Abstract:The existing four-channel bilateral control teleoperating system based on disturbance observer (DOB) can achieve a certain degree of transparency without relying on force sensors. However, the current master/slave controller design is mostly based on the trial and error method, lacks systematic design, and is limited by the DOB operating bandwidth, so transparency cannot be improved. In this work, a systematic method for designing controllers based on robust control theory is proposed. The teleoperated four-channel bilateral control is transformed into two general feedback control systems by obtaining the dynamic equations of the position response error and disturbance residual between the master/slave robot, then systematically integrated controller can be obtained through selecting appropriate weighting functions based on H∞ control theory. The new method presents a general method to realize the bilateral control of master/slave robot under the condition of forceless sensor. Experiments show that the controller designed via the systematic method can ensure the system stability and transparency, and improve the tracking and force transfer between two motors. Besides, this design exhibits stronger robustness to disturbance and model uncertainty in the system compared with the traditional four-channel bilateral control based on PD control design.

Abstract:In order to improve the status quo of inaccurate positioning of high-speed trains during running, this study adopts global navigation satellite system and inertial navigation system to construct GNSS/INS tightly coupled precise positioning system. The equivalent weight principle is proposed to replace the noise covariance in the UKF, so that the Filter gain can be adjusted in the process of information fusion, the influence of observation coarse error can be reduced, and the robustness of UKF Algorithm can be improved. Based on the improved UKF algorithm, this study uses map matching algorithm to reprocess the location data after filter processing, and carries on the simulation comparison. The results show that AVE and RMSE values of the position velocity error curves simulated by the improved UKF algorithm and the standard UKF algorithm are reduced by more than 20%. And the positioning accuracy is higher after the map matching algorithm processing, indicating that this study can improve the positioning accuracy and stability of high-speed trains.

Abstract:Aiming at the requirements of lower limb exoskeleton robot, an adaptive fuzzy PID control system design method based on Verilog HDL hardware description language and FPGA (Field programmable logic gate array) was proposed, which was applied to drive and control the joint of lower limb exoskeleton robot. Firstly, the algorithm model was simulated by MATLAB/Simulink system simulation software to verify the stability of the algorithm and the accuracy of parameters. Then Quartus II was used as the development platform to complete the hierarchical design of Verilog HDL for adaptive fuzzy PID controller, which was realized and verified based on FPGA chip EP4CE10F17C8. The experimental results show that the fuzzy PID control can reach the stable state faster, and the overshoot is reduced by 4.7%, and the control algorithm is realized by FPGA to better meet the control requirements of the lower limb exoskeleton robot.

Abstract:Traditional dance choreography is time-consuming and labor-intensive, causing an expensive and tedious production. In addition, intelligent music-driven dance choreography has a circumstance of lack of data. Therefore, an intelligent music-driven dance choreography display system was designed based on a new framework using directed graph nueral network, which faces to administrators and general users, consisting of managements of scene, network, interaction and data. For normal users, the effectiveness of custom audio file is assesed, and the choreography result is returned based on an directed graph neural network framework, then the system displays dance sequences. For administrators, the functions of resource migration and user management are offered to update available models trained on the growing dataset. While for general users, the upload files are checked then choreography and display. The experimental results show that the system can generate higher quality dance sequences compared to baseline, and recives high user satisfaction, short response time, having a certain application value.

Abstract:In order to improve the real-time performance of the activity recognition algorithm and be suitable for embedded devices with limited resources, a hardware acceleration method of the activity recognition algorithm was proposed and implemented on the FPGA platform. Traditional wearable sensor-based behavior recognition algorithms require strictly labeled data for training and classification, but the labeling process of sensor sequences consumes a lot of manpower and computing resources. To solve this problem, an attention mechanism is introduced into the traditional convolutional neural network model. , for action recognition based on weakly labeled data. Computational modules such as convolution, pooling, and attention mechanisms in the algorithm use a high-level comprehensive design. According to the operation characteristics of the model, the operation speed is improved by pipeline constraints, multi-pixel and multi-channel parallelization, and data fixed-pointization. Experiments are carried out on the Ultra96_V2 platform, and the experimental results show that the designed behavior recognition system has a recognition accuracy of 90% and a computing speed of 25.89frams/s, which is 54.15 times faster than that of a single-core ARM_A53 processor. The average power consumption of the system is 2.204W and the power efficiency is 11.75frams/s, which meets the design requirements of low power consumption and low delay.

