Abstract:As a critical element of integrated circuits, semiconductor chips now demand increasingly higher quality standards. During the miniaturization and high-density manufacturing processes, chips are prone to defects that can impact their performance and longevity. Therefore, detecting and identifying these defects is crucial for enhancing chip reliability. This paper reviews the advancements in chip defect detection methods using machine vision over the past decade, both domestically and internationally. Initially, it introduces the chip manufacturing process and the prevailing chip packaging technologies. It then outlines the mainstream non-destructive testing technologies for chip defect imaging, which include optical imaging, acoustic imaging, infrared thermal imaging, electromagnetic imaging, and X-ray imaging. The paper further explains the methods for detecting surface defects in chips using both traditional technologies and deep learning. Additionally, it compares and analyzes defect detection methods for chip packages based on defect locations. Finally, the paper summarizes the current challenges in chip defect detection and explores potential future research directions

Abstract:Water level is a key element of hydrological measurement. Accurate water level measurement is of great significance for flood disaster prevention and water metering. With the construction of intelligent hydraulic engineering and the large-scale deployment of video equipment, the water level recognition methods based on image processing have been processed rapidly, which is currently cutting-edge research interest in the field of water level measurement. This article proposes a monocular vision-based gaugeless water level measurement method. Firstly, deep learning techniques are used to formulate a water surface segmentation model enabling automated waterline detection from water edge images. Subsequently, utilizing spatial mapping derived from camera calibration and sectional constraints, 3D coordinates corresponding to waterline pixels are computed. Finally, statistical methods are applied to compute the water level. The method is applied to an indoor flume experiment to validate its accuracy. The average number of falsely segmented pixels on the water line is 0. 825, which shows that the water surface segmentation is accurate. The mean absolute error and root mean square error are 1. 5 mm and 1. 9 mm, respectively. The results show that the method can accurately measure the variation process of water level.

Abstract:In the context of laser interference in electro-optical imaging reconnaissance equipment, interference spots often appear in the imagery. These laser jamming spots significantly degrade image quality and obscure target information, severely impacting detection and tracking systems′ performance. For addressing laser jamming images in typical target scenarios, an inpainting network is developed based on global semantic learning and salient object awareness. A gated semantic learning mechanism is specifically proposed. Initially, a contextual attention mechanism is employed to establish long-range correlations between the interfered and known regions, enabling the inference of content in the interfered regions. Then, a multi-scale feature aggregation module refines the inferred content across different receptive fields, reconstructing rich semantic information in the interfered areas. Finally, a gating mechanism adaptively fuses features from the known and reconstructed regions, enhancing the global semantic consistency of the restored image. Additionally, a salient target consistency loss is designed to guide the inpainting network in perceiving salient targets, improving the sharpness of object contours and texture coherence using a gradient penalty method based on the salient target mask. Experimental results in typical target scenarios such as aircraft, bridges, and roads demonstrate that the proposed network outperforms other methods in generating visually realistic and complete content, with good generalization performance in dealing with complex interference spots.

Abstract:Considering the advantage of visualization and non-invasiveness of electrical impedance tomography ( EIT), it′s broadly applied in industrial and biomedical fields. However, due to the highly nonlinear and ill-posed nature of inverse problem, numerical imaging methods face spatial resolution limitations. These limitations are especially evident in multiphase media distributions, where current EIT technology encounters boundary distortions and conductivity errors, thereby affecting the final imaging accuracy. To address mentioned issues, this paper introduces a learning-based model for EIT reconstruction, referred as MAT-UNet, which is mainly composed of U-shaped backbone and optimized multi-head attention block. The MAT-UNet integrates convolutional block attention module (CBAM) in the Encoders for feature extraction to construct the spatial and channel latent feature. In addition, the Squeeze-andExcitation Vision Transformer (SE-ViT) is introduced in the skip connection between Encoder and Decoder, which optimizes the global feature learning. Also, the Multi-Head Cross-Attention (MHCA) module facilitates multi-scale information fusion between the encoder and decoder. MAT-UNet is trained on extensive simulation data to obtain optimal model parameters and is experimentally validated on diverse complex shapes and lung simulation models. The quantitative evaluation metrics indicate that this method achieves a Root Mean Square Error (RMSE) of 2. 315 6 and a Structural Similarity Index (SSIM) of 0. 943 7 in reconstructed images. The visualized results closely match the true distribution and boundaries. Experimental outcomes demonstrate that the proposed MAT-UNet model exhibits robust performance and generalization capability. Compared to traditional single convolution structures, the integration of the Transformer structure provides more accurate EIT image reconstruction, presenting significant potential and value in non-destructive measurement and detection applications.

