Abstract:The defect detection of automotive assembly parts is an important part in the automotive manufacturing process, which can not only improve product quality, reduce the return rate, avoid cost waste, but also provide safety protection for drivers. The earliest defect detection relies on expert experience, which is low accuracy and high labor cost. The nondestructive testing technology relies on media and is not efficient. The introduction of machine vision can not only balance the problem of detection accuracy and efficiency, but also improve the robustness of the detection system, which is one of the most promising defect detection technologies. This article firstly gives the definition and main process of visual defect detection, briefly introduces the hardware of image acquisition in visual defect detection system. Then, the research progress of automobile assembly defect detection in recent years is reviewed from three aspects of commonly used defect segmentation methods, feature extraction methods and convolutional neural networks. The advantages and disadvantages of related methods are compared and analyzed. The automobile assembly parts are roughly divided into four categories, such as wheel tires, body paint, parts and engines. The research status of defect types and defect detection algorithms are summarized. Next, 10 data sets related to the automobile industry and defect detection performance evaluation indicators are introduced. Finally, it is pointed out that the defect detection of automobile assembly is faced with many technical challenges, and the prospect of further work is given.

Abstract:To address the problem of anomalous events occurring in a specific local region of the foreground in an image, with the background region posing interference for anomaly detection, proposes a dual-stream multi-task anomaly detection model. The model architecture consists of a future frame prediction network and an optical flow reconstruction network. Firstly, the optical flow information of the video frame image is extracted by the deep optical flow network, and the foreground detection algorithm is used to obtain the foreground object region of the natural image and the optical flow image. Secondly, the encoding-decoding network is used to complete the future frame prediction and motion reconstruction, and the motion features and apparent features are extracted. Finally, the deep probability network is used to give the probability as the decision to judge the anomaly, and it is combined with the reconstruction loss and the prediction loss to determine the anomalous nature of the images. In this article, the anomalousness of the proposed model is evaluated on three video surveillance datasets (UCSD pedestrian dataset, Avenue, Shanghai Tech) of large scenes, and the proposed method achieves AUC values of 97. 4% , 86. 4% and 73. 4% on the three datasets, respectively. Compared with existing works, the proposed model architecture is simple and easy to train, and the anomaly detection results are more accurate.

Abstract:When a helicopter rotor blade rotates at high speed, the measurement and analysis of blade lead-lag motion in the full field are of great significance for the load design of helicopter blades and rotor structure design. In this article, a stereo vision measurement system for blade lead-lag motion is designed, and the lead-lag motion mode fitting analysis is carried out. First, the existing stereo vision measurement system of the subject group is improved, and the 3D coordinates of the rotating blade marker points at different collective pitch and cyclic pitch are measured in wind tunnel tests. Secondly, the blade lead-lag motion in the hub coordinate system is calculated. Finally, the first order polynomial is used to fit the blade lead-lag motion pattern at a specific moment, and the compound sine function is used to fit the blade lead-lag motion law during rotation. Experimental results of wind tunnel measurements show that the root mean square error of the measured lead-lag motion in the 4. 6 m×4. 6 m scenario is less than 1 mm. Blade lead-lag motion modes and law models fit well, with root mean square error less than 1 mm, providing data support for helicopter blade design.

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.

Abstract:To address the problems of inaccurate camera pose estimation and insufficient utilization of environmental semantic information in most SLAM systems in dynamic environments, proposes a dynamic object detection algorithm based on the instance segmentation, keyframe detection, and Bayesian dynamic feature probability propagation, and three-dimensional reconstruction of static objects in the environment. To construct a multi object monocular SLAM system in a dynamic environment, the system performs instance segmentation and feature extraction on key frame input images, which could obtain a set of potential moving object feature points and a set of static object feature points. A set of non-moving object feature points is used to obtain inter frame pose transformation, Bayesian probability propagation of dynamic and static feature points are utilized for ordinary frames, and a set of static feature points is used to achieve accurate estimation of camera pose. Joint data association is performed on static objects in key frames, and after sufficient data is available, multi object 3D reconstruction is performed to construct a multi object semantic map. Finally, multi object monocular SLAM is achieved. The experimental results on TUM and Boon public dataset show that in dynamic scenarios, compared to the ORB-SLAM2 algorithm, the RMSE of APE decreases by 54. 1% and 58. 2% on average.

