Most downloaded articles
2021(11):150-159.
Abstract:
Aiming at the high missed detection rate of casting defects, a casting defect detection method based on deep learning model fusion is proposed. Firstly, the Faster RCNN network is improved, the feature pyramid structure is used to improve the feature extraction network module, multi-scale feature fusion is realized, and the feature extraction of casting defects is completed. Then, the ROI pooling layer in the network is improved based on ROI Align, and the IOU score is introduced into the judgment process of NMS algorithm. And the improved network is integrated with Cascade RCNN and YOLOv3. Finally, an experiment study was carried out to verify that the fusion model can effectively reduce the missed detection rate of casting defects. The experiment results show that the defect recall rates in the Faster RCNN model and the network model proposed in this paper are increased by 1. 73% and 4. 08% , respectively after the pooling improvement of the region of interest. Using the method of model fusion, in the condition without considering the classification accuracy, the defect recognition rate of the entire model reaches 95. 71% . Compared with single model, while guaranteeing the detection accuracy of casting defects, the method also improves the defect detection recall rate and meets the requirements of industrial applications.
Aiming at the high missed detection rate of casting defects, a casting defect detection method based on deep learning model fusion is proposed. Firstly, the Faster RCNN network is improved, the feature pyramid structure is used to improve the feature extraction network module, multi-scale feature fusion is realized, and the feature extraction of casting defects is completed. Then, the ROI pooling layer in the network is improved based on ROI Align, and the IOU score is introduced into the judgment process of NMS algorithm. And the improved network is integrated with Cascade RCNN and YOLOv3. Finally, an experiment study was carried out to verify that the fusion model can effectively reduce the missed detection rate of casting defects. The experiment results show that the defect recall rates in the Faster RCNN model and the network model proposed in this paper are increased by 1. 73% and 4. 08% , respectively after the pooling improvement of the region of interest. Using the method of model fusion, in the condition without considering the classification accuracy, the defect recognition rate of the entire model reaches 95. 71% . Compared with single model, while guaranteeing the detection accuracy of casting defects, the method also improves the defect detection recall rate and meets the requirements of industrial applications.
2023, 44(5):61-70.
Abstract:
The existing data-driven methods in the early detection of rolling bearing degradation have problems of low sensitivity and high false alarms. To address these issues, a dynamically adjustment grey incidence analysis ( DAGIA) method for transient mechanical equipment health monitoring is proposed. First, the Hilbert transform is applied to demodulate the vibration data of the rolling bearing to obtain the envelope signal. To weaken the influence of the value of the resolution coefficient to highlight the degree of discrimination of the correlation value, the feature-to-noise energy ratio (FNER) method is introduced into the traditional grey incidence analysis (TGIA) to dynamically adjust the resolution coefficient, which can characterize the strength of bearing faults. Then, the first set of data is extracted at the initial stage of bearing operation as reference data. The dynamic grey incidence analysis is calculated between the remaining data and the reference data and the bearing performance degradation index is established. Finally, according to the normal samples and combined with Chebyshev′s inequality, the control line is set to identify the starting position of the early degradation of the rolling bearing. The IMS and XJTU-SY databases are used to complete the early degradation recognition of rolling bearings. The results show that the proposed method can accurately recognize the starting position of early degradation and the false alarm is close to 0. It has both sensitivity and robustness, which is beneficial for equipment maintenance personnel to better grasp the operating status of rolling bearings.
The existing data-driven methods in the early detection of rolling bearing degradation have problems of low sensitivity and high false alarms. To address these issues, a dynamically adjustment grey incidence analysis ( DAGIA) method for transient mechanical equipment health monitoring is proposed. First, the Hilbert transform is applied to demodulate the vibration data of the rolling bearing to obtain the envelope signal. To weaken the influence of the value of the resolution coefficient to highlight the degree of discrimination of the correlation value, the feature-to-noise energy ratio (FNER) method is introduced into the traditional grey incidence analysis (TGIA) to dynamically adjust the resolution coefficient, which can characterize the strength of bearing faults. Then, the first set of data is extracted at the initial stage of bearing operation as reference data. The dynamic grey incidence analysis is calculated between the remaining data and the reference data and the bearing performance degradation index is established. Finally, according to the normal samples and combined with Chebyshev′s inequality, the control line is set to identify the starting position of the early degradation of the rolling bearing. The IMS and XJTU-SY databases are used to complete the early degradation recognition of rolling bearings. The results show that the proposed method can accurately recognize the starting position of early degradation and the false alarm is close to 0. It has both sensitivity and robustness, which is beneficial for equipment maintenance personnel to better grasp the operating status of rolling bearings.
2022, 43(4):26-37.