Abstract:In order to improve the detection ability of low, slow and small unmanned aerial vehicle(UAV) targets with very small pixels in dynamic, wide-angle scenes, this paper proposes a detection method that integrates inter frame information with template matching. Firstly, a dynamic information extraction module was designed to guide the algorithm to focus on dynamically changing small target areas by filtering out background information interference. Secondly, a multi template matching strategy is adopted to determine the similarity of the selected dynamic regions and complete drone target detection. Finally, drone target detection experiments were conducted under different backgrounds such as sky, mountains, and buildings, with different sizes and modes. The results show that the method proposed in this paper can effectively compensate for the shortcomings of deep learning methods in detecting extremely small pixel targets in wideangle views. The detection accuracy of low, slow and small targets reaches 0. 81, with a false alarm rate of 0. 06, and the accuracy can reach 0. 70 on datasets with pixel ratios not less than 0. 01% . The method is suitable for data processing in different modes such as visible light and infrared, and can meet the application needs of various intelligent algorithm combinations for detection in the future.

Abstract:Unstructured scenarios play a critical role in specialty industries, where ensuring the health of their internal surfaces is essential for the safe operation of the device. However, the large depth-of-field variations within these surfaces often lead to insufficient or mismatched depth-of-field in conventional camera imaging, resulting in out-of-focus blur and poor defect detection. Therefore, this paper proposes a fast construction method of focused tracking curve based on an extreme value search algorithm, and realizes fast highdefinition imaging. Firstly, the basic construction method of the focused tracking curve was established. Secondly, an extreme value search algorithm is proposed to simplify the curve construction process. Finally, the construction of single / multi-focus tracking curves based on A/ B calibration plates was designed, and its influence on imaging clarity was analyzed. The results show that,the proposed algorithm effectively reduces the number of image acquisition times and increases the focusing speed by 34. 8% while focusing accurately. Based on the multi-focus tracking curve, the average value of the objective evaluation index of the image is increased by 10. 35% . Clear resolution of 0. 5 mm cracks is achieved under 100~ 1 000 mm depth of field span. Therefore, this method can provide a solution for high-definition full depth imaging in unstructured scenes with conventional cameras.

Abstract:Airhole is the most common apparent quality problem in reinforced concrete structures. Limited by the complex site environment and the computing power of equipment, existing evaluation methods have challenges such as poor accuracy and slow operation efficiency. In this paper, a quantitative assessment method of airhole defects based on hierarchical fusion architecture of point clouds is proposed, to realize efficient quantitative assessment of the apparent quality of concrete from end to end. Firstly, the set, shape, and depth feature information of the point cloud of the target scene were extracted, and a new hierarchical fusion architecture of multi-dimensional information of point cloud was given. Secondly, a planar linear search method based on depth line distortion is proposed to effectively overcome the influence of environment and noise in defect detection. Then, in order to reduce the influence of shooting angle and other interference information, the maximum heavy plane defect volume quantification model was established. In addition, to address the issue of key defect point loss during oblique scanning, a compensation strategy is proposed to improve the evaluation accuracy of different shooting angles. Finally, the accuracy and robustness of the proposed method are verified by the comprehensive evaluation index of defects and field experiments. The results show that the proposed method has a good evaluation effect on stomatal defects under various conditions, with a front scan error of less than 6. 0% and a compensation error for oblique shooting of less than 19. 8% . This method can provide an effective reference for on-site construction quality assessment.

Abstract:Contact wire wear (CWW) is an important and comprehensive indictor to represent the catenary servicing performance. Due to large spatial range, high accuracy and high efficiency requirements of CWW measurement, a visual method with active structured-light is proposed for the application of CWW onboard monitoring based on the triangulation and Scheimpflug principles. To solve the vision modeling and calibration problems, the Gaussian-Newton nonlinear least square method is investigated. The parameters of homograph matric between image plane and structured-light plane, the coefficients of lens distortions are extracted simultaneously by the crossinteractive numerical optimization approach. The camera calibration for the CWW measurement is established with large field of view and high accuracy, and the corresponding onboard monitoring equipment is developed to meet the field demand. The experimental studies of static calibration and field dynamic test are both carried out. It′s found that the calibration re-projection root-mean-square (RMS) errors of the proposed imaging model is 0. 083 mm, which improves the accuracy of 53. 1% compared to the traditional imaging model. Meanwhile by the comparing the onboard test and manual results, the RMS errors of CWW width, depth and area are within 0. 119 mm, 0. 115 mm and 0. 788 mm 2 , respectively.