Abstract:In order to obtain reasonable parameters for the serpentine liquid-cooled structure of the power battery pack, an optimal design method for the liquid-cooled structure of the power battery pack combining the uniform design method, BP neural network algorithm and multi-objective genetic algorithm is proposed. Firstly, a single cell temperature rise test is carried out to verify the cell simulation calculation model and provide support for the data accuracy of uniform design test and parameter processing. Then it was determined that the temperature difference of the battery pack and the pressure drop of the liquid cooling structure were the design objectives, and the coolant inlet mass flow rate, coolant inlet diameter and liquid cooling tube pipe width were the design parameters. CFD simulation was conducted through uniform design test to obtain the specific parameters of the liquid cooling structure, and the agent model between the design objectives and the design parameters was obtained by training with BP neural network algorithm. Finally, the NGSA-II multi-objective genetic algorithm is used to calculate the proxy model to obtain the Pareto solution set, and the optimal Pareto solution is selected according to the engineering experience to verify the optimization results and compare the simulation results before and after optimization. The comparison results before and after optimization show that: the maximum temperature of the battery pack is reduced by 5.06℃, with a decrease of 14.3%; the maximum temperature difference of the battery pack is reduced by 4.88℃, with a decrease of 51.5% compared with that before optimization; the pressure drop of the liquid cooling structure is increased by 122.8%, which solves the negative pressure problem and reduces the coolant pressure loss, which verifies the effectiveness of the optimization method.

Abstract:In order to solve the inverse kinematics of manipulator with poor real-time performance and low precision, a particle swarm optimization (PSO) algorithm for LSTM was proposed in this paper.Firstly, the model of the series 6-DOF manipulator is established for kinematic analysis, and the training data are obtained. Next, Optimizing the quantity of hidden level neural units and learning percentage of the long-term and short-term memory network by PSO. LSTM after parameter optimization learns the mapping relationship between the position and pose of the manipulator's end effector and joint variables. Finally, the trained PSO-LSTM model is used to predict the joint variables of the manipulator to obtain the inverse kinematics solution. The experimental results show that the inverse kinematics solution speed of the model is kept within 10 ms, which is tens of times higher than that of the traditional solution, and the mean square error of the model is as low as 0.001, which can not only improve the solution speed but also ensure the solution accuracy.

Abstract:Focus on the uncertainty of Medical closed-loop control System caused by differences among patients, a model-based closed-loop adaptive control architecture was proposed based on the concept of MCPS. BIS signal was used as the control variable to control the depth of hypnosis during anesthesia. The pharmacokinetic and pharmacodynamic model of the patient was used in the control scheme, and the calculated results were used as the feedback signal of the standard PID controller to correct the uncertainty of the model. In the model, parameter adjustment is performed offline using genetic algorithm to optimize the performance indicators of the patient data set. The infusion rate of anesthetic drugs can be automatically adjusted to keep the depth of anesthesia at a stable target value. The robustness of the proposed method was tested by adding noise blocks. Monte Carlo method was used to verify the effectiveness of the proposed method on a wide range of people. Simulation results show that the proposed method can reach the target value stably in a specified time and has good interference rejection.

Abstract:In order to improve the measurement efficiency of digital spectrometer, an embedded high-speed data acquisition and processing technology based on FPGA+ARM architecture and two-level cache is investigated and implemented. The FPGA is used to provide a sampling clock for a high-speed A/D converter. The sampled data is then cached by a FIFO so as to realize cross clock domain data transmission. A DDR3 integrated with the ARM is used as a second cache to avoid data jam and loss phenomenon during the high speed transmission due to the relatively slow data processing by the ARM. Experimental tests show that the acquisition rate is up to 65 MHz, and the transmission rate is up to 25.6 Mbytes/s, and the normalized spectral intensity error is less than 0.5%. The achievements present herein can be generalized into such applications as precision instruments, equipment, digital devices and so on, wherein high speed and large throughput data acquisition and real-time data computation are usually indispensable.