Abstract:
In this article, an active fault-tolerant control method based on the optimized adaptive threshold and fault reconstruction strategy is proposed for the nonlinear mechatronic system with uncertain parameters and sensor fault. Firstly, the linear fractional transformation form is used to model the nonlinear mechatronic system with parameter uncertainty. The optimized adaptive threshold based on the particle swarm optimization is established to improve the fault detection performance in the presence of parameter uncertainty. Secondly, the dynamic equations of the system are derived by the analytical redundancy relations, and the tracking control strategy of the healthy system based on the recursive terminal sliding mode is proposed to realize the tracking of load position. An adaptive sliding mode observer is formulated to reconstruct the sensor fault when the fault occurs in the system, based on which the adaptive active fault-tolerant control law is established. The switching of control law can be implemented online by using the fault detection result. Experimental results show that the proposed fault detection and active fault-tolerant control method can accurately achieve fault detection and fault-tolerant control of sensor within 0. 06 s, which evaluates the feasibility of the method.
In this article, an active fault-tolerant control method based on the optimized adaptive threshold and fault reconstruction strategy is proposed for the nonlinear mechatronic system with uncertain parameters and sensor fault. Firstly, the linear fractional transformation form is used to model the nonlinear mechatronic system with parameter uncertainty. The optimized adaptive threshold based on the particle swarm optimization is established to improve the fault detection performance in the presence of parameter uncertainty. Secondly, the dynamic equations of the system are derived by the analytical redundancy relations, and the tracking control strategy of the healthy system based on the recursive terminal sliding mode is proposed to realize the tracking of load position. An adaptive sliding mode observer is formulated to reconstruct the sensor fault when the fault occurs in the system, based on which the adaptive active fault-tolerant control law is established. The switching of control law can be implemented online by using the fault detection result. Experimental results show that the proposed fault detection and active fault-tolerant control method can accurately achieve fault detection and fault-tolerant control of sensor within 0. 06 s, which evaluates the feasibility of the method.
2019, 40(7):178-186.
Abstract:
The existing indoor crowd counting methods face the problems limited scenarios, and low detection accuracy, etc. A crowd counting method based on deep neural networks without carrying equipment is proposed in this study. Multiple wireless fidelity (WiFi) sensor nodes are employed to cover indoor areas. The crossover WiFi link data are obtained by detecting signals among sensor nodes. Deep neural network is utilized to learn and extract the features of the effect of the change of the indoor crowd number on WiFi signals. The crowd counting model is trained for the indoor area, and it can be used to estimate the number of crowd by inputting realtime WiFi signals into the model. Evaluation experiments are implemented in a complex indoor office environment. Results show that the proposed method can realize accurate crowd counting with an accuracy of 8223% and the mean error of 037 people. Compared with other machine learning methods, the deep neural network perception model has higher detection accuracy and can be applied to various application scenarios.
The existing indoor crowd counting methods face the problems limited scenarios, and low detection accuracy, etc. A crowd counting method based on deep neural networks without carrying equipment is proposed in this study. Multiple wireless fidelity (WiFi) sensor nodes are employed to cover indoor areas. The crossover WiFi link data are obtained by detecting signals among sensor nodes. Deep neural network is utilized to learn and extract the features of the effect of the change of the indoor crowd number on WiFi signals. The crowd counting model is trained for the indoor area, and it can be used to estimate the number of crowd by inputting realtime WiFi signals into the model. Evaluation experiments are implemented in a complex indoor office environment. Results show that the proposed method can realize accurate crowd counting with an accuracy of 8223% and the mean error of 037 people. Compared with other machine learning methods, the deep neural network perception model has higher detection accuracy and can be applied to various application scenarios.
2023, 44(9):16-40.
Abstract:
The performance of the inertial sensor directly determines the accuracy of the inertial navigation system. The quantum inertial sensor based on the atomic system is expected to achieve the performance of the traditional inertial sensor with a smaller volume and lower cost, and theoretically can obtain higher measurement sensitivity and long-term stability than the existing technology. In recent years, with the rapid development of the field of quantum precision measurement, the practical and engineering research of quantum inertial sensors has made remarkable progress. In the future, by replacing traditional accelerometers and gyroscopes, it is possible to form a highly integrated, low-power, and low-drift quantum inertial navigation system. This article briefly introduces the basic principles of quantum inertial sensors based on atomic systems, and summarizes the current research status of quantum inertial sensors mainly including atomic interference gyroscope, atomic spin gyroscope, atomic interference accelerometer, atomic interference gravimeter, and gravity gradiometer. The article also reviews and analyzes key technical issues that need to be solved. It provides valuable insight for the development of quantum inertial sensors.
The performance of the inertial sensor directly determines the accuracy of the inertial navigation system. The quantum inertial sensor based on the atomic system is expected to achieve the performance of the traditional inertial sensor with a smaller volume and lower cost, and theoretically can obtain higher measurement sensitivity and long-term stability than the existing technology. In recent years, with the rapid development of the field of quantum precision measurement, the practical and engineering research of quantum inertial sensors has made remarkable progress. In the future, by replacing traditional accelerometers and gyroscopes, it is possible to form a highly integrated, low-power, and low-drift quantum inertial navigation system. This article briefly introduces the basic principles of quantum inertial sensors based on atomic systems, and summarizes the current research status of quantum inertial sensors mainly including atomic interference gyroscope, atomic spin gyroscope, atomic interference accelerometer, atomic interference gravimeter, and gravity gradiometer. The article also reviews and analyzes key technical issues that need to be solved. It provides valuable insight for the development of quantum inertial sensors.