Abstract:Traditional SLAM technologies based on explicit scene representations, such as point clouds, have matured in accuracy and robustness but fall short in capturing the texture and semantic information of the map. To address this limitation, this paper introduces neural radiance fields (NeRF) with differentiable rendering capabilities into the traditional visual SLAM system, proposing a novel visual SLAM method: DRM-SLAM (dense radiance mapper-SLAM). This method uses ORB-SLAM3 for camera pose estimation and combines the RGB and depth information of keyframes to generate dense point clouds. By utilizing a dynamic voxel grid, the method samples within the grid according to the three-dimensional geometric information provided by the point cloud data, thereby reducing the frequency of NeRF calling the multilayer perceptron (MLP). Additionally, the method incorporates multi-resolution hash coding and the CUDA framework′s NeRF implementation, significantly accelerating NeRF training speed. Tests on the TUM, WHU-RSVI, Replica, and STAR datasets demonstrate that DRM-SLAM effectively uses dense point clouds and NeRF volume rendering technology to fill gaps in point clouds, maintaining the pose estimation accuracy of traditional SLAM methods while enhancing texture and material continuity in the map. The DRM-SLAM algorithm achieves a frame rate of 22. 3 on the Replica dataset, which is significantly higher than NICE-SLAM, iMap, and Co SLAM algorithms, showcasing its high real-time performance. Ablation experiments in the same scenario show that NeRF rendering based on dense point clouds increases the frame rate threefold compared to traditional NeRF methods, further proving that dense point clouds can accelerate NeRF convergence and demonstrating the effectiveness of DRM-SLAM in map reconstruction.

Abstract:Simultaneous Localization and Mapping (SLAM) is widely used in the field of mobile robots since it can solve the problem of localization and mapping in unknown environments. This paper proposes a SLAM method named 3D-MultiFus, utilizing radar, camera, IMU, and wheel odometry. The radar-IMU-odometry subsystem ( Ls) rapidly constructs the geometric structure of global map by minimizing point-to-plane errors to estimate the system′s positional state. Meanwhile, the camera-IMU-wheel odometry subsystem (Vs) removes occluded or depth-discontinuous feature points to minimize inter-frame map photometric errors, which further estimates the pose state and enables color rendering of the point cloud map within the sub-map. The tightly coupled data resulting from the fusion of IMU and odometry, radar system′s point-to-plane errors, and camera system′s photometric errors are processed using an error-state-based iterative Kalman filter (ESIKF), ensuring precision and robustness while simultaneously achieving rapid localization and mapping. To validate the localization and mapping accuracy of proposed algorithm, the 3D-MultiFus algorithm was compared with related algorithms based on an established indoor motion experimental scenario. Simulation and experimental results demonstrate that the 3D-MultiFus algorithm completes data processing in 185 ms, outperforming the operational efficiency of other algorithms. The long-term positional error between the initial and final positions is merely 0. 085 6 m in complex indoor scenarios, significantly enhancing the global map accuracy of the 3D-MultiFus mobile robot. The constructed global map exhibits excellent consistency, validating the robust and reliable performance of the proposed algorithm in indoor environments.

Abstract:The high cost and large volume of three-axis high-precision optical gyroscope have hindered the widespread application of the gyroscope-based north-finding technology in the field of mobile robotics. In response to this issue, a method for indoor mobile robot north-finding based on a single-axis optical gyroscope and Micro-Electro-Mechanical System ( MEMS) Inertial Measurement Unit is proposed. This method achieves or approaches the heading accuracy of three-axis optical gyroscope while reducing costs and volume by approximately two-thirds. When the mobile robot is stationary at its initial position, two potential heading angles are obtained by combining the output of the single-axis optical gyroscope with the output of a MEMS three-axis accelerometer. Similarly, two potential heading angles at the second position are obtained in the same manner. By subtracting the heading angles at the two positions, four potential heading change values are derived. A unique heading angle for the second position is determined by comparing these values with the heading change value obtained from the MEMS strapdown inertial navigation system. Error analysis shows that the proposed method achieves a heading accuracy close to three-axis optical gyroscopes, especially when the heading angles are near 90° and 270°. Experimental results from on-vehicle tests confirm the effectiveness of the proposed method. Keywords:gyroscope-based north-finding; inertial navigation; single-axis optical gyroscope; M

