Abstract:Vector control is the current mainstream control method for permanent magnet synchronous motors (PMSM). Aiming at its disadvantages such as complex calculations and dependence on motor parameters identification, a formal method of signal temporal logic (STL) is proposed to identify the running states of the motor, so that the maximum torque per ampere (MTPA) can be realized by controlling the pulse width modulation (PWM) of the motor drive circuit. The voltage data of the shunt resistance in series with the DC bus of the driving circuit under the same working condition and different running states of the motor were collected. The 5-fold cross validation is adopted, and the STL formula is learned based on the decision tree. Finally, the STL formula is used to determine whether the motor is running normal, under modulation or over modulation. The first-level and second-level primitives are defined as the nodes of the decision tree respectively. Particle swarm optimization (PSO) is used in the learning process, and different impurity measures are used as the loss functions. The experimental results show that the accuracy of motor states recognition by STL with first-level primitives can reach 98.78%, and the program takes 0.1509s. The recognition accuracy with second-level primitives can reach 95.06% and the program takes 2.3979s. It is of great significance to the implementation of motor control algorithms based on STL.

Abstract:The health assessment of power equipment can reduce the blindness of equipment maintenance. Aiming at the problem of high ambiguity of state division near the boundary point of adjacent states in current circuit breaker state evaluation, a method for evaluating the state of circuit breaker with extension gray cloud clustering considering the difference of state interval is designed in this paper. This method uses extension cloud theory and gray cloud clustering function as the basic theoretical framework, adopts different constraint ranges to construct the optimal cloud entropy, and simultaneously considers the difference in the state interval range. Taking a vacuum circuit breaker in a substation as an example, in the early warning state, the cloud entropy constructed by the "3En" rule is 0.6712, and the cloud entropy constructed by the "50% correlation" rule is 0.5776. The optimal cloud entropy constructed in this paper is 0.6127. This method will The overall judgment is an early warning state, which is consistent with the results of other evaluation methods, but its reflection on the degree of membership is more accurate. The analysis of the example results shows that it can effectively deal with the uncertain factors in the evaluation of the circuit breaker's status, and has high feasibility. At the same time, the accuracy of the evaluation results has been improved to a certain extent.

Abstract:In view of the large and complex space-based network constellation composed of hundreds, thousands, or even tens of thousands of satellites, the ground-based measurement and control network alone cannot meet the problem of real-time satellite management and control, and the long-term deviation from the ground reference caused by the independent measurement and control of the constellation network alone ( Time and space benchmarks) and other risks, this paper comprehensively considers the advantages of ground-based measurement and control network and inter-satellite networking, and proposes a parallel system method for autonomous operation of large-scale networking constellations. Through the construction of a space-earth integrated parallel system, under the action of the ACP framework, a space-earth integrated parallel operation mechanism is formed with the network constellation autonomous processing as the main and the ground parallel system support as a supplement, so as to realize the large-scale network constellation autonomous control, autonomous observation, and autonomous processing And autonomous monitoring and other autonomous operation functions. The research results show that the method proposed in this paper has reference value in the autonomous operation of large and complex constellations.

Abstract:Ferromagnetic metal foreign Bodies are one of the common interference objects in the actual operation of wireless power transmission systems. It indirectly affects the output characteristics of the wireless power transmission system by affecting the electrical parameters of the magnetic coupling mechanism. DD/BP structure is the most common type of magnetic coupling mechanism for electric vehicle wireless power transmission system. This paper takes DD/BP magnetic coupling mechanism as an example to analyze the influence of ferromagnetic metal foreign bodies on the coils' self-inductance and mutual inductance, and introduces the influence of ferromagnetic metal foreign bodies in different positions on the magnetic coupling mechanism of DD/BP based on the analysis of magnetic theory. An electromagnetic simulation model of the DD/BP magnetic coupling mechanism is built to simulate the electrical influence by using Maxwell. Ferromagnetic metal foreign Bodies are set in different horizontal or vertical position of the model to verify the correctness of the theoretical derivation. The simulation results show that when the ferromagnetic metal foreign body is on both sides of the DD/BP coil, it will increase the coil self-inductance and the mutual inductance between the coils, while at the junction of the two D coils, it will increase the coil self-inductance and reduce the mutual inductance between the coils. In the vertical direction, the closer the ferromagnetic metal foreign body is to the coil, the greater the influence on the self-inductance of the corresponding coil, and the closer to any coil the greater the influence on the mutual inductance between the coils.