6 Fluxgate magnetometer with large dynamic range and its application in orientation error correction
2024, 44(1):81-89.
Abstract:
The accuracy of geomagnetic vector observation data will be affected by Instrument orientation error, and the existing correction methods require known geomagnetic field modulus or referring to standard instruments, and are difficult to be applied in downhole and marine situations where manual installation and adjustment of instruments are impossible. A large dynamic fluxgate magnetometer and a vector correction method based on Euler rotation transformation are proposed by the paper. The directional error angle of fluxgate can be obtained and the measurement is self-calibrated by this method without referring to standard instrument. The results show that when the orientation error of experimental instrument is set to the large angle of four different quadrants, the correlation coefficient with the corresponding component of station is still above 0. 99 after correction, and the length of confidence interval in the Bland-Altman diagram is reduced by more than 86% and the RMS error is reduced to less than 15% , which proves the effectiveness of this method. Correspondingly, the observation data quality of geomagnetic stations can be improved, and reference solutions for the orientation of downhole and marine geomagnetic observations can be also provided.
The accuracy of geomagnetic vector observation data will be affected by Instrument orientation error, and the existing correction methods require known geomagnetic field modulus or referring to standard instruments, and are difficult to be applied in downhole and marine situations where manual installation and adjustment of instruments are impossible. A large dynamic fluxgate magnetometer and a vector correction method based on Euler rotation transformation are proposed by the paper. The directional error angle of fluxgate can be obtained and the measurement is self-calibrated by this method without referring to standard instrument. The results show that when the orientation error of experimental instrument is set to the large angle of four different quadrants, the correlation coefficient with the corresponding component of station is still above 0. 99 after correction, and the length of confidence interval in the Bland-Altman diagram is reduced by more than 86% and the RMS error is reduced to less than 15% , which proves the effectiveness of this method. Correspondingly, the observation data quality of geomagnetic stations can be improved, and reference solutions for the orientation of downhole and marine geomagnetic observations can be also provided.
2023, 44(12):297-306.
Abstract:
A fusion information enhanced method based on Transformer is proposed to address the issue of misalignment when the current 3D object detection methods fuse different modal data, which mitigates the disruption of correlation between data and data loss. Firstly, a region proposal network of dual fusion feature module based on transformer is designed, which utilizes the deformable attention mechanism to fuse the extracted lidar point cloud features and image features into dual domain features and generate pre-selected boxes. Then, the refinement of box is designed by using a feature information enhancement module, which utilizes a deep completion mechanism to complement the dense depth and feature semantic information. Finally, a multimodal feature cross attention module is designed, which uses a dynamic cross attention mechanism to obtain correlations between different modalities, thereby aligning and fusing feature information effectively. The experimental results based on the Kitti, Nucences, and Waymo datasets demonstrate the effectiveness of method. A large number of ablation experiments have proven the effectiveness and efficiency of each module in the algorithm. The experimental results based on a real vehicle platform show that the algorithm possesses strong robustness in complex practical environments.
A fusion information enhanced method based on Transformer is proposed to address the issue of misalignment when the current 3D object detection methods fuse different modal data, which mitigates the disruption of correlation between data and data loss. Firstly, a region proposal network of dual fusion feature module based on transformer is designed, which utilizes the deformable attention mechanism to fuse the extracted lidar point cloud features and image features into dual domain features and generate pre-selected boxes. Then, the refinement of box is designed by using a feature information enhancement module, which utilizes a deep completion mechanism to complement the dense depth and feature semantic information. Finally, a multimodal feature cross attention module is designed, which uses a dynamic cross attention mechanism to obtain correlations between different modalities, thereby aligning and fusing feature information effectively. The experimental results based on the Kitti, Nucences, and Waymo datasets demonstrate the effectiveness of method. A large number of ablation experiments have proven the effectiveness and efficiency of each module in the algorithm. The experimental results based on a real vehicle platform show that the algorithm possesses strong robustness in complex practical environments.
2023, 44(12):34-43.
Abstract:
The linear motor feed drive axis has the advantages of fast velocity, large acceleration, and short response time. However, the large heat generated by the primary coil easily causes thermal deformation of the external components and affects the positioning accuracy of the feed axis. To solve the problem, a thermal error modeling method for the direct-drive feed axis considering the multi-field coupling of “electromagnetic-thermal-fluid” is proposed. The electromagnetic, thermal, and flow fields of the direct-drive feed axis are analyzed, and the control equation of the temperature field is established under multi-field coupling. A simplified method for solving the control equation is proposed, which separates the equation into three stages, including heat generation, steady-state heat convection, and heat conduction. Then, the three stages are recoupled to obtain an explicit analytical model of the temperature field and the deformation field. The experimental results show that this model reveals the mechanism of the thermal error of direct-drive feed axes. It can accurately predict the thermal error, and automatically adjust the linear and nonlinear intervals, which is of great significance for improving the motion accuracy of direct-drive machine tools.
