ensure that the time-varying fault characteristic component has a high energy concentration after generalized demodulation, and the result is not interfered by other signal components. Experimental analysis shows that the proposed method can accurately extract the compound fault characteristics of rolling bearings.

Intelligent vehicles motion control system is a kind of nonlinear system with many heterogeneous disturbances such as modeling inaccuracies, model coupling, parameters perturbation, speed fluctuation and external environmental disturbance. This paper accordingly proposes a nonlinear longitudinal-lateral coupling model predictive control method. Firstly, to effectively describe the inherent nonlinearity and coupling of the vehicle dynamic model, an approximate nonlinear longitudinal-lateral coupling model with a paradigm bounded uncertainty is established using linear fractional transformation formulation. Secondly, a final-value theorem-based constrained feedforward compensation strategy and a speed compensation are designed to obtain the longitudinal-lateral coupling control law, which can effectively handle heterogeneous disturbances of the system. Thirdly, a Sobol-based global sensitivity analysis method is designed to analyse and quantify the extent to which different disturbances affect the control performance of the system. Finally, the recursive feasibility of the closed-loop system is analysed and system asymptotic stability is proved through Lyapunov theorem. The simulation results of Carsim-Simulink and the statistical comparison analysis show the superiority of the proposed controller, which can effectively ensure the stability margin and has good tracking accuracy, robustness. Further, real-vehicle experimental results also show good accuracy and real-time performance.

The rapid development of deep learning has promoted the application of rolling bearing fault diagnosis techniques. However, in practical applications, the researchers often faces the challenge of a serious imbalance in the proportion of normal and fault states. This imbalance greatly affects the accuracy of diagnosis. Therefore, this paper proposes a novel fault diagnosis framework based on an auxiliary classifier generative adversarial network (ACGAN). Firstly, the stacked contractive autoencoder is cleverly combined with the discriminator network to improve its feature extraction capability and fault diagnosis accuracy. Subsequently, the original algorithm's focus on different types of samples is optimised to improve the generalisation of the diagnostic network. Finally, the stability of the generator network training is optimised with the help of the metric properties of the Kullback–Leibler scatter, and thus the variational stacked contractive-ACGAN model is proposed. The experimental results show that the fault diagnosis accuracy reaches 99.75$\%$ with 200 training samples of each class on the Case Western Reserve University (CWRU) bearing dataset, which is significantly better than other algorithms. Under the same conditions, on the Jiangnan University bearing dataset, the accuracy reaches 99.25$\%$, which shows good generalization and provides a broad prospect for future applications.

Intelligent diagnostic technology is helpful in the detection of welding quality issues which have a significant impact on product reliability, and attracts increasing concern. Welding diagnostic technology typically extracts feature values from raw signals and makes quality judgments based on these feature values. The conventional feature values extracted from electrical signals are not sensitive enough to indicate the welding quality and their amplitude varies with different welding powers, giving rise to difficulties in welding quality judgment. To address this problem, this paper proposes a normalized amplitude-invariant feature value (NAIFV), its extraction algorithm, as well as NAIFV-based threshold categorization (TC) and NAIFV-based support vector machine (SVM) for welding quality diagnosis. Experiments were conducted to validate the proposed approaches. It was found that NAIFV has advantages in rapidity, normalization and consistency compared to conventional feature values. Results also showed that the diagnosis accuracy of the NAIFV-based TC and NAIFV-based SVM reached 97.5% and 98.7% respectively, much higher than those based on spectral kurtosis, which is the best among the four conventional feature values discussed in this study.

This paper proposes an autonomous anti-disturbance control algorithm for vertical takeoff and landing unmanned aerial vehicle (VTOL-UAV) in satellite positioning rejection environment. In this kind of environment, the external satellite information is unavailable. This paper proposes a VTOL-UAV environment perception and precise positioning algorithm based on tightly coupled LiDAR and inertial navigation, on the basis of environment perception and precise positioning, the autonomous motion planning algorithm for VTOL-UAV is proposed. Motion planning is carried out according to the environmental information and the dynamic characteristics of the UAV, and the planning results include the spatial path information, the velocity information, the acceleration information and the JERK information of the UAV. Finally, the path tracking method of the VTOL-UAV with an extended state observer is proposed. Based on accurate estimation and feed-forward suppression of the external disturbances of the complex environment, the motion planning results can be accurately tracked, so as to realize the autonomous control of VTOL-UAV in complex environment of ultra-low altitude and improve the intelligence level of UAV.