Abstract:In order to realize the intelligent control of the micro knob, the key task is to obtain the accurate position and pose of the knob. Based on this, a new method of knob pose measurement combining edge detection and deep network is proposed. Firstly, to solve the problem of measurement error caused by image tilt, an improved Canny algorithm was proposed to extract the accurate edge and correct it by combining perspective transform. Secondly, the improved YOLO-V4 algorithm is used to achieve the precise segmentation of the knob. Finally, the improved Canny algorithm and bicubic spline interpolation were used to extract the high-precision knob groove sub-pixel contour, and the PCA algorithm was used to fit the contour rectangle and measure the position and pose. Experimental results show that the proposed improved Canny algorithm improves the accuracy of edge extraction and effectively reduces false edges. Compared with the YOLO-V4 knob detection algorithm, the average detection accuracy of the improved YOLO-V4 knob detection algorithm is improved by 2.92%, reaching 99.49%. The measurement accuracy of the center position and deflection angle of the knob grooves reaches 1.5pixel and 1.5° respectively, which can realize the high-precision measurement of the knob position and pose.

Abstract:With the continuous improvement of the industrialization level, the measurement of large-size parts based on machine vision has been hot in research. Firstly, this paper explains the research background of machine vision measurement technology and the current research status at home and abroad, points out the difficulties of current vision measurement research, advocates to improve measurement accuracy and efficiency by studying image processing algorithms. Secondly, this paper researches the edge detection technology, which mainly uses the coarse and fine positioning edge detection algorithm in vision measurement. Sub-pixel edge detection algorithms in precise edge positioning are highlighted for analysis. Then, the image stitching techniques used in the measurement of large size parts are investigated and analyzed. The image registration applied by this technology is mainly based on two types of methods: region and feature. This paper also analyzes the advantages and disadvantages of these two types of methods. Finally, the characteristics and limitations of the measurement methods for large size parts are summarized, and future further exploring directions for improvement are pointed out.

Abstract:In order to improve the accuracy and reliability of the transformer fault diagnosis, a fault diagnosis method of transformer based on MPC(modification of the PC, for short: MPC) algorithm optimized by Bayesian network was proposed, and the fault diagnosis technology of transformer was studied. Firstly, according to the analysis of dissolved gas in oil, the 9-D fault features of oil-immersed transformer were extracted by the non-coding ratio method, and the data samples were normalized. Secondly, a fault diagnosis model based on Bayesian network was established with normalized training samples as input, and the Bayesian network model was optimized with the MPC algorithm. Finally, the fault diagnosis model was tested with test samples. In order to prove the superiority of the proposed method, the proposed method was compared with the existing fault diagnosis methods. The result shows that the proposed method has higher diagnostic accuracy and better diagnostic effect.

Abstract:In order to solve the over-segmentation and incorrect segmentation in tree detection of aerial image under complex terrain background, a watershed segmentation algorithm based on optimal marker was proposed. In this method, firstly, the contrast between the target and the background is improved by nonlinear gray transformation. Secondly, the foreground and background areas are initial marked according to the shape characteristics of the target, then the distance transformation in the traditional watershed algorithm is replaced by the centroid of the foreground area, and the foreground is marked again. Finally, the image is segmented by the watershed algorithm. The experimental results show that the average segmentation accuracy of this algorithm is 92.5%, which is more anti-noise and accurate than the existing image segmentation methods, and also solves the over-segmentation problem of the watershed algorithm.

