Abstract:When the magnetic flux of a permanent magnet synchronous motor changes rapidly, it is approximately assumed that a magnetic flux change rate of 0 will constrain the performance of the PMSM fault-tolerant control system. This paper proposes a model free super twisting sliding mode composite fault-tolerant control method based on intelligent proportional integral differential (iPID-MFSTSMC). Firstly, a mathematical model for PMSM demagnetization fault was constructed, taking into account the variation of permanent magnet magnetic flux. Based on the constructed model, a magnetic flux sliding mode observer was designed to observe the magnetic flux and its rate of change. In addition, a load torque derivative sliding mode observer was designed for rapidly changing load torque. Then, an iPID model free fault-tolerant control method was designed based on observations of magnetic flux and its change rate, as well as load torque derivatives. The proposed control method uses a super twisting sliding mode algorithm to suppress observation errors and other disturbances, improve dynamic performance, and reduce the difficulty of controller parameter design. The stability of the designed fault-tolerant control method was proved using Lyapunov stability theorem, and parameter design conditions were provided. Finally, under the condition of rapid change of flux and load torque, simulation and experimental analysis show that the proposed method can effectively reduce the system jitter and reduce the overshoot.

Abstract:In order to investigate the evolution of the vibration response in the multi-scalp gear system, a multi-scalp gear dynamics model is constructed by integrating the time-varying meshing stiffness. Firstly, a multi-peeling gear meshing stiffness model is established by integrating the modified energy method and the single/double tooth pair meshing angular displacement calculation method; Secondly, a 6-degree-of-freedom multi-stripping gear dynamics model is constructed by considering the time-varying meshing stiffness, tooth side clearance and initial pressure angle, and the model validity is analysed from the simulation, theoretical and experimental dimensions; lastly, the effects of changes in the tooth root, pitch line, and tooth apex positions on the vibration response of the multi-peeling gear are investigated on the basis of the model; the evolution of the vibration response is obtained when the simultaneous peeling occurs. Finally, based on the constructed model, we study the influence of the change of spalling size on the meshing stiffness when spalling occurs at the same time, and obtain the evolution of the vibration response of multi-spalling gears. The results show that the error between the constructed model and the theoretical and experimental errors is less than 0.5%, and the evolution of the meshing stiffness and vibration response of the flaking at the tooth root, pitch line and tooth apex position is obtained. With the expansion of the flaking width, the vibration amplitude near the pitch line increases significantly; with the increase of the flaking depth, the vibration impact on the tooth apex position is more intense; and with the increase of the flaking length, the range of flaking is enlarged. The obtained conclusions provide theoretical guidance for the health monitoring and early fault diagnosis of the gear system.

Abstract:In order to deeply analyze the influence of local failure of rolling ball bearing on the overall vibration characteristics of the bearing, a numerical simulation model of ball bearing-rotor system dynamics under rotor eccentric force excitation is constructed by using Hertz contact theory, and the impact excitation response equation between rolling body and various types of fault areas is deduced by combining the bearing clearance changes caused by local failure of outer ring, inner ring and rolling body of deep groove ball bearing. On this basis, by analyzing the change of internal contact load of balls caused by single-point damage, the influence law of single ball contact stress on the load of adjacent rollers and the overall vibration characteristics of the system is revealed. By analyzing the changes in the number of effective bearing balls during the bearing rotation, the relationship between the ball load occupancy time and the amplitude of fault impact oscillation is established. The analysis results show that the contact separation of the ball and the fault area causes its load to oscillate, and the oscillation frequency of each fault state is the system resonance frequency; the larger the running time share on the 2 equivalent load-bearing rollers, the greater the intensity of the fault shock oscillation. The Poincar- mapping point of each type of early fault state keeps near the same amplitude. When the degree of local failure increases to a certain degree, the chaotic characteristics of the system vibration are enhanced and the vibration amplitude increases accordingly.