Blockages in centrifugal pumps affect their flow rate and performance and can even interrupt their continuous operation. Health monitoring of the pump helps to avoid unwanted stoppages that can further lead to the failure of the whole system. Various supervised machine learning models have been developed in the past to monitor pumps for these faults. These models perform well on large amounts of labelled data, but a shortage of labelled data is a common problem in industrial applications. However, unlabelled data acquired from real-time operation of the pump are easily available but not utilised for training these models. Therefore, this study presents a semi-supervised methodology to detect blockages in pumps along with their severity. First, the support vector machine (SVM) model and the state-of-the-art long short-term memory (LSTM) model are individually trained with only labelled data using the statistical features acquired from the discharge pressure signal. The hyperparameters of both these models are optimised using the grid search optimisation method. Next, pseudo labels are generated for the unlabelled data through a trained SVM model. Pseudo labels defined by the SVM with a confidence greater than 0.9 are further selected to be combined with labelled data to train the LSTM model. The results show that the proposed approach effectively identifies blockage faults with a validation accuracy, test accuracy and F1 score of 97.51 %, 97.39% and 97.9%, respectively.

Path planning plays a crucial role in determining the shortest and safest route for autonomous mobile robots, and metaheuristic optimization algorithms have demonstrated their efficacy in solving complex problems involving numerous and variably shaped obstacles. The dung beetle optimizer (DBO) emerges as a novel heuristic optimization algorithm, drawing inspiration from the behavior patterns exhibited by dung beetles. This paper presents an improved algorithm called the enhanced DBO (EDBO) designed to address the mobile robot path planning problem. First, a node selection strategy based on the search radius is proposed to reduce the probability of path reciprocation while concurrently enhancing the initial population quality. Second, the conventional ball-rolling strategy employed in DBO is replaced with a dynamic step size search strategy. This strategy dynamically adjusts the step size based on the relative distance between the current position and the worst position of each iteration, enabling the population to update its position adaptively in response to environmental changes. Additionally, the global optimal position is incorporated to guide the search process, thereby improving the global search performance of DBO. Finally, a path fine-tuning strategy is implemented to refine the new generated individuals, with geometric principles of triangles being introduced to locally adjust the optimized path and guide the population out of local optima. Experiments are conducted to validate the effectiveness of the three improvement strategies, comparing the path optimization performance of EDBO with five other algorithms across five different complexity environments. Additionally, two statistical testing methods are employed to evaluate the experimental results. The experimental results indicate that, in all five environments, EDBO outperforms the other five algorithms in terms of path length and achieves superior path smoothness in 80% of the cases. Moreover, EDBO consistently demonstrates better overall performance than DBO, excelling in both convergence accuracy and speed.

Global navigation satellite systems (GNSSs) are often integrated with inertial measurement units (IMUs) for navigation in urban environments. The ability of these integrated systems to achieve precise positioning and navigation is highly dependent on accurate time synchronization. While the current time synchronization algorithms are capable of achieving millisecond-level synchronization in open environments, their performance deteriorates significantly in complex urban environments due to the impact on GNSS measurements of non-line-of-sight signals and multipath effects. Here we propose a loosely coupled GNSS/IMU integration time synchronization error modeling and compensation algorithm to tackle this problem in the context of urban vehicle navigation. In the algorithm, the time synchronization error is augmented as a state variable in the robust extended Kalman filter (REKF), with the robustness factor threshold being dynamically adjusted based on the time synchronization error. Furthermore, we have designed a system time synchronization detection scheme based on two detection factors, with the aim of achieving high-precision GNSS/IMU time synchronization in complex urban environments. Experimental results show that the proposed algorithm synchronizes data time to the millisecond level in urban positioning environments. This represents an improvement of more than 90% in time synchronization precision compared to the traditional extended Kalman filter (EKF)-based time synchronization algorithm, and an improvement of more than 60% compared to the REKF-based time synchronization algorithm. Furthermore, compared to the EKF-based GNSS/IMU fusion algorithm, the proposed algorithm enhances the accuracy of the determination of both position and velocity in the horizontal and 3D directions by more than 50% and heading accuracy by 85%.