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.

Abstract:With the continuous fusion and optimization of multi-link, multi-equipment manufacturing process and information in the current intelligent manufacturing scenarios of aviation, aerospace and large scientific instruments, the demand for distance measurement is being promoted from the traditional local, single-parameter, off-line and time-sharing to the global, multi-objective online, real-time and synchronous. The existing laser ranging methods are limited by principle and face bottlenecks in precision, real-time performance and parallel capability. This article develops a dual-femtosecond laser multi-target absolute distance measurement system based on the time-of-flight method. A specific fiber beam splitting scheme is designed to make femtosecond laser pulses reach multiple targets at the same time. Multiple sets of pulses with distinguished feature matching in the temporal domain are constructed. The pulses carrying distance information are transferred into the FPGA data processing unit after the intensity cross-correlation module, and the real-time multiple absolute distances and realizing the integrated design of the whole system are calculated. Finally, the system for verification and the experiment results show that the ranging accuracy of the system is better than 4 μm and the allen deviation is better than 10 -5 m at the updating speed of 2 kHz. On this basis, the parallel optical path structure is established to realize the simultaneous measurement of multiple degrees of freedom, and the measurement deviation is better than 5″.

Abstract:To address the problems of low echo signal-to-noise ratio ( SNR) and Lorentz force failure in the online monitoring of fiberwrapped gas cylinder by electromagnetic acoustic transducer (EMAT), an attached EMAT shear horizontal (SH) guided wave detection method based on coding compression is proposed. Through finite element modeling, the effects of Barker code sequence bits and subpulse carrier period, chirp signal pulse width and band width, combined Barker code combination on pulse compression signal SNR and resolution are analyzed and evaluated experimentally. Finally, the 20 mm×0. 5 mm×2 mm crack defect of the fiber-wrapped gas cylinder is detected. The results show that the metal attached method can effectively solve the problem of EMAT failure in non-metallic materials. Increasing the sequence bit and carrier period of Barker code, increasing the chirp pulse width, decreasing the chirp band width, and increasing the sequence length of combined Barker code can improve the SNR of pulse compression signals. After optimization design, compared with the traditional Tone-burst excitation method, the SNR of defect echo can be improved by at least 20. 4 dB by pulse compression technology. Among different coding algorithms, chirp signal excitation has the highest SNR, reaching 23. 9 dB. When 3 × 13-bit combination Barker code is adopted, the main lobe width of pulse compression signal is 28. 8 μs. The resolution is highest, but the SNR is low.

Abstract:To realize the in-situ stress measurement of large composite plates, an air-coupled ultrasonic stress measurement method for composite plates based on Lamb waves is proposed. At present, due to the severe impedance mismatch of the air-coupled ultrasonic transducer, energy loss is severe, and the air-coupled ultrasonic signal is relatively weak, making it difficult to accurately extract acoustic features. Meanwhile, the application of the air-coupled ultrasonic Lamb wave stress measurement method in composite materials is still in the exploration stage of theory and experiment. The Lamb wave acoustic elastic effect of carbon fiber composite materials is explored through experimental methods. According to composite plates ’ the phase velocity and the optimal incident dispersion characteristics analysis, the center frequency of the air-coupled ultrasonic transducers, and the excitation method of the relatively pure A0 mode Lamb wave is determined. The accuracy of sound time extraction is ensured by a good signal-to-noise ratio of the spatially coupled ultrasonic Lamb wave. To evaluate the effectiveness of the proposed method, seven different tensile specimens are obtained along the fiber directions of 0°, 15°, 30°, 45°, 60°, 75°, and 90° for stress measurement. The experimental results show that the measurement error is less than ±8. 1 MPa in the range of 0~ 100 MPa, and the measurement repeatability is 7. 5 MPa. This method has significant advantages in measurement accuracy and repeatability and could provide an advanced and feasible technique for in-situ stress measurement of large composite plates.