Abstract:To address the issue of interdimensional coupling interference in three-dimensional vibration acceleration sensors, this paper focuses on the spatial layered structure fiber bragg grating (FBG) three-dimensional vibration acceleration sensor. It outlines the basic principle of three-dimensional vibration acceleration sensing. An experimental platform for dynamic calibration of the vibration acceleration is constructed, and the structural coupling characteristics of the sensor are analyzed. A neural network model based on the whale optimization algorithm and extreme learning machine (WOA-ELM) is proposed for non-linear decoupling experiments. The results show that the average measurement errors in the x, y, and z axes are reduced to 1. 58% , 1. 17% , and 0. 17% , respectively. Additionally, the maximum values of the average class I and class II errors are reduced to 0. 73% and 0. 37% , respectively. The decoupling effect of the WOA-ELM is compared with other algorithms, and the results demonstrate that WOA-ELM is more effective in reducing inter-dimensional coupling interference in the three-dimensional vibration accelerometer sensor, thereby improving measurement accuracy.

Abstract:Due to the high flexibility and environmental adaptability of remote-operational manipulators, they are increasingly used to complete tasks in special situations. An FBG sensor-based glove for the remote operation of manipulator is proposed to meet the demand for high-precision remote operation of manipulator in hazardous environments. A bending measurement model of fiber Bragg grating sensors in gloves is established. The central wavelength variation of fiber Bragg grating sensor is found to be directly proportional to the bending angle. Combined with the control of the manipulator servo, a method for remote operation of manipulator using fiber Bragg grating sensors is proposed. The upper computer software is developed. The process from the center wavelength of the fiber Bragg grating to the bending angle of the finger, and finally to the controlled encoding of the manipulator servo is implemented. The design and packaging of the fiber Bragg grating sensors are completed. There are a total of 12 FBG sensors with a length of 10 mm in the four detection fibers of the gloves. Among them, 11 FBG sensors are deployed above the finger joints for bending sensing, and 1 FBG temperature sensor is used for temperature compensation. The FBG sensors are packaged with 704 silicone rubber. A remote operation system for the manipulator based on fiber Bragg grating sensors is built, and the fiber Bragg grating sensors are calibrated. The experimental results showed that the average bending angle error was 1. 147° within the range of 0° ~ 90° bending of the human hand joint. The determination coefficients of all fitted lines are greater than 0. 993. Finally, a human-computer interaction demonstration module is designed to display the movement of the human hand. The glove based on fiber Bragg grating sensors proposed in this article has the characteristics of miniaturization, high precision, and resistance to electromagnetic interference. It has potential applications in spacecraft in orbit maintenance, remote medical surgery, high-risk mine clearance operations, and so on.

Abstract:Advanced photonic chip testing systems have attracted significant attention to date. However, existing photonic chip testing systems are primarily developed for communication chips and lack environmental parameter control functions, making them insufficient for the development needs of sensing chips. In this study, we present a photonic chip sensing test system based on grating coupling techniques. The system consists of automatic grating coupling, environment control and perception, as well as data processing and interaction. Using this system, we demonstrated gas concentration sensing and temperature sensing by using a silicon micro-ring resonator. Experimental results show that under standard temperature and pressure conditions, the system can detect CO2 gas concentrations ranging from 20% to 80% , with a sensitivity of 0. 152 GHz/ % (2. 113 pm/ % ). Moreover, the temperature sensing can be achieved in a range of 30℃ to 35℃ with a sensitivity of 4. 996 GHz/ ℃ (74. 891 pm/ ℃ ). This work provides a rapid and efficient approach for the development of optical sensor chips.

Abstract:In this paper, a stochastic networking algorithm based on global state extended kalman-based particle filter on labeled multiBernoulli (GS-EPF-LMB) is proposed for distributed cooperative navigation of multiple robots in intermittent observation or no absolute observation environments. The algorithm models the states and observations using random finite sets and generates labeled multi-Bernoulli particles through three state update strategies: time update, observation update, and display communication. To improve the consistency and localization accuracy of the algorithm, this paper couples relative and absolute observations based on labeled multi - Bernoulli particles, using particle filters to optimize the labeld particle states and constrain state estimation with historical information. In addition, it employs probabilistic data correlation for navigation system state estimation and uses a hierarchical Gaussian model combined with variational Bayesian methods to achieve globally optimal state estimation. The experimental results show that the proposed algorithm achieves a localization accuracy of 0. 11 m. The convergence of localization state covariance is improved by 48. 6% and the accuracy is increased by 11% compared with the GS-CI algorithm.