Most visual simultaneous localization and mapping (vSLAM) methods assume a static scene, limiting their effectiveness in complex, real-world dynamic environments. This paper presents RED-SLAM-a real-time dynamic visual SLAM method based on the ORB-SLAM3 framework for RGB-D sensors, designed to effectively address the impact of dynamic objects. RED-SLAM leverages spatial-geometric observations combined with semantic cues to identify dynamic points within the field of view, thereby utilizing only static points for state estimation. In the geometric verification module, the initial distinction between static and dynamic points is achieved by checking the spatial projection ray distance error based on the matching between map points and feature points. To conserve computational resources, semantic segmentation is performed exclusively on designated semantic frames, which are constructed based on the changes of dynamic objects. The detected dynamic semantic objects are subsequently spread to successive frames using a semantic propagation technique. All map points associated with dynamic semantic objects are excluded to further enhance the identification accuracy of dynamic map point. Compared to existing methods that apply semantic segmentation across all frames or keyframes, RED-SLAM only performs segmentation when dynamic objects change, improving system's performance and efficiency. Experimental results on public datasets and real-world scenes demonstrate that our method enhances pose estimation accuracy in dynamic environments, achieving competitive accuracy compared to state-of-the-art methods, while maintaining reliable real-time performance.

This study proposes an enhanced multi-sensor fusion perception and localization algorithm named EMF-PAL, which leverages the strengths of three distinct sensor types to achieve high-precision localization. The algorithm employs a lightweight object detection network to extract semantic information from the environment, which is combined with feature points obtained through parallel bidirectional optical flow tracking. A dual-strategy approach is introduced to identify reliable static feature points, enabling accurate pose estimation. Sensor data is dynamically updated using factor graph optimization to ensure accurate and continuous real-time localization. In the feature preprocessing module, the lightweight object detection algorithm is integrated with Shi-Tomasi corner detection to provide prior information for subsequent object recognition and filtering. Additionally, the high-precision inertial measurement unit (IMU) supplies translation and rotation data, enabling short-term, high-accuracy positioning and motion compensation, effectively reducing misclassification during rapid movements. The Global Navigation Satellite System-Real-time Kinematic provides all-weather, drift-free global localization information without the need for additional transformations, further enhancing the factor graph optimization for position updates and supplying auxiliary data to the IMU. EMF-PAL maximizes the advantages of multiple sensors, making it well-suited for complex environments. Extensive comprehensive experimental validation using real outdoor environment sequences under different environmental conditions is conducted to demonstrate the real-time accuracy of EMF-PAL and its ability to cope with complex environments. The experimental results demonstrate that the proposed method enhances localization accuracy by up to 50.3% in challenging outdoor environments compared to state-of-the-art algorithms utilizing three-sensor fusion, effectively fulfilling the localization and perception requirements of real-world applications.

Accurate calibration of the inertial navigation system (INS) is essential for enhancing its performance. However, gravity disturbance significantly impacts the calibration accuracy of the INS, directly reducing the navigation performance of the high-precision INS. This paper investigates a method for further improving the calibration accuracy of the INS through compensating gravity disturbance. A full-parameter error model incorporating gravity disturbance for INS is established, the impact of gravity disturbance on INS calibration accuracy is analyzed, and gravity disturbance compensation method is proposed. Finally, the 19-position calibration experiment, the static-based navigation experiment and vehicle experiment are carried out to validate the effectiveness of the proposed method. Experiment results indicate that a 20'' meridian vertical deviation introduces a calibration error of 0.001° h⁻¹ in the X-axis and Y-axis gyro drift, a 10 mGal gravity anomaly results in a 10 ppm calibration error in the three-axis accelerometer scale factor, while the impact of the prime vertical deviation on the INS calibration accuracy can be ignored. After compensating gravity disturbance during the INS calibration process, the 60 min static navigation accuracy improved from RMS (root mean square) 332.81 m to 272.80 m and the 70 min dynamic navigation accuracy improved from RMS 1371.8 m to 1067.2 m.