Abstract:In order to reduce the number of model coefficients while ensuring the accuracy of power amplifier modeling, this paper creatively proposes a simplified power amplifier model based on compressed sensing algorithm, which uses remove duplicates sparse adaptive matching pursuit (RDSAMP) algorithm to simplify the model coefficients. Compared with the generalized memory polynomial (GMP) and the piecewise simplified dynamic deviation reduction (PSDDR) full coefficient model, the digital pre-distortion test results show that the modeling accuracy of the simplified model proposed in this paper is improved by 1dB and-0.6dB to-46.01dB, and the adjacent channel power ratio (ACPR) is increased by 3.2dB and-1dB to-50dBc, respectively. At the same time, the number of model coefficients is greatly reduced by 72% and 65%, respectively. Therefore, the model proposed in this paper can greatly reduce the model coefficient on the basis of maintaining the modeling accuracy, which has a high reference value for the existing power amplifier model simplification problem.

Abstract:For twisted pair in the process of signal transmission with high speed over long distances, hundred meters above the signal attenuation loss will aggravate the situation, this paper designed the hundred meters cable parameter model, RLCG parameter model as the research object, and on the basis of its estimated attenuation characteristic of cable, the results of two common types cable test data under the same experimental conditions, Compared with the simulation results of the model, it is found that the error between the two is less than 0.5dB in the frequency range of 10-200 MHz, which indicates that the cable loss characteristics estimated by the parameter model are basically consistent with the actual test results within a certain frequency range. This model provides a theoretical basis for the study of twisted-pair fast and long distance reliable transmission of LVDS (low voltage differential signaling).

Abstract:With the development of industrial production to intelligence, unmanned has become a trend.In the operation of mechanical equipment, the rolling bearing is easy to be damaged. Remote monitoring of the operation of mechanical equipment has become an urgent problem to be solved.This paper designs a remote monitoring system for the running state of rolling bearing based on Lora base station. The system takes STM32 as the core and forms a collection sub node to collect the bearing vibration signal, and then sends the collected data to Lora wireless gateway through SX1278 RF chip.Lora wireless gateway sends the received data to the remote service platform through GPRS module. The remote service platform uses the complementary ensemble empirical mode decomposition method to analyze the vibration signal to determine the working state of the bearing. Experiments show that the system can effectively monitor mechanical equipment in real time to ensure long-term stable operation and safe and effective production.

Abstract:Aiming at the problem of poor detection accuracy of OFDM signals caused by channel time-varying and non-stationary under dual-selection fading characteristics, a signal detection scheme based on CNN-GRU (Convolutional Neural Network-Gated Recurrent Unit Neural Networks, CGNN) is proposed. First, use the channel model to generate data to fully mine the prior knowledge of the channel; then use the one-dimensional convolutional neural network in offline training to reduce the dimensionality and feature extraction of the original signal, and use the memory characteristics of the gated loop unit to restore the fading signal; finally In order to reduce the interference caused by the severely fading sub-carriers, an attention mechanism is added to the network training, and weights are assigned to each sub-carrier, so as to perform differentiated training. The simulation results show that the error performance of the detection method proposed in this paper is significantly improved. In a flat fading channel, CGNN can obtain an error performance gain of 0.3dB~1dB. In a frequency selective fading channel, CGNN can obtain an error performance gain of 2dB~5dB, and it has strong robustness.

Abstract:Studies have shown that in lower limb movement category, imagine eeg signals based on the motion to imagine joining steady-state somatosensory evoked potentials can get higher classification results, but most studies are based on double side auxiliary body feeling potential induced by electrical stimulation, the unilateral electrical stimulation auxiliary study is less, in this paper, the unilateral posterior tibial nerves and bilateral ankle ankle posterior tibial nerve stimulation, respectively To find out what kind of stimulus can get better classification results. Two control experiments were designed: unilateral left foot electrical stimulation mode vs unilateral right foot electrical stimulation mode and unilateral right foot electrical stimulation mode vs bilateral simultaneous electrical stimulation mode. Based on the analysis of the characteristics of spectrum, time-frequency map and brain topography, it was concluded that the ERD characteristics of the unilateral right foot electrical stimulation mode were the most significant and the activation degree was the deepest. The average classification accuracy of unilateral right foot stimulation mode was 5.57% higher than that of dual stimulation mode, and the classification accuracy of single subject's unilateral right foot stimulation mode was 15% higher than that of dual stimulation mode, which proved that the assistance of unilateral right foot electrical stimulation is more conducive to the classification of lower limb motor imagination EEG signals.