The linear motor feed drive axis has the advantages of fast velocity, large acceleration, and short response time. However, the large heat generated by the primary coil easily causes thermal deformation of the external components and affects the positioning accuracy of the feed axis. To solve the problem, a thermal error modeling method for the direct-drive feed axis considering the multi-field coupling of “electromagnetic-thermal-fluid” is proposed. The electromagnetic, thermal, and flow fields of the direct-drive feed axis are analyzed, and the control equation of the temperature field is established under multi-field coupling. A simplified method for solving the control equation is proposed, which separates the equation into three stages, including heat generation, steady-state heat convection, and heat conduction. Then, the three stages are recoupled to obtain an explicit analytical model of the temperature field and the deformation field. The experimental results show that this model reveals the mechanism of the thermal error of direct-drive feed axes. It can accurately predict the thermal error, and automatically adjust the linear and nonlinear intervals, which is of great significance for improving the motion accuracy of direct-drive machine tools.
2022, 43(12):39-48.
Abstract:
Current intensity is the basic parameter of the electrolytic aluminum process. The accurate measurement of electrolytic cell current is beneficial to improve the production efficiency and work stability. This article analyzes in principle about the reason why the fiber-optic current sensor has high accuracy in electrolytic aluminum current measurement compared with the traditional current measurement technology. To reduce the repeated disassembly and assembly error and high current temperature error of fiber-optic current sensor in field applications, innovatively proposed key technologies such as the sensing optical fiber plug-in scheme based on flexible packaging, two-dimensional compensation of temperature error and nonlinear error, etc. The handheld fiber-optic current sensor is designed and developed for the first time. The test results show that the error of the sensor under repeated disassembly and assembly is less than 0. 12% , the error when measuring 0. 5 ~ 30 kA current in the temperature range of -40℃ ~ 70℃ is less than 0. 2% . When measuring the column bus current in an electrolytic cell, the error with the displayed value in the control room is less than 0. 4% . Through the continuous monitoring and analysis of the distributed current of the cathode soft discharge, the early warning of the damage of the electrolytic cell is realized.
Current intensity is the basic parameter of the electrolytic aluminum process. The accurate measurement of electrolytic cell current is beneficial to improve the production efficiency and work stability. This article analyzes in principle about the reason why the fiber-optic current sensor has high accuracy in electrolytic aluminum current measurement compared with the traditional current measurement technology. To reduce the repeated disassembly and assembly error and high current temperature error of fiber-optic current sensor in field applications, innovatively proposed key technologies such as the sensing optical fiber plug-in scheme based on flexible packaging, two-dimensional compensation of temperature error and nonlinear error, etc. The handheld fiber-optic current sensor is designed and developed for the first time. The test results show that the error of the sensor under repeated disassembly and assembly is less than 0. 12% , the error when measuring 0. 5 ~ 30 kA current in the temperature range of -40℃ ~ 70℃ is less than 0. 2% . When measuring the column bus current in an electrolytic cell, the error with the displayed value in the control room is less than 0. 4% . Through the continuous monitoring and analysis of the distributed current of the cathode soft discharge, the early warning of the damage of the electrolytic cell is realized.
2023, 44(8):63-73.
Abstract:
To solve the problem of pixel coordinate mismatch and effective pixel missing caused by pixel size difference and phase error between camera and projector, an improved wrapping phase-coordinate mapping method based on virtual field of view is proposed. Firstly, the fringe phase under different overlapping field of view is analyzed to determine the optimal projection mode. Secondly, two sets of horizontal and vertical fringes with high and low frequencies are designed, and the low frequency phase extremum is extracted to calculate the virtual projection field of view. The high frequency period numbered the real and virtual field of view step by step to realize the coarse matching between the small field of view. Finally, by using the improved wrapping phase coordinate mapping method and the criterion of phase difference threshold, the projection pixel coordinates are numbered to obtain the accurate mapping relationship between pixels. Experimental results show that the enveloping phase root mean square error in the same pixel interval is reduced by 78. 6% compared with the continuous phase. In planar and complex surface experiments, the number of effective pixels increases by 9. 21 times and 9. 43 times compared with traditional matching. The proportion of pixel mismatched coordinates decreases from 80. 55% and 59. 4% of traditional phase matching to 14. 26% and 12. 56% . It provides a feasible solution for pixel matching with the same name in adaptive fringe measurement technology.
