Abstract:This paper designs a new integrated structure of limiting and filtering that works in the 5G mobile communication frequency band, which is suitable for RF front-end and has good point-to-point protection. By integrating the two modules of limiting and filtering, the structure avoids the construction of a matching network between the modules and reduces the circuit volume. Compared with the traditional ferrite frequency selective limiter, the structure achieves the characteristics of wide span and high power capacity in the target frequency band. The test results show that: in the 2.55-2.65GHz frequency band, the insertion loss of the structure is less than 1.8dB, and the clipping level is about 7.5dBm; in the 4.8-4.9GHz frequency band, the insertion loss of the structure is less than 2dB, and the clipping level is less than 2dB. About 2.5dBm. Outside the selected frequency band, the leakage power is attenuated to be negligible. Therefore, the structure designed in this paper can effectively improve the sensitivity of the RF front-end receiver while protecting the sensitive components.

Abstract:To reveal pressure distribution and attenuation law of shock wave in bifurcated tubes, a modular bifurcated tube model was built for shock wave propagation test. The overpressure of shock wave at each measuring point under different membrane breaking pressures and different bifurcation angles was obtained, and the attenuation curve was fitted with Rankine-Hugoniot equation. The pressure velocity correlation algorithm of FLUENT was used to simulate the propagation process of shock wave at the tube bifurcation and observe the distribution of pressure flow field in tubes. The results show that the bifurcation angle has a significant impact on the pressure distribution of the main tubes and branch tubes. The attenuation rate of shock wave in the tubes is related to the membrane breaking pressure. Among the three bifurcation tubes of 30 °, 90 ° and 150 °, the platform pressure of 90 ° branch tube is the lowest. Under the same membrane breaking pressure, the attenuation rate of branch tubes is opposite to main tubes.

Abstract:Owing to the outstanding commercial value, electric-drive-reconstructed onboard chargers (EDROCs) have been a hotspot of research in electric vehicle (EV) field. This paper studies the fault diagnosis scheme and fault-tolerant control of a six-phase EDROC, considering the single-phase open-circuit condition. At first, the circuit topology and the basic operation principle of the six-phase EDROC are explained. Then, a simple fault diagnosis scheme is proposed for the open-circuit fault. The proposed scheme is composed of two parts, namely the fault detection and the fault localization, which are realized by estimating the derivative of the phase of winding current vector and judging the polarity of winding currents, respectively. Thereafter, the fault-tolerant control of the considered single-phase open-circuit fault is concerned. The amplitudes and phases of stator currents are adjusted, acquiring the purposes of balanced grid currents and non-generation of rotating filed. At last, an experimental test rig is built and the results verify the fault diagnosis scheme and the fault-tolerant control proposed in this work.

Abstract:In order to solve the problem of incomplete representation when the wearable MEMS sensor is used to detect human fall behavior in multiple scenes, a SVM human fall detection and recognition method is proposed based on improved sparrow search algorithm (ISSA). Firstly, the wearable MEMS sensor is used to collect the discrete attitude data of human body. Then, the acceleration threshold and angular velocity threshold eigenvectors are found through the time sliding window and the first-order judgment was performed. At the same time, an ISSA-SVM detection model of fall state is constructed, that is, the kernel parameters and penalty factors of SVM are adaptive optimized by the improved sparrow search algorithm to obtain the optimal classification model. Finally, according to the SVM classification model, the data of the first-level decision are analyzed to judge whether the fall is real. Experimental simulation and product application results show that the test accuracy of the proposed ISSA-SVM model for the detection of human accidental falls in different scenarios is more than 98%, and the failure rate is reduced. After many tests, the fall detector shows good robustness.

Abstract:To meet the need for a ultra-reliable and low-latency service transmission of large-scale equipment access in substations, we propose a multi-frequency heterogeneous wireless communication network access selection algorithm for substation services. Considering the reliability and effectiveness requirements of these services, we firstly construct a substation scenario model under heterogeneous wireless network coverage. Secondly, to effectively improve the freshness of the information received, we utilize the Average Age of Information (AAoI) as the optimization target function of wireless network access selection and propose the optimization problem of substation service access selection based on the age of information. Finally, we implement the Deep Q-Learning (DQN) method to obtain the best access selection scheme. It can be seen from the analysis of the application examples and the test data that the proposed access selection optimization theory and algorithm can reduce the average age of information during service transmission and improve the freshness of data.