The multimodal perception is crucial for mobile robots to achieve safe autonomous steering in complex and changing environments without colliding with obstacles. In spite of the fact that rich environmental information can be provided by the multimodal data, how to integrate the complementary features from different sources effectively remains an open issue. In this paper, a simple yet effective multimodal deep reinforcement learning (DRL) method, concise and effective multimodal DRL, by fusing two input modalities such as depth images and pseudo-LiDAR data generated by RGB-D cameras, is proposed to improve the performance of single sensor in environmental perception tasks. And the convolutional LSTM network layers are also adopted to extract spatiotemporal features and capture temporal relationships in consecutive time steps, which allow the robot to predict dynamic changes of the environment. In addition, A cross-modal fusion module is designed to optimize the multimodal data fusion scheme, with dynamic weighting and the merging of features from different modalities, allowing the agent to focus more reasonably on the current state and improving the accuracy and efficiency of the obstacle avoidance strategy. The experimental results show that the proposed approach achieved significant performance in cumulative rewards, convergence speed, and success rate. Additionally, tests in both simulated and real-world environments further verify that collisions are effectively avoided using only a low-cost RGB-D camera, demonstrating the method's strong generalization capabilities.

Proximity navigation of asteroid exploration missions requires high accurate point cloud registration for pose estimation of unstructured terrain. To enhance the registration accuracy, a neighbor surface variation variance (NSVV) based key points selection method is proposed to accurately extract key points and improve the pose estimation accuracy of the unstructured terrain. The proposed method extracts key points by top-percentage NSVV ranking in ascending order, which generates high-quality matching pairs with high data removal percentage and reduces the registration time-consuming. The matching pairs are further used for the coarse and fine registration to obtain the optimal pose estimation. The simulation experiment results show that the rotation error and translation error with 90% data removal percentage are 0.692° and 0.0513 m, which can be reduced by 59.44% and 45.48%, respectively. The proposed method offers an efficient, accurate and practical solution for unstructured terrain target pose estimation, paving the way for further advancements in proximity navigation of asteroid exploration missions.

The orbit quality of BDS-3 inclined geosynchronous orbit (IGSO) satellites is limited by the observation geometry as well as dynamic models. Regarding ground L-band data, the maximum nadir angle for IGSO satellites is approximately 9°. However, with BDS-3 Ka-band inter-satellite link (ISL) data, the maximum nadir angle can reach up to 45°, which significantly improves observation conditions. This study investigates the impact of ISL data on 5-parameter reduced extented CODE orbit model (ECOM1) compared with that from ground L-band data only. It is observed that the estimates of the ECOM1 parameter become more precise by using ISL data, particularly for D₀ as well as those in B direction. Y₀ bias shows noticeable thermal radiation effects for C38 and C39. The Sun-elevation angle dependent clock linear-fit residuals are reduced significantly for C39 or almost vanished for C38 and C40. By analyzing the solar radiation pressure (SRP) accelerations for the solutions with L-band only or L and ISL data, we observed that the radial acceleration differences show opposite dependency with these in the B direction. The similar phenomenon can be observed in the differences of solutions based on ECOM2 and ECOM1 with L-band data only. These reveal that the deficiency of the ECOM1 for the non-cubic satellites can be partly compensated by using better observation geometry from ISL data, as better estimates in B direction can be obtained. Based on the adjustable box-wing model, the optical coefficients of BDS-3 IGSO are calibrated with L-band and ISL measurements. By incorporating these estimates into the box-wing model used as the a priori SRP to augment ECOM1, the Sun-elevation angle dependent errors in clock offsets are mitigated. And the RMS of clock linear-fit is further reduced by approximate 1.2 cm by using the estimates from L-band and ISL data than that from L-band-only data. This demonstrates the positive contribution for ISL on SRP modeling of BDS-3 IGSO satellites.