Abstract:Rolling bearing is an important component of mechanical transmission equipment. The prediction of its performance degradation trend is the key to ensure the safe and stable operation of the equipment. In order to improve the accuracy of rolling bearing performance degradation trend prediction, a rolling bearing performance degradation trend prediction method based on the combination of multi-head-attention (MHA) and long short-term memory (LSTM) is proposed. Firstly, the multi domain features of time domain, frequency domain, time-frequency domain and Weibull parameters are constructed, and the multi domain features are screened according to the comprehensive performance degradation index. Secondly, the attention mechanism is used to enhance the weight of key features, PCA is used for feature fusion, and LSTM model is further used to predict the performance degradation trend of rolling bearing. Finally, the method proposed in this paper is verified by using the bearing fatigue life experimental data of NSF I / UCR center, and compared with several other models, which shows that the method proposed in this paper can more accurately predict the performance degradation trend of rolling bearing.

Abstract:In view of the lack of specific operation methods of single particle effect simulation in large capacity NOR Flash memory, this paper proposes three methods of single particle effect software fault injection in NOR Flash memory: single particle flip, single particle function interrupt and single particle latching. The board level test system is designed for large capacity devices and its function is verified. The single particle effect simulation experiment is carried out by fault injection method. The single particle effect test system of NOR Flash memory consists of FPGA control logic，Flash detection board and host computer software. Results indicate that single particle flip, single particle latching and single particle function interruption software fault injection methods are verified by single particle effect test system of NOR Flash memory. This paper can provide a reference for related single particle simulation and the electrical system reliability analysis for single particle effect.

Abstract:UHF time difference location method is difficult to be widely used in pd monitoring scenarios because of its high requirement on the calculation accuracy of arrival time difference of discharge pulses. In this paper, a technique of partial discharge location in GIS based on micro-information and statistical information is proposed. Firstly, the micro-information of discharge pulse is used to verify the arrival time and correct the time difference, and then the probability distribution of multiple location results is used to determine the final location result. Field experiments and case studies show that this method can locate the power supply accurately no matter it is a single discharge defect or multiple discharge defects.

Abstract:A novel Underwater Acoustic Covert Communication Scheme based on sparse multiple access (SCMA) is proposed to solve the problems of high bit error rate (BER) and low spectrum utilization in traditional covert underwater acoustic communication. The scheme generates new SCMA codewords at the transmitter using Eisenstein integer and Logistic chaotic sequences, and then a codebook via a mapping matrix. At the receiving end, the original information is recovered by multi-user Detection Algorithm and decoding. Finally, the feasibility of the scheme is verified by experimental simulation, and the influence of different number of users on the performance of underwater acoustic communication is analyzed. At the same time, the sparse multiple access communication scheme is compared with the spread spectrum communication scheme, the simulation results show that the former has lower bit error rate and better performance.

Abstract:To address the shortcomings of the current local stereo matching in weak texture regions with low matching accuracy and over-dependence on the central pixel, an adaptive local stereo matching algorithm based on the improved Census transform is proposed. Firstly, the Census transform is improved by using adaptive support window according to the texture complexity of the central pixel domain, introducing Tanimoto coefficients combined with Hamming distance algorithm, and fusing the absolute value of color or luminance difference as the new initial matching cost calculation. The cost aggregation is performed by the cross-cross domain algorithm and the winner-take-all algorithm is used to calculate the parallax. The left-right consistency method, iterative voting, interpolation filling and sub-pixel refinement are used in the parallax optimization stage, and the improved adaptive median filtering is used as noise suppression for edge blurring to obtain the final parallax map. The experimental results show that the proposed algorithm has an average mismatch rate of 4.39% on the Middlebury dataset, which is a significant improvement over other improved Census transform algorithms, and is robust and adaptive in terms of noise immunity.