To solve the problem of pixel coordinate mismatch and effective pixel missing caused by pixel size difference and phase error between camera and projector, an improved wrapping phase-coordinate mapping method based on virtual field of view is proposed. Firstly, the fringe phase under different overlapping field of view is analyzed to determine the optimal projection mode. Secondly, two sets of horizontal and vertical fringes with high and low frequencies are designed, and the low frequency phase extremum is extracted to calculate the virtual projection field of view. The high frequency period numbered the real and virtual field of view step by step to realize the coarse matching between the small field of view. Finally, by using the improved wrapping phase coordinate mapping method and the criterion of phase difference threshold, the projection pixel coordinates are numbered to obtain the accurate mapping relationship between pixels. Experimental results show that the enveloping phase root mean square error in the same pixel interval is reduced by 78. 6% compared with the continuous phase. In planar and complex surface experiments, the number of effective pixels increases by 9. 21 times and 9. 43 times compared with traditional matching. The proportion of pixel mismatched coordinates decreases from 80. 55% and 59. 4% of traditional phase matching to 14. 26% and 12. 56% . It provides a feasible solution for pixel matching with the same name in adaptive fringe measurement technology.
2017, 38(10):2351-2363.
Abstract:
Robot teleoperation is the key technology to implement tasks in space, telesurgery, deep ocean areas. Robot teleoperation based on virtual reality is an efficient way to overcome the time delay problem, which can provide with a teleoperation with hightransparency and strongstability performance, and it is currently the main way to realize robot teleoperation. Firstly, the key components of the robot teleoperation system based on virtual reality are analyzed, and the function of each module is introduced. Secondly, methods to realize virtual reality environment modelling are reviewed, including the geometric modelling and the dynamic modelling methods. The performance of each modelling method is analyzed. Besides, the construction methods and application areas of virtual fixtures are discussed. Furthermore, current problems in this area are summarized and analyzed, and some research directions are proposed.
Robot teleoperation is the key technology to implement tasks in space, telesurgery, deep ocean areas. Robot teleoperation based on virtual reality is an efficient way to overcome the time delay problem, which can provide with a teleoperation with hightransparency and strongstability performance, and it is currently the main way to realize robot teleoperation. Firstly, the key components of the robot teleoperation system based on virtual reality are analyzed, and the function of each module is introduced. Secondly, methods to realize virtual reality environment modelling are reviewed, including the geometric modelling and the dynamic modelling methods. The performance of each modelling method is analyzed. Besides, the construction methods and application areas of virtual fixtures are discussed. Furthermore, current problems in this area are summarized and analyzed, and some research directions are proposed.
2024, 44(1):301-310.
Abstract:
To solve the problem of requiring global optimization, real-time obstacle avoidance, and safe and reliable trajectory in surface ship path planning, a surface ship path planning method based on A ∗ algorithm and DWA algorithm is proposed. Firstly, the heuristic function dynamic weighting strategy is introduced to improve the search efficiency of A ∗ algorithm. Then, considering the motion characteristics of surface ships, an Angle weakening strategy of the path Angle node is adopted to reduce the angle and shorten the global path length. Finally, the trajectory evaluation function of the DWA algorithm is improved based on the influence of the global factors and track safety constraints, and the algorithm fusion is completed by providing subentry points of global path to guide the DWA algorithm to carry out local planning. Experimental results show that the total steering Angle of the proposed algorithm is reduced by 45. 6% and 46. 0% , respectively, compared with the existing fusion algorithms, which verifies the effectiveness and feasibility of the proposed fusion algorithm, and has more advantages over other traditional algorithms.
To solve the problem of requiring global optimization, real-time obstacle avoidance, and safe and reliable trajectory in surface ship path planning, a surface ship path planning method based on A ∗ algorithm and DWA algorithm is proposed. Firstly, the heuristic function dynamic weighting strategy is introduced to improve the search efficiency of A ∗ algorithm. Then, considering the motion characteristics of surface ships, an Angle weakening strategy of the path Angle node is adopted to reduce the angle and shorten the global path length. Finally, the trajectory evaluation function of the DWA algorithm is improved based on the influence of the global factors and track safety constraints, and the algorithm fusion is completed by providing subentry points of global path to guide the DWA algorithm to carry out local planning. Experimental results show that the total steering Angle of the proposed algorithm is reduced by 45. 6% and 46. 0% , respectively, compared with the existing fusion algorithms, which verifies the effectiveness and feasibility of the proposed fusion algorithm, and has more advantages over other traditional algorithms.
2022, 43(2):225-233.
Abstract:
The pseudoinverse-type solution may exceed the physical limit of a manipulator. To address this problem, two kinds of jointvelocity constraint schemes of redundant manipulator based on the pseudoinverse algorithm are proposed. Firstly, according to the tracking task of the end-effector, the pseudoinverse algorithm is used to analyze the redundancy at the velocity level. Secondly, two constraint schemes are utilized to limit and compress the specified joint velocity. A new velocity solution is obtained to perform the specified trajectory-tracking task. Then, the error compensation function is designed to eliminate the position error of the end-effector to ensure the smooth execution of the tracking task. Finally, simulation results based on the MATLAB software are analyzed and discussed. The algorithm is further evaluated using a six-DOF manipulator which is controlled by the Arduino platform. These results show that the maximum tracking errors do not exceed 3×10 -4 m with two proposed constrained schemes. And the time-varying function constrained scheme can achieve better velocity stability when the joint velocity is limited.