Abstract:The superconducting electromagnetic detection is highly susceptible to interference from environmental electromagnetic noise, which seriously affects the accuracy of data interpretation. To address this issue,this paper proposes a superconducting electromagnetic noise suppression method based on the principle of signal-to-noise separation and denoising. Firstly, the FastICA algorithm is utilized to extract the secondary magnetic field signal from the observed signal. Then, an optimization constraint model is formulated, which is based on maximizing non-gaussianity and minimizing distortion to address the uncertainty of the signal amplitude after separation. Finally, the CS search algorithm is used to iteratively solve for the optimal parameters of the separation matrix. Simulation experiment results show that the proposed method outperforms the PCA, WA and ICA methods in terms of signal-to-noise ratio and mean square error metrics, with a 16. 6 dB improvement in signal-to-noise ratio after noise reduction. Field experiment results show that the proposed method has good suppression effects on various environmental electromagnetic noise. The effective signal duration is increased by nearly 4 times after denoising. The observed signal quality is significantly improved, and the imaging interpretation depth reaches 1 000 m. The effectiveness and practicality of the proposed method is fully evaluated.

Abstract:Girth weld is an important reference for the pipeline test data analysis which can be used to correct cumulative errors of odometer wheel. Thus, girth weld localization is the necessary part in test data analysis task. In this article, a lightweight convolution neural network with multi-scale receptive field is proposed, which is based on the magnetic flux leakage (MFL) data feature and the girth weld spatial distribution characteristics on the wall of pipeline. The features of girth weld can be extracted efficiently due to the axial 1D convolution with single sensor receptive field and circular ring convolution with circular global receptive field. Inspired by the label smoothing, the sample label is augmented. Moreover, some optimization design for loss function and activation function are also achieved. And a girth weld intelligent localization model is established. Finally, the model is trained and evaluated by the dataset including 5 676 samples that collected from various pipelines on-line MFL inspection data. The experiment results show that model has good convergency stability, and the test precision and recall rate reach 93. 90% and 94. 79% , respectively. Furthermore, the model is tested by MFL data of Φ610 pipeline which never participate in model training. The F1 score reaches 0. 93 which shows that the model has a good robustness and generalization ability, and has certain application engineering value.

Abstract:To improve the accuracy of positioning and quantifying delamination defects in carbon fiber reinforced polymer, a precise frequency domain total focusing method three-dimensional imaging technology based on two-dimensional equivalent sound velocity mapping is proposed. This technique first changes the division standard of the sub-matrix from the excitation point position to the excitation-receiver spacing, and obtains a new full matrix data. Then, the transformed two-dimensional sub-matrix is transformed into the frequency domain. According to the change of depth and spatial frequency, an accurate two-dimensional equivalent sound velocity mapping is derived to match the transceiver separation signal in the sub-matrix data, and the sub-matrix frequency domain reconstruction diagram is obtained by combining the angular spectrum operation. The sub-matrix focusing image is obtained by inverse fast Fourier transform and fused to obtain the full focusing image. Finally, the final three-dimensional image is obtained by using multi-plane threedimensional reconstruction technology. Compared with the traditional F-TFM and F-SAFT algorithms, the results show that the proposed algorithm effectively suppresses the generation of side lobe effect. The quantitative error of double defect spacing is reduced by 20. 31% , and the quantitative error of defect width is reduced by 5. 43% and 6. 3% , respectively. When the defect depth and the double defect spacing change, the higher detection sensitivity can still be guaranteed.

Abstract:The sensitivity of different modes to cracks at different locations are different. In response to the problem of inaccurate mode selection when detecting cracks, the selection of optimal crack-sensitive mode ( SOCSM) method evaluated by three indicators is proposed, which considers the effects of cumulative second harmonic modes combination, mode vibration pattern and crack location under 50 kHz excitation. The sensitivity of modes to cracks is evaluated by the trend of relative nonlinear coefficients with crack growth. The optimal mode excitation (OME) algorithm is proposed to accurately excite the required modes. To evaluate the effectiveness and accuracy of the SOCSM and OME methods, numerical simulation analysis and experimental verification are conducted by using the crack at the waist of the CHN60 rail as an example. The results show that the OME method can accurately excite mode combination 11 and mode combination 1. Under 50 kHz excitation, the mode combination 11 β monotonically increases with crack growth, while mode combination 1 β has almost no change. The mode combination 11 β is more suitable for detecting cracks at the rail waist and is very sensitive to small changes in cracks. Both modal combinations can generate cumulative second harmonic. The presence of cracks can affect the propagation of guided waves, which affects the amplitude of the signal.