With the rapid development of wireless communication technology, WiFi indoor positioning has become an important method for achieving indoor localization. Achieving high accuracy in WiFi positioning is a challenging issue. To enhance the accuracy of positioning systems, this paper proposes a WiFi indoor positioning algorithm that uses the random forest (RF) algorithm for access point (AP) selection and the Crested Porcupine Optimizer (CPO) algorithm to optimize support vector regression (SVR), referred to as RF-CPO-SVR. The RF algorithm selects APs by evaluating the feature importance of each AP, reducing the negative impact of redundant and unstable APs on the performance of the positioning system. After AP selection, the CPO algorithm is used to optimize the hyperparameters of SVR, further improving the system's performance. Comprehensive tests of the proposed RF-CPO-SVR algorithm were conducted on public datasets, and the results show that 90% of the positioning accuracy is within 4 m, with an average positioning error of 2.1082 m. Experimental results demonstrate that the RF-CPO-SVR algorithm outperforms traditional positioning methods and existing classical optimization algorithms, improving positioning accuracy by 23.5%, 27.4%, and 24.7% compared to particle swarm optimization-SVR, GA-SVR, and K nearest neighbors, respectively.

Integrated navigation systems, combining inertial navigation system (INS) and global navigation satellite system (GNSS), are extensively used in land vehicle navigation. However, they often exhibit poor performance during GNSS outages and lack robustness in challenging urban environments. In order to reduce reliance on GNSS and fully leverage the vehicle dynamics hidden in the data, this study proposes a novel robust navigation algorithm without GNSS, based on deep learning and variational Bayesian methods. Firstly, a cascaded long and short term memory (LSTM) model with an 'attitude-velocity' two-step sequential structure is proposed. By incorporating quaternions, it overcomes the non-differentiable property of attitude caused by its bounded nature, resulting in more accurate attitude predictions. Subsequently, the attitude prediction results are utilized to enhance speed prediction, leading to a significant improvement in accuracy. Building upon the cascaded LSTM model, variational Bayesian filtering is employed to further address the complex time-varying error of its results. Through an adaptive data fusion of LSTM predictions and INS data, the data fluctuations are reduced, and the accuracy is significantly improved. The efficacy of the proposed robust navigation algorithm is validated through multiple vehicle experiments conducted on diverse paths. Experimental results demonstrate that even in scenarios where GNSS fails continuously for 15 min, the proposed algorithm maintains of speed and horizontal attitude errors within 0.6 m s⁻¹ (root mean squared error (RMSE)) and 0.3° (RMSE) respectively. Consequently, the proposed algorithm has broad application prospects in GNSS-denied environments such as building occlusion, underground parking garages, and indoor positioning. In addition, we also provide attempts to improve the model for situations such as odometer assistance and different motion states.

Data processing of the Global Navigation Satellite System using uncombined (UC) observations has been a hot topic in recent years. However, the computation complexity and time consumption of the UC approach is much higher than those of the linear combination approach due to there being more parameters of interest. Concerning this issue, we propose a three-step approach by combining the ionosphere-free (IF) observation model and the UC observation model in a network scheme to reduce computation complexity. The first two steps are based on the IF observation model to detect cycle slips and gross errors using preprocessing and post-fit residual screening algorithms, respectively, while the third step is based on the UC observation model in which the ambiguities estimated from the IF model are used for UC ambiguity resolution (IF-IF-UC). To validate this approach, a network of 140 globally distributed stations is chosen and 1 year of GPS/GALILEO/BDS-3 dual-frequency data in 2021 is processed. The 1D RMSs of satellite orbit difference and the average STDs of satellite clock differences with respect to ESA's products for GPS, Galileo and BDS-3 are 1.8, 2.4 and 4.1 cm and 0.024, 0.019 and 0.028 ns, respectively. Compared with the GIM products provided by the Center for Orbit Determination in Europe, the mean RMSs of the VTEC differences for GPS, Galileo and BDS-3 are 2.38, 2.30 and 3.09 TECU. For comparison purposes, the approach where all three-steps are based on the UC model is also conducted (UC-UC-UC). Results show that satellite orbits, clocks and VTECs from the IF-IF-UC approach are almost the same as those from the UC-UC-UC approach. By introducing high-performance linear algebra libraries, the elapsed time of the proposed IF-IF-UC approach is about 1 h, which is an improvement of 36% when compared to the UC-UC-UC approach. Nevertheless, the elapsed time is nearly 6 d when the UC-UC-UC approach is conducted without introducing high-performance linear algebra libraries.