The pseudoinverse-type solution may exceed the physical limit of a manipulator. To address this problem, two kinds of jointvelocity constraint schemes of redundant manipulator based on the pseudoinverse algorithm are proposed. Firstly, according to the tracking task of the end-effector, the pseudoinverse algorithm is used to analyze the redundancy at the velocity level. Secondly, two constraint schemes are utilized to limit and compress the specified joint velocity. A new velocity solution is obtained to perform the specified trajectory-tracking task. Then, the error compensation function is designed to eliminate the position error of the end-effector to ensure the smooth execution of the tracking task. Finally, simulation results based on the MATLAB software are analyzed and discussed. The algorithm is further evaluated using a six-DOF manipulator which is controlled by the Arduino platform. These results show that the maximum tracking errors do not exceed 3×10 -4 m with two proposed constrained schemes. And the time-varying function constrained scheme can achieve better velocity stability when the joint velocity is limited.
2022, 43(4):89-97.
Abstract:
Aiming at the problem that the uniformity of the uniform motion coordinate system is reduced due to the end effect of the planar magnetic field linear time-grid displacement sensor developed in the early stage, a method for suppressing the end effect of the planar coil is proposed, and an alternating magnetic field with higher uniformity is constructed. A new type of plane linear time grating displacement sensor that can suppress the end effect is developed. A mathematical model of planar coil excitation is formulated to analyze the influence of the end effect on the uniform magnetic field, and a double-layer complementary excitation coil structure is proposed to suppress the end effect. The dual-column excitation unit realizes the synthesis of travelling wave signals and verifies the effectiveness of the scheme through simulation. A simulation model to analyze the influence of the end effect on the measurement accuracy of the sensor is established, and the sensor parameters are optimized. Based on the PCB process, a new sensor prototype with a measuring range of 228 mm is fabricated and compared with the traditional sensor prototype. Experimental results show that the new planar linear time grating displacement sensor can effectively suppress the end effect of the sensor and improve the measurement accuracy. The intrapolar raw measurement accuracy is improved from ±20 μm to ±10 μm.
Aiming at the problem that the uniformity of the uniform motion coordinate system is reduced due to the end effect of the planar magnetic field linear time-grid displacement sensor developed in the early stage, a method for suppressing the end effect of the planar coil is proposed, and an alternating magnetic field with higher uniformity is constructed. A new type of plane linear time grating displacement sensor that can suppress the end effect is developed. A mathematical model of planar coil excitation is formulated to analyze the influence of the end effect on the uniform magnetic field, and a double-layer complementary excitation coil structure is proposed to suppress the end effect. The dual-column excitation unit realizes the synthesis of travelling wave signals and verifies the effectiveness of the scheme through simulation. A simulation model to analyze the influence of the end effect on the measurement accuracy of the sensor is established, and the sensor parameters are optimized. Based on the PCB process, a new sensor prototype with a measuring range of 228 mm is fabricated and compared with the traditional sensor prototype. Experimental results show that the new planar linear time grating displacement sensor can effectively suppress the end effect of the sensor and improve the measurement accuracy. The intrapolar raw measurement accuracy is improved from ±20 μm to ±10 μm.
2022, 43(1):198-219.
Abstract:
In semiconductor, printed circuit board (PCB), automobile assembly, liquid crystal display (LCD), 3C, photovoltaic cell, and textile industries, the appearance of the product is closely related to the performance of the product. Surface defect detection is an important way to prevent defective products from entering the market. The utilization of machine vision technology to perform inspections with high efficiency and low cost is the main direction of future development. This article reviews the research progress of surface defect detection methods based on machine vision in recent ten years. Firstly, the definition of defect is given, and the general steps of defect detection are described. Then, it focuses on the principle of defect detection using traditional image processing methods, machine learning, and deep learning. The advantages and disadvantages are compared and analyzed. The traditional image processing methods are divided into segmentation and feature extraction. Machine learning consists of unsupervised learning and supervised learning. Deep learning mainly covers most of the mainstream networks for detection, segmentation and classification. Then, 30 kinds of industrial defect data sets and performance evaluation indexes are introduced. Finally, the existing problems of defect detection methods are pointed out and the further work is prospected.
In semiconductor, printed circuit board (PCB), automobile assembly, liquid crystal display (LCD), 3C, photovoltaic cell, and textile industries, the appearance of the product is closely related to the performance of the product. Surface defect detection is an important way to prevent defective products from entering the market. The utilization of machine vision technology to perform inspections with high efficiency and low cost is the main direction of future development. This article reviews the research progress of surface defect detection methods based on machine vision in recent ten years. Firstly, the definition of defect is given, and the general steps of defect detection are described. Then, it focuses on the principle of defect detection using traditional image processing methods, machine learning, and deep learning. The advantages and disadvantages are compared and analyzed. The traditional image processing methods are divided into segmentation and feature extraction. Machine learning consists of unsupervised learning and supervised learning. Deep learning mainly covers most of the mainstream networks for detection, segmentation and classification. Then, 30 kinds of industrial defect data sets and performance evaluation indexes are introduced. Finally, the existing problems of defect detection methods are pointed out and the further work is prospected.