Abstract:Accurate load forecasting is beneficial to the stable operation of the power system, improving economy and reliability. In order to improve the short-term power load forecasting accuracy, a short-term load forecasting model based on the improved chimp optimization algorithm to optimize long-short-term memory network is proposed. Because the chimp optimization algorithm is prone to fall into local optimum and has low optimization accuracy, the Circle mapping strategy is used to initialize the population to generate a uniformly distributed chimp population, improve the diversity of the chimp population, and lay the foundation for global optimization; secondly, the introduction of a spiral The position update strategy enables the chimp population to have multiple search paths, expand the search space, and improve the global search ability of the population; then, the Levy flight strategy and the adaptive t mutation strategy are introduced to perform disturbance mutation at the optimal solution position to enhance resistance to local extremes. It can improve the convergence accuracy of the algorithm. Aiming at the problem that parameters such as the number of hidden layer neurons and the learning rate of the LSTM network are difficult to select, ICOA is used to automatically find the optimal parameters for the LSTM network, and an ICOA-LSTM load prediction model is established. Combined with the actual data of a certain area, the prediction analysis is carried out. The results show that compared with the BP, LSTM, PSO-LSTM, and COA-LSTM prediction methods, ICOA-LSTM model has higher short-term power load forecasting accuracy, its forecast mean absolute error is 17.07kW, the root mean square error is 21.80kW, and the mean absolute percentage error is 0.37 %.

Abstract:Large indoor scenes usually have similar structures in the elevation direction, which leads to the degradation of the features of lidar scanning point clouds in the elevation direction, and the traditional lidar SLAM is prone to the mismatch of elevation features. In view of this, a decoupling SLAM algorithm for airborne 3D lidar based on inertial/altitude sensor information assistance is proposed: altitude sensor and inertial attitude are introduced into the point cloud initialization process to improve the initial pose matching accuracy; Decouple the point cloud registration algorithm based on multivariate normal distribution in horizontal and height channels, restrict the direction of point cloud registration, and improve the positioning accuracy in the environment of elevation degradation; At the same time, the traditional SLAM six-dimensional pose solution is reduced to three-dimensional, which reduces the amount of calculation. The cabin simulation scene is built by Gazebo, and the proposed method is verified. The results show that the proposed method can improve the positioning accuracy of lidar SLAM under the degradation of elevation features, which is more than 40% higher than the traditional algorithm, and effectively improves the calculation efficiency.

Abstract:In view of the fact that permanent magnet synchronous motor (PMSM) will be affected by nonlinear factors such as parameter perturbation and external uncertain interference in actual operation, which will lead to the decline of motor control performance and position tracking accuracy. To improve the control performance of permanent magnet synchronous motor(PMSM) under nonlinear factors, a nonlinear controller is designed by combining sliding mode control (SMC) with Backstepping Control, the reaching law in backstepping sliding mode control is improved, and a new sliding mode reaching law is designed by using a hyperbolic tangent function instead of symbolic function in traditional exponential reaching law. The nonlinear disturbance observer (NDOB) is used to observe and estimate the disturbance and compensate the disturbance, which is combined with the designed nonlinear control to track the motor position. Finally, the improved backstepping sliding mode control module and nonlinear disturbance observer module are built by using Matlab/Simulink. The results show that this method improves the motor position control accuracy and tracking response speed to a certain extent, reduces the position tracking error, and enhances the anti-interference ability of the system.

Abstract:For the temperature control system of the capacitor hot press, there are problems such as large lag and nonlinearity. The principle of the temperature control system of the capacitor hot press is analyzed and the mathematical model of the temperature control system is established. Through the research of traditional PID control algorithm, particle swarm algorithm and fuzzy control algorithm, a scheme of applying fuzzy particle swarm PID algorithm to temperature control system is proposed. Use MATLAB to simulate the temperature control system. The adjustment time and overshoot of the temperature control system using the traditional PID control algorithm are 66s and 58.173%, respectively. The temperature control system using the fuzzy particle swarm PID algorithm has an adjustment time and overshoot of 34s and 6.295%, respectively. The research results show that the temperature control system of the capacitor hot press based on the fuzzy particle swarm PID algorithm is superior to the traditional PID control algorithm in terms of adjustment time, overshoot and anti-interference ability.