Abstract:Aiming at the problem that the setting of the complete failure threshold of rolling bearings is mostly selected according to artificial experience, and the degradation trajectory adaptation ignores the overall morphological trend change of the time series, a method for setting the failure threshold and predicting the remaining life of rolling bearings based on the consistent offset distance of morphological fluctuation is proposed. Firstly, the forward difference (FD) is introduced to preprocess the vibration signal, and the root mean square (RMS) value of the processed signal is calculated as the degradation indicator (DI). Secondly, the double exponential model is used to fit the DI curve to determine the total failure threshold (TFT) of the final reference bearing, so as to reduce the setting deviation of TFT. Finally, the similarity of the DI curve is calculated by using the morphology fluctuation conformance deviation distance (MFCDD) to complete the setting of the failure threshold of the test bearing, and the remaining useful life (RUL) prediction of the rolling bearing is completed by using the particle filter to update the double exponential model. The experimental results on the XJTY-SY dataset show that the score of rolling bearing RUL prediction is 82.97% and 73.64% higher than that of dynamic time warping matching method, convolutional neural network and bidirectional long short-term memory network prediction method, respectively. The experimental results on the PHM2012 dataset show that the score of rolling bearing RUL prediction is 99.99%, 60.65% and 99.90% higher than that of dynamic time warping matching method, convolutional neural network and bidirectional long-term and short-term memory network prediction method, long-term and short-term memory and self-attention mechanism prediction method.

Abstract:Aiming at the variable speed condition, the bearing vibration signal is prone to signal feature aliasing, frequency shift, signal truncation and noise pollution, a fault classification model combining angular resampling, principal component analysis (PCA) and extreme gradient boosting tree (XGBoost) is proposed. Secondly, the time-frequency feature parameters are extracted by principal component analysis (PCA), and the main elements with total contribution greater than 95% are selected as input samples of XGBoost model; finally, the main parameters of XGBoost are tuned by grid search method, and the model is trained by dividing the training set and the test set to verify the accuracy of its fault classification. The results show that the accuracy of fault diagnosis is 96.44%, the running time is shortened by 27.24 s compared with that of the data without dimensionality reduction, and the diagnosis effect after angle resampling is obviously better than that of the diagnosis effect without angle resampling, and the fault recognition rate is improved by more than 7%, which proves that the proposed method can make diagnosis more quickly and accurately.

Abstract:Aiming at the problems of difficult selection of criteria and threshold setting for series arc fault detection, this paper proposes a voltage and current energy information fusion method for AC arc fault detection based on the traditional current detection method by integrating the use of voltage information. Based on the analysis of the respective fault characteristics of switching power supply and non-switching power supply loads, a direct determination method of Arc fault of switching power supply using the total voltage halfwave energy is proposed, and the fault line is selected based on the correlation of voltage and current characteristic energy waveform. A fault detection method suitable for non-switching power supply loads is proposed based on the phase matching of maximum instantaneous characteristic energy of voltage and current in the sensitive domain. The criterion is constructed with characteristic energy phase information, and the difficulty of threshold setting in traditional detection methods is overcome. Although the detection method in this paper uses the idea of load classification, it is not necessary to identify the type of load in practical application because the fault detection under switching power supply loads can be directly determined by using the total energy amplitude of the voltage half-wave to determine the fault directly. Compared to traditional detection methods based on current characteristics, this method has the advantages of simple criteria and easy threshold setting. Experimental results confirm the effectiveness of the proposed method in detecting arc faults in various load types, with detection time meeting industry standards.

Abstract:A new method for transformer fault diagnosis has been proposed to address the issue of low diagnostic accuracy. This approach involves the use of a multi-strategy improved slime mould algorithm (ISMA) for phase optimization in conjunction with a hybrid kernel support vector machine (HSVM). Firstly, principal component analysis (PCA) is employed to eliminate information redundancy among variables and reduce the dimensionality of the dataset. Secondly, the slime mould algorithm (SMA) is introduced, and a Logistic chaotic mapping, quadratic interpolation, and adaptive weight multi-strategy improved SMA are proposed to enhance the convergence speed and local search capability of the SMA algorithm. Subsequently, optimization tests are conducted by comparing the improved SMA algorithm with the original SMA, WHO, and GWO algorithms to validate its superiority. Finally, the improved SMA algorithm is utilized in a phased manner for parameter optimization of HSVM, leading to the construction of the ISMA-HSVM transformer fault diagnosis model. After inputting the dimension-reduced feature data into the HSVM model and comparing it with BPPN, ELM, and SVM, the diagnostic accuracy of the HSVM model improved by 5.55%, 8.89%, and 5.55%, respectively. By optimizing the HSVM model using ISMA and comparing it with WHO, GWO, and SMA algorithm optimizations, the accuracy increased by 13.33%, 12.22%, and 5.55%, respectively. Specifically, the diagnostic accuracy of the ISMA-HSVM model reached 93.33%. The experimental results indicate that the proposed model effectively enhances fault diagnosis classification performance and demonstrates a high level of diagnostic accuracy.