Abstract:The stone cultural relics have been weathered and cracked for many years and are in urgent need of repair. To address this issue, presents a defect location algorithm for stone cultural relics based on defect feature enhancement and convolutional neural network (CNN) in view of the deficiency of depth learning in the application of crack detection for cultural relics. The algorithm extracts features through wavelet packet decomposition and selects an effective frequency band containing rich feature information as the input of the CNN network. Through model training and waveform classification recognition, it narrows the location range and improves the generalization ability and recognition rate of the crack location algorithm. An ultrasonic detection platform is established with field programmable gate array ( FPGA) as the core, and experimental verification is implemented on a cube specimen with a side length of 40 cm. The experimental results show that the accuracy rate of waveform identification is 11. 3% higher than that of traditional algorithms, and the average positioning error is less than 10% , which provides a reliable basis for defect detection of stone cultural relics and is helpful for the protection and restoration of cultural relics.

Abstract:In response to the difficulty in detecting and locating HWPN leaks, this article fully considers the multipath effect of negative pressure wavs (NPWs) propagation in the pipeline network and proposes a HWPN leak location method based on the shortest path planning of NPW propagation. First, the shortest paths from each potential leak point to each pressure transmitter are searched, and the shortest time delay of NPW propagation in the HWPN is calculated to form a time delay standard library. Then, the actual arrival time delay of NPWs that is measured in real time is compared with the time delay standard library to determine the leak location. Small leak location experiment in an on-site 13 km×5 km HWPN is implemented to evaluate the proposed method. When the number of effective pressure monitoring points is only 5, this method can realize zero-error leak location within the time-delay match range of 1 000 m radius. This method utilizes accurate and non-fuzzy principle, which can also avoid installing pressure transmitters on all branches of huge HWPN in field. Therefore, it has important application value.

Abstract:Transparent materials are widely used in aerospace and optical equipment and other fields. But, due to their specular properties and optical properties, such as transmission and multiple reflections, efficient automatic detection methods for their tiny defects are difficult in the field. The main defect is the extremely low contrast of defect imaging. To obtain high-contrast images of defects, the existing methods need to adopt complex system structures or repeated angle adjustments. In this article, based on the modulation effect of defect morphology on structured light, a high-contrast defect imaging method with simple structure is proposed. The method is not sensitive to camera angle. Compared with the current dark field illumination method, the experimental results in different thickness defect specimens show that the contrast of the images collected by this method has an average of 27. 58% and a maximum of 37. 41% improvement in the impact damage, scratches, and abrasion defects. The proposed method improves the defect contrast from the perspective of imaging, and contains richer defect feature information. When using a variety of deep learning detection algorithms for comparative experiments, the current best dark-field lighting method has a maximum mAP of only 34% , while the method in this article is close to 80% , which is a significant improvement.

Abstract:Fault diagnosis is an important part of industrial system health monitoring. Existing data-driven diagnosis methods often use balanced datasets for fault modelling. However, in practical applications, industrial systems often produce many samples with imbalanced distribution, which pose challenges to data-driven fault diagnostics. This issue receives extensive attention from the academic and industrial communities. Many results have been achieved in this area. However, there have been a few reviews on the imbalanced data-driven fault diagnosis. It is difficult to clarify the real challenges and future research directions. In response to this problem, a comprehensive review on the research progress in data-driven diagnostic methods and diagnostic application scenarios is provided. It proposes the challenges and future prospects facing the field, which could provide a reference for the research and application of the fault diagnostics.

Abstract:Gas flow meters are important instruments for natural gas trade measurement, and changing of their health status can cause measurement deviation. To reduce economic loss for gas companies, this article proposes a gas flow meter health status assessment method based on multimodal data augmentation, morphological feature learning, and multiscale adaptive weighted morphological network. Firstly, data augmentation is performed by using the ACGAN algorithm based on Wasserstein distance and spectral normalization to achieve sample balance. Secondly, considering the complexity and noise impact of the gas flow meter vibration signal data, a morphological method based on average hat transform is proposed to extract the positive and negative pulse information from the signal. Finally, to address the non-stationary and variable operating conditions in industrial settings, a multiscale adaptive weighted morphological network is introduced. Multiple components with different structural element scales are used to extract pulse information, and adaptive weighted fusion is employed to enhance the scales that provide strong pulse components. The experimental results show that the proposed method has an accuracy of over 94% in the health status assessment of gas flow meters. This method has significant practical value for actual gas trade measurement.