Abstract:Aiming at the demodulation of vibration measurement signals by sinusoidal phase modulation laser interferometer, this paper proposes a demodulation method that integrates adaptive filtering and normalized PGC-Arctan. Based on the traditional PGC-Arctan algorithm, this method normalizes the quadrature interferometric signal pair by accurately identifying the carrier phase delay and phase modulation depth, thereby reducing the nonlinearity error of phase demodulation, and introducing an adaptive filter based on the minimum mean square algorithm to filter and reduce the noise of the demodulated signal, further improving the signal's SNDR (signal-to-noise distortion ratio) of the signal. The effectiveness of the algorithm is verified by numerical simulation and experimental testing, and the micro-vibrations of solid surfaces excited by sound waves at frequencies of 100Hz-3KHz are detected and demodulated under laboratory conditions. The results show that the method described in this paper can achieve accurate demodulation of the vibration signal, and the SNDR of the demodulated signal after adaptive filtering is increased by an average of 12dB.

Abstract:Aiming at the difficulty of trajectory tracking of helmet wearers caused by target occlusion in construction sites, a helmet wearing detection and trajectory tracking method combining YOLO v5 and centroid matching algorithm is proposed in this paper. Firstly, YOLO v5 network is used to precisely detect the personnel who do not wear safety helmets and calculate their centroid coordinates. Further, the extended Kalman filter is used to predict the target position information. Finally, the centroid matching association algorithm based on Mahalanobis distance and histogram correlation is adopted. Combined with the prediction information, the target trajectory anomaly correction in the target occlusion environment is realized, and the accurate target trajectory can be obtained. The experimental results show that the proposed method effectively solves the problems of target exchange and loss caused by target occlusion in target tracking, and obtains more than 10% target tracking accuracy higher than the traditional algorithm in the self-built data set, It provides strong technical support for the development of smart construction sites.

Abstract:Aiming at the characteristics of small retinal vessels and complex scale changes, a multi-level adaptive scale U-shaped retinal vessel segmentation algorithm is proposed. Firstly, the residual module is introduced based on the encoder-decoder structure to enhance the channel feature propagation capability. Secondly, a multi-scale feature extraction module is embedded at the bottom of the network to adjust the receptive field to effectively extract multi-scale features. At the same time, an improved adaptive feature fusion module is added to the skip connection part to promote effective fusion between adjacent hierarchical features to extract more small blood vessel features. Finally, the multi-level attention structure output on the setting side of the decoding part performs adaptive refinement on the multi-level features. The experimental results show that the accuracy of the algorithm on the DRIVE, STARE and CHASEDB1 datasets reaches 0.9645, 0.9694 and 0.9671, respectively, the sensitivity reaches 0.8417, 0.8465 and 0.8545, and the AUC reaches 0.9866, 0.9908 and 0.9877, respectively, and the overall performance is better than the existing algorithms.

Abstract:Aiming at the problem of line collision accident of large cranes under transmission lines, a method based on object detection and binocular distance measurement is proposed to control the risk. Firstly, the proposed method uses YOLOv4 algorithm to detect the transmission lines. Then, considering that binocular cameras at different angles can lead to image brightness differences, A binocular ranging algorithm based on AD-Census cost SGBM (Semi-Global Block Matching) is proposed. Finally, the effectiveness of the proposed method is verified by experiments. The results show that the average confidence of the method can reach 81.67%, the measurement error can be controlled within 0.4 meters within 5~8 meters, and the average detection time is 50ms. Compared with the original binocular ranging algorithm, the measurement accuracy of the improved algorithm is improved to some extent. The proposed method can accurately measure the distance between the crane boom and the transmission line, which has certain significance to prevent the crane from hitting the line.