Abstract:Aircraft cable networks play a crucial role in electrical, signal, and data transmission functions. During supersonic flight, aircraft cable networks face challenges such as high temperatures, vibrations, current overload, and low pressure, which affect the safety and reliability of the aircraft electrical system. This study designs a multi-sensor monitoring system for cable networks and a cable network health monitoring algorithm based on multi-sensor fusion. The monitoring system achieves functions including data collection, storage, and wireless transmission of voltage, current, temperature, acceleration, and pressure. In the preprocessing stage, the algorithm comprehensively considers the effects of steady-state and transient values such as high temperatures, vibrations, current overload, and low pressure on the health status of cable networks through normalization. For the health status classification part, a multi-layer classification network is designed to classify the cable network states. In both practical experimental datasets and simulated datasets, the multi-layer classification network in this study achieves an average increase in accuracy of 6.4% and a decrease in false alarm rate of 77.2% compared to the SVM classification network. Compared to single-channel monitoring algorithms, the multi-sensor monitoring algorithm in this study significantly improves accuracy. Experimental results validate the effectiveness of the algorithm in cable network health status classification tasks. The results indicate that the multi-sensor monitoring system for cable networks can effectively monitor and identify the health status of aircraft cable networks, providing strong assurance for the operation of aircraft electrical systems.

Abstract:Ultra-broadband harmonic mixers generally use LO-IF duplexers to extract IF signals from the LO circuit. However, when the LO frequency and the IF frequency are similar or overlap, it is difficult to achieve high LO-IF isolation. In this paper, RF-IF duplexer is used to replace the LO-IF duplexer, different from the common harmonic mixer structure, and IF signal is extracted from the RF circuit. A 30~110 GHz fourth harmonic mixer chip is designed and tested. After the four harmonic mixer testing, RF sweep frequency range 30~110 GHz, frequency conversion loss is less than 25 dB at IF frequency 1 GHz. In DC-15 GHz, the isolation between the LO and the IF ports is up to 30 dB. When the local vibration is fixed, the frequency conversion loss of IF sweep frequency DC-7 GHz is less than 28 dB. Therefore, this design can effectively isolate LO and IF signals with similar frequencies, and it is possible to broaden the IF broadband.

Abstract:The current cross-modal person re-identification research focuses on extracting modality-shared features from global features or local features via identity labels to reduce modality differences, but ignores the Subtle features of discernment. This paper proposes a feature enhanced clustering learning (FECL) network. The network mines and enhances the subtle features of different modalities through global and local features, and combines a multilevel joint clustering learning strategy to minimize the modal differences and intraclass variation. In addition, this paper also designs a random color transition module for training data, which increases the interaction between modalities at the image input to overcome the influence of color deviation. The experiments on public datasets verify the effectiveness of the proposed methods. In the Allsearch mode of SYSU-MM01 dataset, the Rank-1 and mAP reach 70.52% and 64.02%. In the V2I retrieval mode of RegDB dataset, the Rank-1 and mAP reach 88.88% and 80.93%.

Abstract:In recent years, video based human abnormal behavior recognition algorithms have achieved certain research results. However, due to the large amount of data stored in surveillance videos and the long time span of videos, existing recognition methods are not suitable for detecting and recognizing abnormal actions in long videos or multiple pedestrians. To this end, a human abnormal behavior recognition model based on keyframe localization is proposed. Firstly, a keyframe localization network based on standardized flow and attention enhanced spatial temporal graph convolution is used to learn the probability distribution of normal frames, filter and extract sequences of abnormal frames (keyframes) in long videos, and use them as inputs for subsequent network models. Then, in order to better capture the motion characteristics and abnormal situations of human posture, a spatial temporal graph convolutional abnormal behavior recognition algorithm that integrates attention and enhanced residuals is proposed. The keyframe sequence is input into the model network to achieve efficient and accurate recognition of human abnormal behavior in surveillance videos. Validate the effectiveness of this method using publicly available and self built datasets. The experimental results show that the TOP-1 accuracy of human abnormal behavior recognition on the publicly available dataset ShanghaiTech Campus reaches 82.86%, and the TOP-5 accuracy reaches 98.10%. This method can better complete the recognition of human abnormal behavior in surveillance videos.