Abstract:Radiation therapy stands as a crucial method in treating lung cancer. However, the spatial movement of lung tumor by the respiratory movement presents challenge to precision radiotherapy. Aiming at the problem of extra radiation doses to healthy tissues caused by image guidance in current precision radiation therapy, this paper proposes an idea to simulate the respiratory deformation of the lungs in vitro by using a lung-like soft actuator. By modeling and predicting the motion of lung tumors, this approach aims to reduce the additional radiation dose to patients, thereby achieving accurate and safe radiation therapy. Firstly, the CT imaging is used to reconstruct the lung′s shape accurately, capturing the deformation motion pattern of the lungs during respiration. On this basis, the morphology of the lung cavity is designed and the structural design′s feasibility is verified through deformation simulation. A prototype of the lungmimicking soft actuator is then developed. Finally, the prototype is tested by giving discrete periodic air pressure to simulate the actual respiratory process. The experimental results show that the designed lung-like soft actuator can simulate respiration deformation as expected, with the maximum air pressure of 45 kPa, the maximum axial displacement of 5. 9 mm, and the maximum radial displacement of 3. 4 mm. The soft actuator exhibits good deformation mimics performance, and can be used as a reference of accurate and safe radiation therapy for lung tumors.

Abstract:Fixed orthodontic technique is the most effective method for the treatment of canine crossbite. The purpose of space reservation and tooth retraction in crossbite treatment is mainly achieved through the orthodontic force released by the orthodontic archwire. Among the tooth crossbite, the canine crossbite is the most common deformities of teeth. The combination of herringbone and T-shaped loop is often used to close the maxillary dental space. This article aims to quantify the release of orthodontic force by herringbone and T-shaped combination before orthodontic treatment. It can provide a theoretical reference for orthodontists in the treatment of crossbite. The deformation process of herringbone and T-shaped composite curve is analyzed, and the deformation process is divided into elastic deformation and elastoplastic deformation. The differential equations of the orthodontic force under the two kinds of deformation are established, and the released orthodontic force is divided into two kinds, including retraction force and overturning force. The theoretical model of herringbone and T-shaped combined loop is established. Secondly, the error rate of the finite element simulation data and the theoretical value are calculated, and the error rate between the simulated and theoretical values of the retraction correction force is 0. 43% ~ 8. 33% , and the error rate between the simulated and theoretical values of the overturning correction force is 2. 13% ~ 9. 76% . The experimental measurement platform is established. The error rate between the experimental and theoretical values of the retraction force is 0% ~ 9. 30% , and the error rate between the experimental and theoretical values of the overturning force is 0% ~ 9. 76% . The experimental measured data showed the same nonlinear variation trend with the theoretical values, which verified the correctness of the theoretical prediction model of the herringbone and T-shaped combination loop. This model can help physicians to conduct a detailed quantitative analysis of orthodontic force before treatment to provide a basis for the formulation of personalized treatment plans and better meet the needs of patients.

Abstract:To meet the requirement of clear texture of coronary angiography images in interventional surgery, this article proposes a super-resolution image reconstruction method based on the omnidirectional deep weighted and lightweight network. Firstly, the local convolution module is designed to reduce the dimension of the feature map to reduce its parameter quantity and speed up the processing speed of the model. Then, the self-attention mechanism module is used to fuse the channel and spatial information of the image to obtain the rich high-frequency detail features of the image. In addition, to further extract the deep feature information of the image, a cascade and weight matching layer attention structure is designed to assign different weights to the features of different depths of the image to realize the super-resolution reconstruction of the image. Finally, to make the method have a stronger generalization ability in real interventional coronary angiography images, a coronary angiography image dataset (CAID) is constructed for training and testing the network model. The experimental results show that, compared with the Omni-SR algorithm, the proposed algorithm reduces the number of parameters by 32. 3% and the running time by 17. 74% . Meanwhile, the quality of the reconstructed image is better than other comparison algorithms in terms of objective indicators and subjective feelings. The average values of PSNR and SSIM are increased by 0. 72 dB and 0. 0122 on the CAID dataset, and 0. 13 dB and 0. 004 4 on the public dataset, respectively.