Abstract:In the harsh on-site environment when sudden disasters such as earthquakes and mining accidents occur, the urgency and danger of rescue and search and rescue tasks highlight the urgent need for intelligent search and rescue robots. It is aimed at the situation that the search and rescue robot has difficulty walking in the narrow space of the ruins, and in the low-light environment, the image processing time is long and the details are lost. Firstly, the 3D model of the robot is designed based on bionics, and the motion of the robot is controlled; secondly, in order to improve the illumination consistency of the image in the search and rescue work, a structure-aware smooth model based on Retinex retina is added, which provides a high-quality search and rescue robot. The output can be seen, and an image evaluation model is established to eliminate low-quality images. Finally, the algorithm accelerates the solver to reduce the image processing time to meet the needs of real-time output of high-quality images on the Raspberry Pi. The experimental results show that when the hexapod robot works in a low-light environment, the image processing time of the Raspberry Pi is only 0.23 seconds, which greatly reduces the output delay, and the peak signal-to-noise ratio is 14.752 dB. It has great application prospects in future geological surveys, seismic searches, and difficult terrain detection.

Abstract:Panel module is one of the key factors that affect the touch and optical performance of the vehicle touch screen. Traditional manual inspection is unsuitable to modern industrial production with high product quality and high production efficiency. Aiming at the problem of large fluctuations of transmittance and large inspected area for the vehicle touch screen, a panel module machine vision inspection system with high inspection accuracy was developed. According to characteristics of the inspected glass object of vehicle touch screen, an image acquisition system and image processing scheme with online scanning has been designed for vehicle touch screen. By comparing the standard grayscale model, establishing the conversion relationship between grayscale value and transmittance, as well as support vector machine classification, various defects of the glass module were dynamic identified with only one CCD, and its transmittance at anywhere were acquired meanwhile. The on-site application results showed that the efficiency of this inspection system was up to 120pcs/h. The inspection accuracy of glass module inspection system was 0.01mm, and its ratio of flaw inspection was higher than 95%.

Abstract:In order to solve the problem of slow signal processing speed of satellite receiving system in military information reconnaissance, a satellite signal channelization scheme based on GPU is designed. Using digital down conversion cascaded multiphase channelization to improve the flexibility of sub-channel bandwidth location; Two-dimensional Block multiplexing is designed to process the multichannel down conversion process in parallel. A multiphase filtering kernel based on shared memory is designed to perform the weighted stack operation in 72 channels channelization. The test results of NVIDIA RTX 2060 and Intel(R) Core(TM) I5-6400 in CUDA environment show that the 288 channel channelization of four 75M real signals takes 4.51s in parallel processing, which is 40.27 times faster than that of single CPU processing. It greatly improves the processing speed of the receiving end of reconnaissance satellite.

Abstract:Using pulse shape discrimination (PSD) to distinguish between neutrons and gamma rays is an important task in the process of nuclear detection.. Based on the Labview platform, this paper realizes the simulation and signal preprocessing process of n/γ pulse signal. The traditional discrimination method, charge comparison method, pulse gradient analysis (PGA) method, and rise time method are used to perform the n/γ pulse signal screening, screening out the neutron and γ-ray mixed pulse signals with the same results of the above three screening methods as the training set of the KNN classification algorithm. The KNN classification model is constructed by training samples, so that the classification of neutron and gamma-ray pulse signals can be realized through this model. The results show that the accuracy of neutron and gamma-ray pulse waveform discrimination based on the KNN classification algorithm is as high as 99.58%. Compared with the charge comparison method, the rise time method and the PGA method, the discrimination error rate is significantly reduced. And the KNN classification algorithm is simple in principle and easy to implement, so it can be applied to the discrimination of n-γ pulses in the actual mixed field.

Abstract:Aiming at the complexity of clutter caused by the uncooperative radar emitter using an uncontrollable third party emitter, a dynamic method of clutter is proposed according to the characteristics of clutter covariance matrix, and a target detection system for uncooperative radar emitter is designed. The results are as follows: after clutter dynamic processing, accurate information of radiation source, target position, velocity and Angle can be extracted by non-cooperative radar radiation source target detection system. The results show that the dynamic clutter can not easily submerge the target with high power, and the target detection system can collect and process the target data, and the errors of position and velocity are less than 1%, which can be used as reference for target tracking and radar signal processing in actual situations.