Abstract:In general, the performance of electronic products deteriorates before failure. But, the traditional life prediction method does not use the degradation information. The power supply module of electronic trip unit is taken as the research object, the correlation between the MOSFET switching period and the circuit weak link electrolytic capacitor degradation is analyzed, the MOSFET switching period is taken as the characteristic quantity of performance degradation of power supply module, and the performance degradation model of the power supply module of electronic trip unit is formulated with the MOSFET switching period as the characteristic parameter. The power supply module is divided into health state, attention state and danger state, and the health state transition graph. The calculation method of transition time is determined, and the health state evaluation model is formulated. The accelerated degradation experiment of the power module under temperature stress is implemented to evaluate the performance degradation model and health state evaluation model. It predicts that the average transfer time of the power supply module from the health state to the attention state is 3 906 days under 40℃ environment, and the average transfer time of the power supply module from the danger state is 9 296 days.

Abstract:In this article, a non-linear bearing dynamic model considering bearing clearance, friction and slip is formulated, and the kinetic model is combined with Archard wear theory to investigate the wear characteristics of rolling bearing. Firstly, through the kinetic equations, the non-linear contact force and contact surface slip velocity during the operation of the bearing are calculated. Then, the calculation results are brought into the wear model to obtain the wear distribution of the outer raceway of the bearing. With the radial clearance of the bearing after wear, the kinetic model is updated to study the wear characteristics of the bearing as well as the vibration response of the bearing. The trend of the wear distribution of the bearing under different surface roughness is also studied. The results show the roughness has significant impact on the wear of bearings. The wear rate of the bearing outer race under the roughness of σ = 0. 42 μm is 6. 9 times of that of σ= 0. 2 μm. The research results provide a theoretical basis for the evaluation of the operating condition and life prediction of bearing systems.

Abstract:In this article, a hierarchical multi-block stacked performance-relevant denoising auto-encoder (HMSPDAE) is proposed to evaluate the process operating performance for plant-wide industrial processes with multiple sub-processes, low data difference among different operating performances, and strong noise interference. First, the whole process is divided into a hierarchical structure according to the process characteristics. Then, a method of stacked performance-relevant denoising auto-encoder is proposed to extract the performance-relevant deep features from the process data which are used to realize the operating performance assessment of each subprocess as well as the whole process. In further, a HMSPDAE-based whole-process evaluation model is formulated. The proposed method can effectively reduce the model complexity and enhance the interpretability of the model. Finally, simulation experiments are conducted in the wet metallurgical process. The results show that the assessment accuracy of HMSPDAE reaches 99. 5% and 99. 38% in two different experiments, which are both better than other methods.

Abstract:There are wide application requirements for three-dimensional electric field sensors. In the presence of space charges in the surrounding environment, charges may accumulate at the surface of the electric field sensor, which is a key factor affecting the accuracy of electric field measurement. Especially for the MEMS three-dimensional electric field sensor, there is still a lack of relevant mechanism research and solutions. To solve this problem, an anti-charge-interference technology for the three-dimensional electric field sensor is studied. The effect of charge accumulated at the sensor surface on electric field measurement is studied, and the relationship between the interference of accumulated charge and size of the sensor structure is analyzed. The correctness of the theoretical analysis is evaluated by finite element simulation and experiment. On this basis, an anti-charge-interference method for the three-dimensional electric field sensor is proposed. By differential calculation of the signals from the sensing elements, the influence of surface charge accumulation on electric field measurement can be eliminated. An anti-charge-interference three-dimensional electric field sensor is developed. Experiments show that the measurement error is less than 4. 2% under the electric field of 0~ 30 kV/ m in the presence of charge accumulation interference.