2020, 41(4):138-146.
Abstract:
Abstract:The spindle is the core component of the numerical control machine tool. The vibration caused by the mass imbalance seriously affects the machining accuracy of the machine tool. To suppress the spindle unbalanced vibration, it needs to accurately extract feature of the vibration signal. To identify the unbalanced vibration amplitude and phase of the spindle system, a feature extraction method based on the allphase fast Fourier transform is proposed. The all phase fast Fourier transform can accurately extract the phase and amplitude of the signal by using spectrum analysis function. This method is compared with other three methods to extract the vibration feature of the signal collected by simulation and experiment. Results show that the allphase Fourier transform can achieve better vibration amplitude and phase accuracy and stability. The accuracy of the vibration phase after extraction can reach 97%, and the dynamic balance vibration suppression experiment can be reduced by 6521% after extraction. The effectiveness of the method is further verified.
Abstract:The spindle is the core component of the numerical control machine tool. The vibration caused by the mass imbalance seriously affects the machining accuracy of the machine tool. To suppress the spindle unbalanced vibration, it needs to accurately extract feature of the vibration signal. To identify the unbalanced vibration amplitude and phase of the spindle system, a feature extraction method based on the allphase fast Fourier transform is proposed. The all phase fast Fourier transform can accurately extract the phase and amplitude of the signal by using spectrum analysis function. This method is compared with other three methods to extract the vibration feature of the signal collected by simulation and experiment. Results show that the allphase Fourier transform can achieve better vibration amplitude and phase accuracy and stability. The accuracy of the vibration phase after extraction can reach 97%, and the dynamic balance vibration suppression experiment can be reduced by 6521% after extraction. The effectiveness of the method is further verified.
2023, 44(12):55-68.
Abstract:
With the rapid development of the new energy industry, how to deal with a large number of retired batteries is problem. The secondary utilization scenarios of retired batteries need to be determined based on the state of health (SOH). However, the traditional method of obtaining SOH is time-consuming and energy-consuming. Therefore, the study of fast SOH estimation is very meaningful. The unavailability of historical working condition information and the unknown state of charge at the time of detection make fast SOH estimation very difficult. For this reason, this article proposes a fast SOH acquisition strategy for retired batteries based on the difference in state of charges. In this article, the state of charge′s differences of different SOH retired batteries are used to generate multiple health features. Meanwhile, to select suitable hyperparameters for the random forest algorithm, the genetic optimization random forest regression algorithm is proposed to be applied for SOH estimation. Through experiments, the proposed strategy substantially reduces the estimation time of SOH for retired batteries. Through multiple strategies to avoid contact resistance and wire resistance during measurement, the error of health state estimation of 10 retired batteries is lower than 3% .
With the rapid development of the new energy industry, how to deal with a large number of retired batteries is problem. The secondary utilization scenarios of retired batteries need to be determined based on the state of health (SOH). However, the traditional method of obtaining SOH is time-consuming and energy-consuming. Therefore, the study of fast SOH estimation is very meaningful. The unavailability of historical working condition information and the unknown state of charge at the time of detection make fast SOH estimation very difficult. For this reason, this article proposes a fast SOH acquisition strategy for retired batteries based on the difference in state of charges. In this article, the state of charge′s differences of different SOH retired batteries are used to generate multiple health features. Meanwhile, to select suitable hyperparameters for the random forest algorithm, the genetic optimization random forest regression algorithm is proposed to be applied for SOH estimation. Through experiments, the proposed strategy substantially reduces the estimation time of SOH for retired batteries. Through multiple strategies to avoid contact resistance and wire resistance during measurement, the error of health state estimation of 10 retired batteries is lower than 3% .
2023, 44(11):215-223.
Abstract:
To address the issue of sample imbalance in EEG obtained through rapid serial visual presentation (RSVP), a multi-task learning model for EEG classification is proposed. Firstly, a deep shared feature extraction module is established, which utilizes a convolutional neural network to automatically learn shared parameters and extract depth-related features associated with tasks. Then, a multi-task objective function is constructed based on the classification task and hyper-sphere constraint task, utilizing the joint learning of these two tasks to extract more effective discriminative features and improving the model′s generalization performance. Experiments are implemented on a public RSVP EEG dataset. Compared with commonly used EEG classification models such as DeepConvNet, EEGInception, DRL, and EEGNet, the proposed model named Multi-task EEGNet can achieve the average AUC improvement of 3. 57% , 1. 84% , 6. 22% , and 2. 09% respectively, across 32 subjects. The results indicate that the proposed multi-task learning model can extract discriminative features more fully, effectively improve model classification performance, and better solve the sample imbalance problem in EEG classification tasks.