Abstract:The early detection and precise pathological classification of gastric cancer can effectively improve the possibility of cure, posing higher demands on limited medical resources. In response to the various sources of classification for gastric cancer and the shortage of pathologists, this paper, for the first time, constructs a gastric tumor pathological classification recognition system using the ResNet-50-based DeepLab v3 semantic segmentation algorithm. This system assists pathologists in achieving rapid, efficient, and accurate collaborative diagnostic classification. For cases without malignant tumors, this paper also implements the binary classification recognition of low-grade intraepithelial neoplasia in the stomach. After training and testing on 1854 digitally annotated slices of gastric tissue from the Chinese PLA General Hospital, pixel-level annotated by experienced physicians, the system achieved a classification diagnosis accuracy of 61. 8% with a kappa value of 0. 496 for cancer zone identification. For low-grade intraepithelial neoplasia diagnosis, it attained a sensitivity of 100% , a specificity of 75. 8% , and an AUC of 0. 972. This paper presents the first implementation of gastric cancer classification diagnosis, capable of identifying cancerous areas and providing diagnostic references. Additionally, the system demonstrates high sensitivity and relatively accurate results for diagnosing low-grade intraepithelial neoplasia.

Abstract:To address the issue of low accuracy in quantitative analysis of coal quality testing using laser-induced breakdown spectroscopy (LIBS) due to matrix effects or environmental factors, this article proposes a method for fast quantitative analysis of organic elements in coal based on LIBS technology combined with the ASG-LWNet algorithm. Firstly, the LIBS spectra of 34 coal samples are collected using a laser-induced breakdown spectroscopy instrument. Then, an adaptive SG filtering algorithm is used to denoise the collected spectra, continuously updating the filter parameters to adapt to different signal characteristics and achieve better filtering effects. Finally, the corresponding characteristic spectral lines of the elements C, H, and S are selected as inputs to the LWNet model for quantitative analysis. Experimental results show that the correlation coefficients of C, H, and S elements based on the ASG-LWNet model on the test set are 0. 998 4, 0. 973 2, and 0. 995 4, respectively. And the root mean square errors are 0. 379 4, 0. 217 9, and 0. 611, respectively. Compared with before denoising, the prediction accuracy is significantly improved. The results indicate that, in the case of complex spectral noise, this method can reduce the impact of matrix effects and improve the accuracy of quantitative analysis.

Abstract:Aiming at the problem of accurate denoising of magnetic field signals under strong noise interference, a new method of denoising magnetic field signals combining the auxiliary sensor array of center-satellite architecture and the deep noise reconstruction network is proposed. First, the magnetic field sensor array is built, and finite element analysis is used to optimize the sensor array positions and analyze the signal characteristics between the center and satellite sensors. Subsequently, a deep neural network model combining noise-enhanced convolutional neural network (NECNN) and bi-directional long short-term memory (BiLSTM) is constructed. The model is trained using the noise signals captured by the sensor array to reveal the nonlinear mapping relationship between the center sensor signal and the satellite sensor signal. Finally, in the magnetic field detection process, the noise components of the center sensor are reconstructed using the noise of the satellite sensor array. The denoised magnetic field signal is obtained by subtracting the reconstructed noise from the noisy signal captured by the center sensor. The experimental results show that the proposed method outperforms the conventional method in terms of the maximum error and the root mean square error index of magnetic field denoising. This new approach provides a new means of dynamic denoising of signals under strong magnetic field interference, and is expected to be applied in the fields of current detection, magnetic field imaging, and battery quality detection.

Abstract:In complex working conditions, the monitoring signals of mechanical equipment are easily disturbed by multi-vibration sources and background noise, making weak fault features and strong coupling. It brings a great challenge to fault diagnosis. Therefore, a generalized minimax-concave enhanced convolutional sparse mechanical fault features resolution method based on the vibration characteristics atom is proposed to analyze weak features and strong-coupling faults. Firstly, an auto-adapted single-side fading wavelet framework is constructed to obtain the optimal feature atoms. The optimal feature atoms are increased in dimension to match the fault periodic to get the vibration feature atoms with periodic characteristics. Secondly, a convolutional sparse coding method based on GMC enhancement is proposed, which combines vibration feature atoms to obtain the sparse coefficients optimally. In addition, a processing parameter optimal selection method based on the ratio of average kurtosis to harmonic energy is designed, which overcomes the dilemma of selecting key parameters. Finally, the main features of the envelope spectrum are extracted and compared with theoretical fault feature frequencies to determine fault type. The effectiveness and superiority of the proposed method are verified by simulated and real test-bed signals. The spectrum kurtosis and tunable Q-factor wavelet transform Generalized Minimax-concave sparse enhancement method are set as comparison groups. The results demonstrate that different fault features are better decoupled, and the sparse component amplitudes are well improved compared to the comparison method.