Abstract:To address the issue of decreased attitude estimation accuracy caused by environmental interference and sensor noise affecting the attitude and heading reference system (AHRS), a noise data processing method based on variable structure error state Kalman filtering (VS-ESKF) is proposed. The text describes a method for detecting noise data in accelerometers and gyroscopes by analyzing the statistical characteristics of sensor observation data and innovation sequence in AHRS. The method is based on the acceleration norm and forgotten sequential probability ratio test (F-SPRT). Secondly, the smooth variable structure filtering (SVSF) strategy is introduced into the error state Kalman filtering (ESKF) to improve its processing capability on the uncertainty of the noise model, based on the noise detection results. The magnetic disturbances are evaluated and magnetometer compensation weights are adjusted in real-time using the Mahalanobis distance method to obtain accurate AHRS correction data by combining the magnetic field strength and magnetic inclination parameter characteristics. The designed VS-ESKF algorithm can detect AHRS noise data timely and accurately, and effectively suppress noise interference, as demonstrated by experimental validation based on a self-unicycle robot platform. Compared to the ESKF algorithm, the accuracy of estimating the roll angle, pitch angle, and yaw angle has increased by 31.05%, 32.32%, and 40.07%, respectively. This improvement enhances the accuracy and stability of attitude estimation.

Abstract:With the continuous advancement of chip technology and the increasing frequency of chip design, delay faults have become an important factor leading to the failure of high-speed chips. In the post-silicon validation stage, due to the lack of a method for measuring the global path delay of chips, the traditional method of constructing delay measurement circuits can only obtain the delay variation of specific critical paths, and comprehensive path delay analysis cannot be conducted when the chip fails. This paper proposes a frequency sweeping at-speed testing method based on scan chains to measure the delay of a large number of timing paths inside the chip and obtain the timing margin. Addressing the issues of long test vector generation time and reliance on specialized testing equipment, frequency sweeping at-speed testing was successfully implemented on a self-developed hardware platform through the generation of multi-frequency test vectors and an improved data verification algorithm. The measurement error of the chip’s path delay is around 8 ps. Experimental verification on different chips at different temperatures confirmed the effectiveness of this method in characterizing path delay, providing a fast and effective method for future research on environmental adaptability analysis and lifetime prediction of high-speed chips through delay parameters.

Abstract:To address the problems of fingerprint information redundancy, difficulty in spatial boundary division, and lack of accuracy in acquiring RP sets in indoor WiFi localization, we propose an indoor localization method based on dual-metric coordination of signal fingerprint measurement. The low-dimensional fingerprint information is simplified and formed through the fusion of fingerprint matrix under S-metric and European-metric. The correlation degree between fingerprints is considered through the “point-class” correlation degree and “class-class” similarity, taking into account the controllability of the number of new fingerprints on the subregion boundary, and the adjustment mechanism of the fuzzy depth of the subregion boundary is established to form the boundary ambiguity generalization ability. Expansion of the fingerprint database is accomplished by the interpolation method of regional sparsity determination, so as to construct a high-density offline fingerprint database. In the preferred subregion, combining the signal space and the location space, the difference degree of the two kinds of measurements is compared to realize the targeted screening of high-value fingerprint points, and reduce the error influence of online fingerprint matching set. In the global experimental scene, the partition results are regular and orderly, which accords with the actual space structure. The construction effect of fingerprint database is improved by at least 11% compared with other schemes, and the positioning accuracy is improved by more than 12% compared with the same type of algorithm. The proposed scheme has significant positioning accuracy advantages, and has better scene adaptability in complex indoor environments with high disturbance characteristics.

Abstract:Ultra-wideband radar has the advantages of high resolution, strong penetration ability, low power consumption, etc. The human body does not need to contact any electrodes or sensors when the ultra-wideband radar is in operation, and it can penetrate through non-metallic media such as clothes and ruins, and detect the information of human vital signs at a long distance. It has an important application value in non-contact vital signs detection. Since the human heartbeat signal is easily interfered by respiratory harmonics and other noises, in order to accurately extract the human vital signs, vital signs signal denoising method based on the combination of improved adaptive noise ensemble empirical modal decomposition (ICEEMDAN) and wavelet packet decomposition (WPD) is proposed. Firstly, we measure the vital signs of the person to be measured by ultra-wideband radar, obtain the spatial location of the human body to extract the micro-motion signals from the body surface, and perform compensation and under-sampling processing for the vibration signals of the body surface; we use the threshold denoising method of ICEEMDAN-WPD to carry out modal decomposition of the micromotion signals, select appropriate modal components for denoising and reconstruction, and obtain the time-frequency characteristics of the micro-motion signals of the human heartbeat. The experimental results show that the algorithm improves the correlation coefficient to 0.9405 and the signal-to-noise ratio to 9.0938 dB compared with the traditional denoising algorithm, which retains more vital signs information and has higher signal-to-noise ratio, and it can be effectively used in the field of vital signs detection.