To address the issue of sample imbalance in EEG obtained through rapid serial visual presentation (RSVP), a multi-task learning model for EEG classification is proposed. Firstly, a deep shared feature extraction module is established, which utilizes a convolutional neural network to automatically learn shared parameters and extract depth-related features associated with tasks. Then, a multi-task objective function is constructed based on the classification task and hyper-sphere constraint task, utilizing the joint learning of these two tasks to extract more effective discriminative features and improving the model′s generalization performance. Experiments are implemented on a public RSVP EEG dataset. Compared with commonly used EEG classification models such as DeepConvNet, EEGInception, DRL, and EEGNet, the proposed model named Multi-task EEGNet can achieve the average AUC improvement of 3. 57% , 1. 84% , 6. 22% , and 2. 09% respectively, across 32 subjects. The results indicate that the proposed multi-task learning model can extract discriminative features more fully, effectively improve model classification performance, and better solve the sample imbalance problem in EEG classification tasks.
2023, 44(1):201-211.
Abstract:
Person re-identification is a fundamental problem in the smart video surveillance system. However, the traditional RGB-based feature extraction method cannot be used in dark environment. A new method for person shape feature extraction using depth measurement is proposed in this article. The depth data are independent from lighting condition. Therefore, the proposed method can be used for person re-id in the dark. Specifically, the point cloud of person is generated from depth data after segmentation and filtering. Then, the point cloud is registered to the initial human body model. The shape and pose parameters of the body model are estimated jointly based on the registered point cloud. Finally, the re-id is achieved by calculating the Euclidean distance in the vector space of shape parameters. The author applies this method on public and self-collected datasets in the laboratory to calculate performance indicators, including Rank-n, cumulative matching curve, and mean average precision, etc. Among the indicators, the Rank-1 of BIWI datasets in Single shot evaluation mode reaches 70. 71% and the Rank-5 of BIWI datasets is up to 92. 32% , which indicate that the proposed algorithm can effectively improve the re-recognition accuracy.
Person re-identification is a fundamental problem in the smart video surveillance system. However, the traditional RGB-based feature extraction method cannot be used in dark environment. A new method for person shape feature extraction using depth measurement is proposed in this article. The depth data are independent from lighting condition. Therefore, the proposed method can be used for person re-id in the dark. Specifically, the point cloud of person is generated from depth data after segmentation and filtering. Then, the point cloud is registered to the initial human body model. The shape and pose parameters of the body model are estimated jointly based on the registered point cloud. Finally, the re-id is achieved by calculating the Euclidean distance in the vector space of shape parameters. The author applies this method on public and self-collected datasets in the laboratory to calculate performance indicators, including Rank-n, cumulative matching curve, and mean average precision, etc. Among the indicators, the Rank-1 of BIWI datasets in Single shot evaluation mode reaches 70. 71% and the Rank-5 of BIWI datasets is up to 92. 32% , which indicate that the proposed algorithm can effectively improve the re-recognition accuracy.
2023, 44(8):41-50.
Abstract:
To solve the problem that industrial robots cannot place products correctly due to mechanical positioning error in fast beat handling task, this article proposes a two-step strategy for two-point corrective positioning based on the improved Blob analysis. The first step identifies the positioning points based on machine vision and image processing techniques using the improved Blob analysis method. The second step quickly locates the position of a large number of products using the principle of planar affine transformation. The improved Blob analysis method is based on image grayscale inversion, histogram equalization, and image filtering techniques combined with threshold watershed segmentation algorithms to achieve reliable segmentation and circle center coordinate extraction for locating circular features in black pallet images that are not easily imaged under limited lighting conditions. Therefore, the two-point positioning strategy could achieve the planar affine transformation matrix before and after the pallet offset based on the reliable pallet circle center position results to complete the pallet offset correction function. Finally, the experiments at the handling station of the actual production line demonstrate that the accuracy and recall rate of the two-step strategy of two-point deflection correction positioning based on the improved Blob analysis are 99. 75% and 99. 75% , respectively, while the product handling time is reduced by 20. 52% . The reliability and efficiency of the robot unloading system are improved.
To solve the problem that industrial robots cannot place products correctly due to mechanical positioning error in fast beat handling task, this article proposes a two-step strategy for two-point corrective positioning based on the improved Blob analysis. The first step identifies the positioning points based on machine vision and image processing techniques using the improved Blob analysis method. The second step quickly locates the position of a large number of products using the principle of planar affine transformation. The improved Blob analysis method is based on image grayscale inversion, histogram equalization, and image filtering techniques combined with threshold watershed segmentation algorithms to achieve reliable segmentation and circle center coordinate extraction for locating circular features in black pallet images that are not easily imaged under limited lighting conditions. Therefore, the two-point positioning strategy could achieve the planar affine transformation matrix before and after the pallet offset based on the reliable pallet circle center position results to complete the pallet offset correction function. Finally, the experiments at the handling station of the actual production line demonstrate that the accuracy and recall rate of the two-step strategy of two-point deflection correction positioning based on the improved Blob analysis are 99. 75% and 99. 75% , respectively, while the product handling time is reduced by 20. 52% . The reliability and efficiency of the robot unloading system are improved.