Abstract:Miniature attitude and heading reference system (AHRS) incorporates tri-axial magnetometer, accelerometer, and gyroscope to achieve three-dimensional attitude measurement. It has be widely used in flight control of unmanned aerial vehicle and other related fields for. However, AHRS uses gravity vector to calculate pitch and roll angles, and thus will have errors caused by motion acceleration. Existing algorithms for motion acceleration elimination essentially rely on gyroscope, and may cause accumulative errors when affected by persistent motion acceleration. A novel algorithm is proposed to improve the dynamic performance of extended Kalman filter, which utilizes optical flow sensor to measure and compensate motion acceleration, while estimates gravity and geomagnetic vectors in parallel. Experiment shows that with the impact of horizontal acceleration, attitude error of the proposed algorithm can be 50% lower than existing methods, and thus it can significantly enhance dynamic attitude accuracy.

Abstract:Aiming at the problems in service environment, damage situation of oil and gas pipelines in desert burial ground and the invasion situation of third party which threatens the safety of pipelines, which easily causes the difficulties in effective feature extraction and accurate classification identification of invasion vibration signals, a feature identification method for invasion signals from oil and gas pipelines in desert burial ground is proposed. The method first uses distributed optical fiber to acquire the intrusion vibration signal along the pipeline, and then decomposes the vibration signal by the modified ensemble empirical mode decomposition (MEEMD) method to obtain the intrinsic mode function (IMF) component of the signal; then extracts the energy of the IMF component and the MEEMD energy entropy to form a feature vector; finally, the feature vector is input to the extreme learning machine (ELM) classification recognition model. The experimental results show that the method can achieve the recognition of four types by tapping the pipe, manual mining, mechanical construction and sandstorm weather, and compared with BP neural network and support vector machine recognition models, the total recognition accuracy of the method reaches 94%, and the recognition speed is faster. The proposed method has important reference significance for distributed fiber optic desert buried oil and gas pipeline monitoring.

Abstract:When there are multiple partial discharge sources in electrical equipment, mixed partial discharge signals will be generated, which will bring difficulties to subsequent signals identification and other work. Aiming at this problem, this paper to electrical equipment inside at the same time there are three different insulation defects, for example, each with three kinds of signal model with exponential and three kinds of UHF bureau put mathematical model is constructed in the two groups mixed signals, partial discharge to simulate different mixed signals generated by electrical equipment, and put forward using a JADE based on blind source separation algorithm to separate the signal, and then use the similarity coefficient and signal interference ratio of two kinds of evaluation parameters to describe the separation performance of the algorithm, finally add appropriate noise in the signal, test the robustness of the algorithm. The simulation results show that the method can effectively separate the signals, the similarity coefficient of the separated signal and the source signal is above 0.9, and the signal interference ratio is above 9.0, and it has a certain robustness, which lays a certain foundation for the subsequent signal identification work.

Abstract:For the feature extraction requirements of complex and non-stationary vibration sensor signal, this paper researches Hermite interpolation based on triangular Shepard(TSHI) and LMD method based on TSHI for vibration sensor signal feature extraction. TSHI combines binary Hermite interpolation functions with osculatory triangular Shepard basis functions. Then it creates the envelope curve interpolation polynomial of complex and scattered vibration sensor signal. It can adjust the local interpolation curve according to the time distance between the interpolation point and every vertex of the triangle, so that the envelope estimation curve is more suitable. The LMD method based on TSHI decomposes the vibration sensor signal into several Product Functions(PF) which has time-frequency characteristic scale of the signal. Then the method combines the energy of the main PFs to form the signal characteristic vector. The experimental results show that TSHI for complex and high frequency vibration sensor signal’s envelope curve can avoid phase difference, over envelope and under envelope. The RMSE of the TSHI result is smaller. If Relevance Vector Machine(RVM) fault diagnosis model applies the LMD method based on TSHI, its diagnostic accuracy of vibration sensor is close to 100%.