Abstract:The optical sensor is the core part of the optical partial discharge (PD) detection system in gas insulated switchgear (GIS). In this article, simulation and experiment studies are implemented for the design of optical PD sensors based on the light guide rod. Firstly, based on the TracePro simulation software, the front microstructure and length of the light guide rod are designed, and an optical sensor for GIS PD detection is proposed. Then, the GIS optical PD experimental platform is established to study the actual PD detection effects of different front microstructures and lengths. Both the simulation and experiment results show that the light guide rod with its front microstructure being concave cone shape and the length being as long as the inner wall of the tank has better detection effect. The consistence between the simulation and experiment results demonstrates the feasibility of the design and the correctness of the simulation model. Finally, several detection surfaces and detection points are set up in this simulation model, and the total irradiance maps of each detection surface and the light irradiance value of each detection point are observed and analyzed. The light irradiance values received by different detection points are related to component occlusion and the intensity of direct ray. Taking the y-115 PD source as an example, the light spots on the P1 detection surface are concentrated around 150° to 210°, while the optimal detection points on the P2 detection surface are mainly located around 30° and 330°. The research results provide some basis for the development of optical PD sensors in GIS.

Abstract:In response to the common problems of long response time, poor temperature measurement accuracy, and high radiation error of temperature sensors used in high-altitude detection, a temperature sensor with small packaging volume, fast response speed, strong radiation resistance is developed. Mn-Ni-Cu-Fe-O-based chip thermistors with dimensions of 0. 4 mm×0. 4 mm×0. 25 mm and 0. 6 mm× 0. 6 mm×0. 25 mm are prepared by the solid-phase method. On the sensitive surface of the chip thermistor, an insulation film and an aluminum metal reflector film are deposited sequentially. The results show that the resistance range of the developed thermistor is 1 000~ 4. 2 kΩ and 375 ~ 1. 6 kΩ in the temperature range of - 80℃ to 60℃ . Thermal dissipation and response speed tests are conducted on the thermistor, and the results show a thermal dissipation coefficient of 1. 034 mW/ ℃ and a response time of 0. 63 s for the developed sensor. The temperature measurement error of the thermistor is studied by using a solar simulator, and the minimum temperature measurement error is 0. 22℃ under an irradiance intensity of 100 W/ m 2 . The thermistor has the potential to be used in the sounding instrument of high-altitude meteorological detection.

Abstract:The independent decoupling control of two motors in the six-phase series three-phase PMSM system can be realized by controlling the voltage vectors of different planes in the static coordinate system. Based on this, the rotating high frequency voltage is injected into the six-phase plane and the three-phase plane at the same time in the static coordinate system. By demodulation of the negative sequence component of the corresponding plane high frequency current, the preliminary observed rotor position angles of the two motors are obtained by decoupling. To compensate the error caused by stator resistance, speed and filter, the rotor position angle is compensated by combining the error angle information in the positive sequence component of high frequency current with the speed feedforward. The experimental results show that the average absolute value of the steady-state rotor position angle observation error values of six-phase and three-phase motors are 0. 146 and 0. 106 rad, and the maximum error values under dynamic conditions is 0. 2 rad. The sensorless direct torque control system based on the observed rotor position angle has excellent performance in various operating conditions at low speed.

Abstract:The radar emitter signal is susceptibility to noise interference of feature information and low real-time performance of sorting, etc. To address these issues, this article proposes a method for online signal data stream classification based on ambiguity function multidimensional structural metrics. Firstly, the concept of global image similarity is leveraged, and the signal′s ambiguity function is denoised by using an integral-accelerated non-local means smoothing method. Then, multidimensional structural distribution features are extracted from both the main and side perspectives of the processed ambiguity function, and a comprehensive feature vector is formulated. Finally, a semi-supervised learning classification model is optimized and applied in real-time to the continuous input stream of signal feature vectors, and instantaneous classification outcomes are achieved. Experimental results show that with less prior data, the proposed method maintains a classification success rate of 99% or higher in signal-to-noise ratios ranging from 8 to 18 dB, and the accuracy can also reach 91. 8% even at 2 dB. Moreover, the average time required for extracting features from a single signal is a mere 0. 29 second. These results evaluate the effectiveness and real-time capability of the proposed method, and show significant engineering value.