Maintaining the alignment of the shield tunneling axis with the designed tunnel axis is crucial in urban subway construction. However, previous studies have only verified the accuracy of the models in a specific project, ignoring model transferability. This paper proposes a multistep transferable prediction method (PCA-GRU) for shield attitude and employs feature importance analysis of the PCA-GRU model through feature dimensionality reduction experiments. Datasets obtained from the Guangzhou and Fuzhou Metro Line X Projects were used to validate the accuracy of the proposed method. The results revealed that PCA can unify the model input, enabling the model to be applied to different projects, and the proposed model could produce an accurate prediction within 24 steps, with an average R² of 0.94 and 0.97 in the Guangzhou and Fuzhou projects. This indicates that the proposed model possesses transferability and generalization ability, which can aid decision-making during urban subway construction.

LiDAR-based simultaneous localization and mapping (SLAM) methods often degrade in environments lacking geometric features (such as tunnels and long corridors), where planar features are abundant but linear features are sparse. To address this issue, we propose RID-LIO, an intensity-assisted LiDAR-inertial SLAM framework that integrates adaptive intensity feature extraction and intensity-based loop detection, overcoming the reliance on geometric features in existing methods. First, 3D point clouds are cylindrically projected to generate intensity images, from which intensity line features are extracted to enhance constraints in degraded directions. A weighting function is also incorporated to optimize the quality of the pose estimation, while an efficient intensity edge context descriptor improves loop detection efficiency and reduces trajectory drift. Evaluations on the VECtor dataset show an average improvement of 63.59% in trajectory accuracy. Tests on a private dataset demonstrate that RID-LIO outperforms other state-of-the-art methods in terms of end-to-end error and map consistency.

With the development of unmanned aerial vehicle (UAV) technology, UAV navigation based on deep reinforcement learning (DRL) has become a current research focus. In the existing research on UAV navigation based on LiDAR, most of them construct the state space of DRL using the direct measurement data of LiDAR, ignoring the impact of the representation method of LiDAR measurement data on navigation performance. To address this issue, this study analyzed the characteristics of different representation methods of LiDAR measurement data. Considering the impact of UAV angle changes, the LiDAR measurement data were represented by polar coordinates and Cartesian coordinates respectively to construct the state space of the UAV navigation model. Based on two classic DRL frameworks, through a large number of flight tests in complex static and dynamic scenarios, it was found that when considering the dynamic information of the environment, the models based on polar coordinates and Cartesian coordinates have better navigation performance, which provides new ideas for the effective utilization of LiDAR perception information.

Learning-based visual odometry (VO), that estimates frame-to-frame (F2F) translation and rotation in end-to-end network, has attracted much attention in recent years. However, existing methods perform poorly in complex environments, especially when there are moving objects in the scene. To solve the above issues, a novel monocular VO is proposed based on motion segmentation and attention mechanism under dynamic environments. Firstly, semantic information and optical flow residuals are adopted for motion segmentation through iterative optimization, so as to eliminate the influence of dynamic objects. Secondly, a parallel attention mechanism module is used for extracting the most relevant features for pose estimation. Two types of experiment (odometry dataset and tracking dataset) based on Karlsruhe Institute of technology and Toyota Technological Institute (KITTI) are carried out to show the effectiveness of the proposed method. Experimental results show that the accuracy of pose estimation is greatly improved by adding motion segmentation and attention mechanism. Moreover, the comparison results show that the proposed method performs comparably to geometry-based techniques and outperforms other learning-based approaches. This paper lays the foundation for accurate pose estimation in dynamic environments.