Abstract:Ultrasonic ranging is the most common ranging technology with high cost performance. However, the ranging error is often caused by signal shaping in common ultrasonic echo ranging mainly due to the loss of starting period and the difference of reflected signal amplitude, which makes the measured signal flight time exceed the starting time of received signal. The dual phase detection method proposed in this paper transmits a group of critical modulation signals of 64 cycle carrier and 2 cycle modulation signals, timing starts at the same time, stops timing when the amplitude of the received signal exceeds a specific threshold, then detects the phase of the signal at the threshold point, obtains the starting time point of the received signal, and then uses the carrier phase correction method to eliminate the error caused by signal shaping to obtain the distance measurement result. Through the simulation test and experimental test, the results show that the actual measurement accuracy is almost not affected by the noise and the amplitude of the reflected signal. In the range of 50~200 mm, the standard deviation of the two-phase ranging device is less than 1 mm, and the ranging accuracy is effectively improved.

Abstract:In X-band T/R module, the number, length, arch height, span, solder joint position and other parameters of gold bonding wires will have a serious impact on the microwave transmission characteristics. Through the automatic detection technology to achieve the automatic detection of the above parameters, we can infer whether the bonding quality of X-band T/R module is qualified. In this paper, based on the focus variation measurement technology, the micron scale parameter measurement of the gold bonding wires is realized, and the relative error of the measurement result is less than 0.7%. In this method, a group of images of the gold bonding wires are obtained through the self-designed image acquisition platform, and then the arch height and span of the gold bonding wires are measured through image processing technologies such as multi focus image fusion and focus evaluation. This technology is helpful to improve the detection efficiency of gold wire bonding products and the production efficiency of X-band T/R module.

Abstract:In order to improve the accuracy of the Tiny YOLOV3 target detection algorithm in pedestrian detection tasks, the algorithm is researched and improved. Firstly, deepen the feature extraction network of Tiny YOLOV3 to enhance the feature extraction capabilities of the network. Then, add the channel attention mechanism to the two detection scales of the prediction network, and assign different weights to different channels of the feature map to guide the network to pay more attention the visible area of pedestrians. Finally, the activation function and loss function are improved, and the K-means clustering algorithm is used to reselect the initial candidate frame. Experimental results show that the improved Tiny YOLOV3 algorithm has an average precision(AP) of 77% on the VOC2007 pedestrian subset and 92.7% on the INRIA data set, which is 8.5% and 2.5% higher than Tiny YOLOV3, and the running speed is 92.6 frame per second(FPS) and 31.2 FPS. The improved algorithm improves the accuracy of pedestrian detection, maintains a faster detection speed, and meets real-time operation requirements.

Abstract:The grid-based image speeds up the implementation of the algorithm in the GMS (grid-based Motion Statistics) matching algorithm. However, the feature points at the edge of the Grid are not effectively processed, which leads to the existence of wrong matching pairs. This paper proposes an image mismatching elimination algorithm based on adaptive margin mesh motion statistics. Firstly, the adaptive algorithm is used to calculate the optimal distance of the grid edge, and the feature points of the grid edge are assigned to other adjacent grids, so that these feature points can effectively play a supporting role for the correct matching points and improve the score of the correct matching points. Finally, the statistical characteristics representing the motion smoothing constraint were used to eliminate the wrong matching points in the initial matching. Simulation experiments show that the recall rate of the proposed method is about 10% higher than that of the GMS algorithm, and the real-time performance is also about 30% higher. Compared with the SIFT algorithm, the running time is shortened by 17 times on average. Compared with SURF algorithm, the number of correct matches is increased by 8 times on average, which fully indicates that the wrong matching points can be removed effectively and efficiently, and the image matching quality can be further improved.

Abstract:Gan (Generative Adversarial Network) is applied to the generation of defect samples for power inspection to solve the problem of insufficient defect samples. The current Gan-based technology of insulator defect sample generation has the following limitations: 1) A large number of defect samples are required for training and the number of generations is insufficient; 2) The quality of the generated samples is poor, the size is small, making it difficult to use for target detection neural network model training. To address the above limitations, a style transfer Tw_Cycle (A Target weighted Cycle consistent) Gan is proposed based on Starganv2. The network can use non-defective samples for training, and realize one-to-many defect sample generation based on non-defective samples. In order to ensure the semantics of defects remain unchanged, the Unet segmentation network is added, and the Roi_cyc loss and the Roi_mask loss are used to strengthen the constraint of the insulator target. Through qualitative and quantitative evaluation, Tw_Cycle Gan has achieved better results. In order to verify the validity of the generated samples, an experimental evaluation method for defect detection based on real samples is designed. The results show that the same Yolov3 target detection algorithm that uses synthetic defect samples to amplify training, the AP, increased by about 5% on average, Precision increased by about 4.6% on average, Recall increased by about 10% on average, and F1 increased on average by 0.083.