Abstract:Remote photoplethysmography ( rPPG) is a video-based and the non-contact heart rate ( HR) detection technique that is susceptible to motion noise due to weak amplitudes of blood volume pulse signal that carries, making it challenging to accurately detect video-based HR values in motion scenarios. This article studies HR estimation based on a combination of chrominance ( CHROM) signals and joint blind source separation (JBSS). On the one hand, CHROM is applied to each facial region of interest (ROI) and timedelay operation is followed by constructing two multi-channel datasets, which can highlight quasi-periodic variables while suppressing irregular motion noises. On the other hand, JBSS technique is applied to the generated two datasets to extract the common underlying source component vectors (SCVs), where the one indicating BVP signal is selected and the HR is measured. The proposed method is evaluated on two public databases UBFC-RPPG and ECG-Fitness, and compared with several other typical methods. The results show that the method achieves the best performance of HR estimation during the dramatic sport situation, with HRmae = 9. 93 bpm, HRrmse = 16. 17 bpm and r = 0. 75. It provides a solution for the practical applications of the rPPG technology.

Abstract:A multi wavelet transform sliding feature detection method is proposed to address the difficulty in identifying sliding feature signals during the process of robotic arm grasping targets. Firstly, the mechanism of flexible tactile slip sensing based on FBG sensing is studied, and a double-layer “cross” type distributed sensing unit based on FBG is designed. Secondly, a tactile perception experimental platform is established and tactile perception experiments are implemented on the dynamic grasping process. Then, based on the db10 wavelet denoising method, the sliding perception signal is denoised. Finally, a sliding signal feature separation and perception method using the Mexican hat continuous wavelet and the first-order Haar discrete wavelet is proposed, and relevant experimental research is conducted. The experimental results show that the detection threshold of wavelet detail coefficients is ±2×10 -4 , and the average accuracy of sliding detection with different grip forces can reach 98. 88% , which can accurately identify the sliding state of the target being grasped by the robotic arm.

Abstract:In this article, a remote calibration method of voltage source based on the GPS common-view method is proposed to address the problems of traditional voltage source calibration methods. This method is based on the transfer of non-physical standards and can enable the standard to be placed in the laboratory without having to be transferred to the calibrated site for remote calibration. It can solve the problem of introducing additional errors. By designing the voltage source remote calibration system based on the common-view method, the calibration model is formulated. According to the common-view principle, remote comparison of voltage values between the standard side and the calibrated side can be achieved, and remote calibration of the voltage source can be completed. The calibration results of the remote calibration method and the traditional calibration method at the DC voltage 0~ 1 V calibration point are compared. The results show that the difference between the voltage differences measured by the two methods is within 5. 2×10 -5 V. The uncertainty evaluation process of the two methods at the DC voltage 1 V calibration point is given, and the consistency evaluation of the calibration results of the two methods is carried out by the transfer comparison method. The uncertainty evaluation results show that the extended uncertainty of the remote calibration method at the 1 V calibration point is 9. 182 44×10 -5 V ( k = 2). The calibration results of the remote method are consistent with the traditional results of the remote method.

Abstract:Aiming at the characteristics of large inertia, large hysteresis and unstable parameters of the air and flue gas system of coalfired generator sets, a sliding mode active disturbance rejection control strategy based on generator sets is proposed. The fuzzy radial basis function (RBF) algorithm is selected to identify the model, the gradient descent method is used to coarse-tune the neural network weights, and the genetic algorithm is used to fine-tune the neural network weights. The internal and external disturbances of the system are estimated by the extended state observer, the nonlinear state error feedback rate is designed and sliding mode control strategies are designed to overcome the inertia, hysteresis and disturbances of the system, and Lyapunov functions are designed to evaluate the stability of the control system. The simulation results show that the designed control strategy reaches the set value in 38 s with no overshoot compared with the cascaded proportion integration differentiation (PID) control, sliding mode control and self-rejecting control in the case of model mismatch. When a 20% backward step disturbance is applied to the system, the system regulation time is 39. 5 s with 3. 4% overshoot. The regulation time in the case of model mismatch is 43. 2 s with no overshoot. When the system applies 20% reverse step disturbance, the system regulation time is 46. 4 s with 3. 87% overshoot. The engineering application results show that the primary air volume control deviation is within ±10 000 m 3 / h, which is 21% lower than the fluctuation range of the cascaded PID control, and the anti-disturbance capability and robustness of the system are improved.