The measurement of the celestial Doppler difference velocimetry navigation is the Doppler difference between the solar spectrum and the planetary spectrum. The asynchronous solar/planetary spectral pairs in the existing database lead to the difficulty in the verification of the celestial Doppler difference velocimetry navigation. To solve this problem, we propose a VAE + CycleD2GAN-based generative method of synchronous solar/planetary spectral pairs, which fuse variational auto-encoder (VAE), cycle generative adversarial network (CycleGAN), dual discriminator generative adversarial nets (D2GANs). In CycleGAN, asynchronous solar spectra and planetary spectra are used as the training set to generate synchronous solar/planetary spectral pairs. Due to small samples in the existing spectral database, we explore the strong generalization ability of the latent space in VAE to improve CycleGAN and ensure its learning effectiveness. A double discriminator structure is introduced in VAE + CycleGAN to restrain the disadvantage of training instability and mode collapse, which can effectively capture spectral distribution patterns and improve the quality of generated spectra. Experimental results show that, compared with the asynchronous solar/planetary spectral pairs, the VAE + CycleD2GAN-based generative method produces synchronous solar/planetary spectral pairs, improving the velocimetry accuracy of the synchronous ones by more than 50%, and enhancing the position and velocity accuracies of navigation by more than 12% and 33%, respectively. The research demonstrates that the proposed method provides strong technical support for the feasibility of the celestial Doppler difference velocimetry navigation.

Addressing the challenges of position information solution among multiple unmanned aerial vehicles (UAVs) and the issue of low relative navigation accuracy, an improved collaborative localization algorithm based on the cubature Kalman filter (CKF) is proposed. It utilizes a tightly coupled approach to construct a relative ranging model equipped with ultra-wideband and inertial measurement unit sensors, enabling the correction of self-navigation information. The cosine similarity function is utilized to determine the reliability of the observations, and the observation noise matrix is adaptively adjusted based on the innovation vector from each iteration, thereby refining the positioning trajectory for collaborative navigation. Simulation experiments results demonstrate that the improved CKF collaborative localization algorithm significantly enhances the relative positioning accuracy between UAVs, effectively reduces the maximum relative positioning error, and exhibits strong anti-interference capabilities. It is more suitable for indoor multi-UAV and multi-sensor collaborative localization.

The strapdown inertial navigation system (SINS)/Doppler velocity log (DVL) integrated navigation is one of the primary methods for underwater navigation, providing autonomous underwater vehicles with precise information on position, velocity, and attitude. The core idea of the SINS/DVL integrated navigation using the Kalman filter is to establish the state and observation equations to estimate the SINS navigation error using DVL velocity. The optimal state estimation and correction strategies are critical to the accuracy of integrated navigation. To address this, this paper proposes a hybrid correction method for SINS/DVL Kalman filter integrated navigation based on a convergence factor judgment (HCKF-CFJ). First, the convergence factor for the Kalman filter is constructed using the diagonal elements of the estimated error covariance matrix corresponding to the velocity error state. Then, the filter's stability is assessed based on this convergence factor. When stability is confirmed, direct feedback correction is applied to the SINS velocity, and indirect feedback correction is applied to the SINS position. SINS velocity and position undergo feedforward correction if the stability condition is not met. Finally, ship experiments validate the effectiveness of the proposed method.

Tunable continuous wave (CW) lasers produce the highest optical resolution for pumping and probing atomic and molecular transitions. These lasers adopt sophisticated electronic control of optical components within the cavity to set the wavelength and ensure a narrow bandwidth. The low-cost circuit described here gates a CW laser beam on and off for high resolution timing experiments, using an electro-optic modulator that acts as a switchable waveplate between linear polarizers. The switch can operate at rates of up to 100 kHz with a minimum pulse width of 650 ns, or it can be configured to deliver laser pulses with widths $\unicode{x2A7E}$ 35 ns at lower rates. Multiple laser pulses of equal or non-equal widths can also be delivered within each repetition cycle.

In this Technical Note, we present a simple and low-cost production method for fluorescent particles perfectly suitable for velocimetry applications. We leverage the unique ability of Nile Red (NR) fluorophores to adsorb on the polymer surface and/or embed itself in between polymeric chains. The laboratory procedure to dye polyamide particles with NR and monetary advantages over commercially available fluorescent particles is outlined. Subsequently, the fluorescence behavior of the dyed particles is tested under a laser illumination source in polar and non-polar liquids. The distinct advantage of the emission spectrum of NR-dyed particles is demonstrated with sample test results.