Abstract:The localization problem of wireless sensor network can be transformed into a fitness function optimization problem, which is solved by the classical sparrow search algorithm. However, the fitness function used in this algorithm does not use measured distance data between unknown nodes, resulting in limited improvement in positioning accuracy. To address this issue, a cooperative localization algorithm based on sparrow search is proposed. This algorithm mainly includes two search stages: rough search and fine search. In the rough search stage, the measured distance data between the unknown node and the anchor node is used to determine the initial position of the unknown node. In the fine search stage, the measured distance data between unknown nodes is used to determine the precise position of the unknown node. Firstly, the Cat chaotic mapping method is used to ensure the uniform distribution of the initial population, which helps to determine the optimal location. Secondly, two different fitness functions are constructed, one for rough search and the other for fine search. Among them, the fitness function used for fine search utilizes the measured distance data between unknown nodes to improve positioning accuracy. Finally, a new fine search method is proposed to avoid the convergence of cooperative localization results to the local optimal position. The effectiveness of the proposed method is verified through analysis of simulation and measured data.

Abstract:Video action recognition has been widely used in many fields such as video surveillance and automatic driving. SlowFast network is often used in the field of video action recognition. At present, attention is used to enhance relevant information in SlowFast correlation network. The combination of attention mechanism and network is to embed the attention mechanism among various convolutional blocks of the network. If the attention mechanism is deeply embedded into the specific convolutional layer of the convolutional block, the information extraction capability of the SlowFast network will be further enhanced. Firstly, a deep nested attention mechanism is proposed, which contains an attention SCTM that can extract space-time and channel information, and further strengthens the three information extraction capabilities of SlowFast network. In addition, the current multi-stream network fusion information is not fully interactive and processed. A multi-stream spatio-temporal information fusion network based on cross-attention and ConvLSTM is proposed to make the information of each stream in the multi-stream network fully interact. The improved SlowFast network has achieved 98.5% Top-1 accuracy on UCF101 and 80.1% accuracy on HMDB51. Compared with the original SlowFast network, the SlowFast spatiotemporal information fusion network with deeply nested attention has superior performance in information extraction and fusion.

Abstract:Accurate SoC is an important guarantee for the safe and efficient operation of lithium batteries. Aiming at the problem that the traditional unscented Kalman filter (UKF) has poor tracking ability for the abrupt state of nonlinear systems, which in turn leads to the low accuracy of SoC estimation, a new adaptive fading unscented Kalman filter was proposed for SoC estimation in this paper. First, the UKF error covariance matrix is weighted by designing a novel fading factor, and the design of the AFUKF is completed based on the novel fading factor, which reduces the influence of stale measurements on the estimation results, improves the estimation accuracy and tracking ability of the traditional UKF. Second, based on the test data of the self-built experimental platform, it is verified that the AFUKF proposed in this paper, in the presence of the initial error, compared with the traditional UKF, the mean absolute error (MAE) and root-mean-square error (RMSE) under the ECE condition are decreased by 47.95% and 33.92%, respectively, the MAE and RMSE under the DST condition are decreased by 36.40% and 27.73%, respectively. Compared with the similar improved AUKF, the MAE and EMSE decreased by 43.36% and 33.51% for the ECE condition, 39.01% and 25.63% for the DST condition, respectively. The modeling results show that, AFUKF has higher accuracy and better robustness under initial SoC errors than the traditional UKF as well as the improved AUKF of the same type.