Abstract:Transfer learning techniques can reduce the distribution difference between source and target domain features. However, in cross-device scenarios, existing research is often difficult to measure and reduce the differences in the conditions of data between different devices, resulting in the knowledge obtained by the model in the source domain cannot be migrated to the target domain. Additionally, in real-world failure diagnostic scenarios, decision-makers usually need to understand why the model classifies a specific type of fault. Due to the complexity of deep learning models, they are often seen as " black boxes," making it difficult to explain their internal workings. To address these issues, an interpretable fault diagnosis method based on conditional metric transfer learning is proposed. Firstly, Hilbert envelope spectrum analysis is used to convert time-domain signals into frequency-domain signals. Secondly, a deep twin convolutional neural network and classifier are built to extract high-dimensional features from both source and target domain data in the frequency domain and perform classification training. Then, the interpretable Conditional Kernel Bures is embedded into the loss function of unsupervised learning to enhance feature adaptation and model interpretability under conditional distribution. Finally, the SHAP method from game theory is used to conduct post-event interpretable analysis of the model diagnosis results based on the envelope spectrum. Tests were conducted on 12 cross-equipment bearing fault diagnosis tasks across three types of mechanical equipment, evaluating the proposed method against other related methods. The results show that the proposed method could effectively improve the accuracy of cross-equipment mechanical fault diagnosis, achieving an average diagnostic accuracy of 99. 47% . It also identifies which frequency points played a crucial role in the model′s decision-making process.

Abstract:A stream algorithm is proposed to address the low search efficiency of traditional algorithms in mobile robot path planning. Firstly, the algorithm obtains all the main points through the main point search model. Flow step by step from the starting point, when a single obstacle is encountered, the stream obstacle avoidance algorithm is invoked to avoid the obstacle. When multiple obstacles are encountered, the pseudo-virus algorithm is called to mark these obstacles. Then, the stream obstacle avoidance algorithm is called to avoid obstacles until the end. Finally, the resulting path is smooth. A variety of map environments are modeled by using the grid method. The path length and running time of the stream algorithm are compared with those of the ant colony algorithm, Dijkstra algorithm, and Floyd algorithm in simulation studies. The testing results show that, compared with the A ∗ algorithm which achieves the shortest path and the least time, the average path length obtained by the stream algorithm is reduced by 2. 40% ~ 6. 30% , and the average time is reduced by 35. 71% ~ 53. 51% . To test the application of the stream algorithm in the actual scene, it is applied to the mobile robot Turtlebot2 and conducted a comparative experiment with the A ∗ algorithm. The experimental results show that, compared with the A ∗ algorithm, the measured path is increased by 3. 83% , the running time is reduced by 10. 77% , and the number of inflection points is reduced by 42. 86% .

Abstract:To address the issues with current variations of Convolution Recurrent Networks (CRN), which often extract limited features, capture global characteristics poorly, and have large parameter sizes under single masking or mapping encoder-decoder structures, this paper proposes an efficient single-channel speech enhancement network. This network combines a multi-feature aggregation convolution module, leveraging complex spectrum joint masking and mapping, with an efficient Transformer-based attention mechanism. In the encoder-decoder layer, a Dual-branch Gated Cooperative Unit (DGCU) is designed to interact and aggregate multi-level complex spectral features, addressing the problem of singular feature extraction. The intermediate layer incorporates a Channel-Time-Frequency Attention Fusion Module, focusing on spatial and time-frequency local detail features of speech. Ablation and comparative experiments on the THCHS30 dataset demonstrate that this network achieves lightweight efficiency with the lowest parameter count and relatively low computational cost. It improves PESQ by 10. 5% ~ 50. 6% and 16. 3% ~ 94. 5% under matched and mismatched noise conditions, respectively. Both objective and subjective metrics outperform other comparative network models, exhibiting superior noise reduction performance and network generalization capability.