Abstract:Reference template is used in most methods of PCB defect detection，which is very time consuming and causes a big position error. YOLOv4 is fast but it misses the object easily in PCB detection and its accuracy is not high in detecting the small object. Therefore, the method of PCB defect detection based on improved YOLOv4 algorithm is proposed. Firstly, CSPDarknet53 is used as backbone and the structure of single feature layer is adopted, which avoids the prior boxes assignment problem caused by data imbalance. Then, five convolutions are improved using CSP to increase further the ability offeature extract.Finally, prior boxes are gotten by using K-means++ to improve the training effect. In the experiment, Peking University PCB public dataset is used for training. The result shows that mean average precision of our algorithm achieves 98.71% and it has a better performance compared with other several classical object detection algorithms.

Abstract:Aiming at the problem that the traditional SSD algorithm lacks the exchange of feature information between features at different scales, which is not conducive to target positioning and recognition, so that the detection results of the traditional SSD algorithm often can not meet the actual accuracy requirements, and the detection speed can not meet the actual requirements, so it is improved, and the lightweight network mobilenet-v2 and feature fusion module are introduced, Compared with the traditional SSD algorithm, the detection accuracy (map) has reached 98.32% and the detection speed (FPS) has reached 72 frames / s. It has better application value for practical engineering projects.

Abstract:Facial expressions can correctly reflect people's inner activities, but due to the complexity and subtlety of facial expressions, accurate recognition of facial expressions is still a big problem. This paper designs a method based on convolutional neural network (CNN) and long short term memory (LSTM), so that the computer can recognize the expression of human face The loss function uses focal loss. The framework includes three aspects: (1) two different preprocessing techniques are used to deal with the illumination change and preserve the edge information of the image. (2) The preprocessed image is input into two independent CNN layers for feature extraction. (3) The extracted features are fused with LSTM layer. We use FER2013, JAFFE and CK + data sets to verify the accuracy of the model, and select FER2013 data set to make a mixed matrix. The results show that the accuracy of our model on fer2013 data set is improved by 9.65% compared with the current advanced model, and it also performs well on Jaffe and CK + data sets. The results show that our proposed model has strong generalization ability.

Abstract:Aiming at the prominent "pseudo-change" problem in the detection of changes in heterogeneous high-resolution remote sensing images, this paper proposes an object-level change detection method based on improved bilateral filtering and small targets suppression model. On the grounds of the traditional filtering strategy based on global pixels, this paper designs an improved bilateral filter under the boundary constraint of segmented objects to improve the spatial structure consistency between pixels in the object; Moreover, in order to further weaken the "false change" caused by local outliers, the paper proposes a small target suppression strategy based on high-order neuron on-off channel; Finally, the Otsu method is used to classify the difference information and obtain the final change detection results. The experimental results of multiple groups of heterogeneous high-resolution remote sensing images show that the proposed method can effectively reduce the detection error caused by "pseudo change", the overall accuracy can reach 92.2%, and the false detection rate is less than 8.7%. It is significantly better than the three groups of comparison methods in visual analysis and quantitative evaluation.

Abstract:In recent years, deep learning methods have performed well in human activity recognition. They use time series data obtained by wearable sensors such as gyroscopes and accelerometers to perform training and classification after preprocessing and data-level fusion. This paper proposes a feature-level fusion method of LSTM and CNN in order to solve the problem that the data-level fusion method has certain limitations in the recognition of multiple sensors. This method connects the independent sensor data to the LSTM layer and the convolutional component layer in turn for feature extraction, and then gathers the features of multiple sensors for action classification. The average F1 scores of this method on the three public data sets UCI-HAR, PAMAP2 and OPPORTUNITY is 96.06%, 96.17% and 94.44% respectively. Experimental results show that the method proposed in this paper has better accuracy in multi-sensor recognition of human movements.