Abstract:Aiming at the localisation drift problem of pure laser SLAM algorithm, a coarse matching loopback detection algorithm based on the global search of point cloud feature descriptor is proposed. The algorithm firstly adopts the fast segmentation method based on image distance to remove ground points from the laser point cloud, implements edge feature extraction and clustering based on the point cloud curvature and key point aggregation algorithm, and obtains the feature descriptor of the point cloud in the current frame through the feature descriptor generation algorithm, then completes the global matching search by calculating the similarity scores of the current frame and the historical frames to achieve the selection of candidate looping frames, and completes the loopback detection. The coarse matching process is completed. Then the NICP algorithm is used to accurately match the current frame with the candidate loopback frame to complete the loopback detection process. Finally, the real vehicle platform of the mobile robot is built to complete the acquisition of the campus dataset to verify the positioning effect of this paper’s algorithm, and through the analysis of the results of the experiments on the real vehicle, it can be seen that the average value of the degree of optimisation of the error on the campus dataset acquired by the real vehicle is 13.15%. In order to further validate the overall performance of this paper’s algorithm, test comparisons are performed on the KITTI dataset, and the results show that compared with the Lego_loam and Lio-sam algorithms, the algorithm proposed in this paper effectively improves the localisation accuracy on the basis of guaranteeing the operational efficiency.

Abstract:To improve the reliability and machining quality of deburring process for inner holes of projectile components, an evaluation index for the quality of inner hole abrasive flow machining (AFM) is established based on the inverted fillet radius, material removal rate, and surface roughness. An orthogonal test was designed and completed, with abrasive particle size, abrasive mass fraction, abrasive flow working pressure, and cycle numbers as the control factors. Results indicate that all process parameters can ensure good deburring effect. The working pressure of AFM is the primary and significant influencing factor on the inverted filler radius. The primary influencing factor on the material removal rate is the abrasive mass fraction. And for surface roughness, it is cycle numbers. Furthermore, A prediction model for process parameters and machining quality evaluation indexes was established through nonlinear regression analysis. Multi-objective optimization was performed using the fast elitist non-dominated sorting genetic algorithm (NSGA-II). The optimized parameters (The abrasive grain size is 1 000#, the mass fraction of abrasive is 40%, the pressure is 7.5 MPa, and cycles times is 30) were verified and the optimized results are consistent with the actual machining.

Abstract:Focusing on the problems of machine learning in traffic anomaly detection, including reliance on expert experience for feature selection, insufficient expression ability of raw features, poor robustness of models due to noise and outliers in data, and low detection rates for minority classes in imbalanced high-dimensional datasets, an improved LightGBM for Traffic Anomaly Detection Method with Feature Enhancement is proposed. Firstly, the isolation forest (iForest) method is utilized to handle outliers, and the data processed by outlier treatment is used to train an one-dimensional convolutional denoising auto-encoder (CDAE) with global average pooling (GAP), which indirectly eliminates noise in the data and obtains low-dimensional enhanced expressions of original features. Then, adaptive synthetic sampling (ADASYN) is applied to the data after outlier treatment for data augmentation, and the trained CDAE is used to extract features. The obtained low-dimensional features are used as input for LightGBM, which is trained and optimized with Bayesian parameter tuning. At last, the precision classification of anomalous traffic is achieved through the utilization of the obtained CDAE+LightGBM ensemble model. The proposed method attains accuracy rates of 87.80% and F1 scores of 87.75% in a five-class classification task on the NSL-KDD dataset. Experimental results demonstrate that the proposed approach significantly enhances detection accuracy and reinforces the capability to identify unknown attacks. The test on CICIDS2017 scene data set further verifies the feasibility of the proposed method, which superior to the same type of deep learning algorithm.

Abstract:Due to the complexity and randomness of the environment in the production process of ceramic tiles, it is very difficult to construct large-scale and high-quality ceramic tile surface defect data samples, and the insufficient distinguishable feature information under few-shot conditions has a great impact on the accuracy of ceramic tile surface defect detection. To solve this problem, a tile surface defect detection method based on deep transfer learning and improved two-stage ThunderNet network is explored. Firstly, a tile surface defect detection model based on the improved ThunderNet network is proposed, and the structure and functional characteristics of the model are elaborated. Secondly, the decision-making mechanism for spatial parameter transfer of tile surface defect depth features is constructed to effectively improve the characterization ability of sample feature. Third, the ShuffleNet backbone network is optimized based on Switchable Atrous Convolution (SAC) to enhance the model’s learning ability to the changeable shape of defects. Fourth, a feature fusion algorithm based on multi-scale mapping and squeeze and excitation (SE) is proposed to realize the multi-level differentiated characterization of tile surface defect feature information in a limited feature level. Finally, a tile surface defect detection algorithm for fusion deep transfer learning and improved ThunderNet network is given. The experimental data show that on the same tile surface defect test dataset, the proposed method has superior performance for the detection of tile surface defects under few-shot conditions, and the average accuracy, average recall and average detection speed of the model reach 87.22%, 93.69% and 61.6 ms/img, respectively, compared with the traditional ThunderNet model, the average accuracy and average recall are improved by 9.30% and 4.16%, respectively, among which, based on the SAC optimal atrous ratio combination {1,2}, The model accuracy is improved by 5.51%, the model accuracy is improved by 6.16% based on the optimal compression ratio of SE 24, and the model accuracy is improved by 3.86% based on the transfer mechanism in this paper, and the network convergence is accelerated. Compared with the traditional ThunderNet network and other mainstream detection models, the proposed method improves the expression ability of few-shot object features through knowledge sharing of transfer mechanism, and realizes hierarchical representation of object features by introducing SAC and SE under the premise of controlling the scale of the model, which effectively improves the real-time reliability and reliability of the model.