Abstract:In industrial automated assembling, the phenomenon of mechanical jamming often leads to an increase in contact force between parts, resulting in assembly failures. This article focuses on the assembly task of the door pin and body orifice in automotive assembling, using the admittance control method to address the issue of compliance. Influenced by admittance parameters ( elasticity, damping, and mass), the control system may exhibit varying degrees of fluctuations in steady-state error, overshoot, and response speed. To eliminate the impact of steady-state error, this article successfully addresses it using an integral optimization method based on gain scheduling. In response to the issues of overshoot and imbalanced response speed, a penalty function-based automatic tuning algorithm is proposed. By setting penalty terms, the constraints are gradually satisfied during the solution process. Experimental results show a reduction in oscillatory fluctuations of contact force from 0. 1 N to 0. 05 N, and a 50% reduction in settling time when external impedance disappears. This method successfully avoids part damage caused by oscillations and enhances compliance performance during automated assembling.

Abstract:Aiming at the low power consumption, bluetooth low energy (BLE) connection delay, and narrow band internet of things (NB-IoT) downlink traffic response delay requirements for NB-IoT and BLE dual-mode single-standby terminals, a terminal communication control method based on semi-Markov chain is proposed to enhance the comprehensive performance of the terminal. By integrating the extended idle mode discontinuous reception (eDRX) and power saving mode (PSM) of NB-IoT, where radio frequency is exclusively dedicated to BLE in PSM, a semi-Markov chain model for terminal communication state transition is constructed. This model is used to derive the power saving rate and the calculation model for the average NB-IoT downlink delay. Based on the delay requirements of BLE connection and NB-IoT downlink traffic, and combined with NB-IoT real-time traffic data, the critical parameters of eDRX are dynamically calculated. Subsequently, the terminal optimal comprehensive performance is achieved through optimizing BLE broadcast duration. Experimental results show that the proposed method achieves a power saving ratio exceeding 0. 96 in low-power demand scenarios. Furthermore, in low-delay demand scenarios, the average downlink traffic response delay is less than 10 seconds. This method satisfies the communication delay requirement and reduces the communication power consumption of the terminal.

Abstract:The precision of trajectory tracking in autonomous vehicles is closely linked to the performance of onboard sensors. However, various interferences can cause sensor data loss, impacting the vehicle′s movement. This article focuses on differential drive unmanned vehicles, proposing a trajectory tracking control method that relies solely on wheel odometers. By expanding the state observer to estimate total disturbances, it addresses the issues of reading deviations caused by interference in complex conditions and cumulative errors from long-term odometer operation. Firstly, the paper analyzes the odometer′s positioning process and uses an extended state observer to accurately measure disturbances affecting positioning. Compensation measures are then applied to reduce odometer bias and improve positioning accuracy. Following this, the dynamic tracking errors of the vehicle are thoroughly investigated, leading to the formulation of error equations for a trajectory tracking control strategy. In actual road testing, the vehicle is stabilized within a tracking error range of 0. 21 m, verifying the feasibility and effectiveness of the proposed method.

Abstract:Pneumatic soft actuators are fundamental and critical components of pneumatic soft robot, offering unparalleled advantages in achieving bending motions. However, the nonlinearity of both structure and material property presents great challenge for modeling and control of soft actuators. Based on the assumption of piecewise constant curvature deformation, Yeoh hyperelastic material constitutive model, and the virtual work principle, a bending model of a pneumatic networks ( pneu-net) soft actuator with constant curvature is established in this paper. Finite element simulations were conducted to investigate the effects of structural parameters and input air pressure on the bending performance of the actuator. On this basis, a variable parameter variable curvature actuator design was proposed, and a predictive model for its bending deformation was established. Consistency error analysis for multiple actuators was performed, and the model′ s validity was verified through finite element simulations and experiments. Finally, the force output characteristics of the soft actuators were experimentally tested. A three-finger soft gripper was developed, and the performance of gripping different objects was shown by experiments.

Abstract:The accuracy of downhole drilling attitude relies on the error calibration of the inertial measurement unit ( IMU). However, ellipsoid and plane fitting based on simplified linear models have failed to meet engineering requirements. The periodic residual error of axial tilt angle and tool face angle needs further compensation, and its nonlinearity is confirmed by multi-point tumble tests and Monte Carlo simulations. In this study, we employ post-fit identification through tilt tumble tests and perform parameter optimization, followed by proposing an equidistant balance correction technique. Data is collected under typical cases where ellipsoid fitting, plane fitting, tumble optimization, and equidistant balance correction techniques are applied in experiments. The results indicate that tumble optimization and the equidistant balance correction technique effectively reduce the nonlinearity of attitude error, reducing the nearvertical angle error of axial tilt from an initial ±0. 1° to calibrated ±0. 000 8° (1σ), and the tool face angle error from intial ±1° to ±0. 016 8° (1σ).