Abstract:At present, the automatic test equipment on the production line can only perform the traditional static test of on, off, and short circuit of the high-speed serial interface, and it is difficult to complete the dynamic test of the time domain characteristics of the eye diagram in a limited production cycle time. In order to complete the dynamic test at low cost and quickly on the production line, a two-dimensional bit error rate statistical circuit with multiple phases and multiple decision levels was developed, and the time domain eye diagram was constructed by the method of equal bit error rate interpolation. Firstly, the different source characteristics of jitter are analyzed, and the relationship between jitter and BER, the relationship between Q Scale and BER P is obtained according to the Dual-Dirac model, and the tangent slope of the Q-P curve converges to the reciprocal of the random jitter variance σ. Then, by measuring the tangent slope to solve the σ, the model of Gaussian random jitter is obtained. Finally, according to the Gaussian model the BER curves at all decision levels are interpolated at a giver BER, the equal points on the two-dimensional BER map are found, and these points are connected to obtain the inner contour line of the eye diagram. The calculation and analysis of Gaussian function, eye diagram simulation data and experimental data show that the Q-P curve slope interpolation method can rebuild the inner contour line of the eye diagram in the area below the BER less than 10-4, and can quantitatively evaluate the basic eye diagram time domain parameters such as eye height, eye width, amplitude, rise and fall time, and the accuracy error of the relevant data is about 10%.

Abstract:Aiming at the problems such as low efficiency and inaccurate results in powder pile inventory, a quality measurement method based on modified cam clay model is studied, the feasibility of this method is verified by the quality measurement experiment of small quartz sand. In order to improve the efficiency of material quality inventory and the accuracy of measurement results, a stress-density model is constructed based on the integral principle and the stress-strain relationship described by the modified cam clay model, based on the thin-film pressure sensor, a density measuring device is designed, and two different kinds of stacking materials are designed according to the common stacking materials in the storage yard, the density can be obtained by the pressure at different height and radial depth, and the volume can be obtained by the point cloud data of three-dimensional laser scanner. The experimental results show that the relative error of the results obtained by this method is below 4% for quartz sand piles with masses of 92.1 kg and 86.1 kg. The experimental results prove the feasibility of the measurement method, which can provide reference for the inventory of large-scale stack and help enterprises to make further production planning.

Abstract:Crimped conductor is primarily used in constructing overhead lines, and its edge distance is crucial for ensuring safe operation and a long service life of the line. To address the limitations of poor reliability and low efficiency in traditional manual measurement, a novel method based on image processing was proposed. Firstly, the measured area was automatically cropped based on the cumulative grayscale mutation points of the image and then median filtered. Subsequently, a binary image with continuous edges was obtained based on the improved Otsu thresholding method. Then, the edges were located twice using morphological processing and the Gaussian curve fitting method, respectively, and sub-pixel level edge points were obtained. Finally, the upper and lower edge points were fitted using the least-squares fitting method based on RANSAC to achieve the measurement of crimped edge distance. The simulation test confirmed that the measurement algorithm produces better results after 10 iterations of RANSAC fitting, with a measurement error of less than 0.1 pixels. The actual wire measurement test results demonstrated that the improved Otsu algorithm is adaptable to changes in light intensity. Furthermore, when compared to manual measurement, the image measurement method has a maximum relative deviation of only 1.82%, a 60% increase in repeatable standard deviation, and consumes only 1/10 of the time. The efficient and reliable measurement of crimping edge distance is realized.