Abstract:This paper designs a lightweight EtherCAT master and its test system based on Raspberry Pi. On the test system, the performance parameters of several parallel running masters are measured by multi-channel listeners, the delay jitters in polling/interrupt and single/multi-core modes are measured by means of software and hardware timestamp，and the delays and jitters of frame at each protocol layers are also measured. The EtherCAT master station based on Raspberry Pi adopts VxWorks operating system, the main optimization direction includes system kernel configuration, network adapter driver, message processing, application layer software, etc. By testing, the master has stable performance, the period jitter is about 6μs under 125μs period, which can meet the requirements of industrial field data acquisition, building automation, AGV control and so on.

Abstract:Nowadays, virtual reality (VR) technology has been widely used in various fields, but many VR systems will cause users to have an uncomfortable symptom when they bring immersive experience -- Virtual reality motion sickness. In order to understand the effects of virtual reality motion sickness on brain neural activity, this study recruited subjects to experience virtual reality motion sickness induced scenes by head-mounted VR, and recorded their EEG signals before and during the experience. Sample entropy and Power spectrum were used to extract EEG characteristics of subjects in different states, and significance test was carried out. In the whole frequency band, the mean values of Sample entropy at electrodes F8, F12, CZ, CPZ and OZ were significantly different, and the mean values of Power spectrum at electrodes F7, T7 and T8 were significantly different (P<0.01). In terms of frequency bands, the mean values of Sample entropy and Power spectrum in Delta and Theta bands were significantly different (P<0.01). The results show that the Sample entropy and Power spectrum analysis results may be related to virtual reality motion sickness, which is expected to be an effective index to measure virtual reality sickness.

Abstract:Multiple Depots Multiple Traveling Salesman Problem is an extension of Traveling Salesman Problem. To solve this problem, a parallel K-means clustering simulated annealing algorithm based on Spark platform is proposed. The algorithm firstly classifies all the cities by k-means clustering algorithm, then establishes a traveling salesman problem for each class, and solves the traveling salesman problem by an improved simulated annealing algorithm. The MMTSP solution is calculated from the sum of the shortest paths of these classes. The proposed algorithm adopts the solution strategy of clustering first and then simulated annealing algorithm, which can greatly reduce the simulated annealing search space. Moreover, Spark platform can divide the clustering algorithm into several classes for parallel solution, so as to obtain the optimal solution of MMTSP problem faster. Several test instances in TSPLIB database are selected for simulation experiments, and the solution accuracy and running time are tested. The experiments are compared with other related algorithms. The experimental results show that, compared with the current FCMPGA, IPGA, IWO and other algorithms, the solution accuracy is improved by 5%~40%, and the solution efficiency is improved by 1~5 times compared with other algorithms, especially when the value of K is large.

Abstract:With the development of economy in China and the improvement of social life, the national general aviation industry is rising step by step. Low-altitude airspace is also gradually opening up. At the same time, the opening of low-altitude airspace has further promoted the development of the aviation industry. Some new low-altitude aircraft have been put into use. The increase in the number of low-altitude aircraft has resulted in frequent safety accidents in low-altitude airspace. In order to settle with the security issues in low-altitude airspace, we need to build a healthy low-altitude airspace surveillance system. This paper firstly analyzes the current situation of low altitude airspace security. Secondly, it introduces the basic situation of the ADS-B technology. Thirdly it introduces the functional module of the ADS-B system. Then it expounds the application and trend of ADS-B technology in airspace, and discusses the existing problems this technology and possible solutions. Finally, it elaborates the future development and prospects of ADS-B technology in low-altitude airspace security.

Abstract:In the vector control system of permanent magnet synchronous motor (PMSM), aiming at the problems of insufficient control accuracy, large overshoot and easy to be affected by external disturbance of the traditional PI controller, an inverse sliding mode controller is proposed to replace the traditional PI controller. The inverse sliding mode controller is produced by the combination of inversion method and sliding mode control theory, In the design of reaching law, hyperbolic tangent function and system state variables are introduced to form a new reaching law, which replaces the traditional exponential reaching law, improves the convergence speed of the system and reduces the chattering phenomenon of sliding mode. The simulation model is built in MATLAB/Simulink. The simulation results show that this method makes the system have a good control effect on the motor speed. Compared with PI control, traditional sliding mode control and variable exponential sliding mode control, it has better dynamic response performance, the overshoot is reduced by 22.15%, 18% and 2.75% respectively. The anti-disturbance ability of the system is improved and the system has strong stability.

Abstract:The traditional fuzzy neural PID control algorithm is prone to the problem of poor control effect caused by improper adjustment of network parameters. This paper presents a fuzzy neural network PID control algorithm optimized by improved grasshopper algorithm. Aimed at the problem of insufficient traditional algorithms of grasshopper particle diversity firstly introduced Levy random flight strategy, secondly introduce nonlinear reduction factor and simulated annealing algorithm to improve the optimization ability and the ability to jump out of local optimal solution, then the improved grasshopper algorithm combined with fuzzy neural PID neural network is optimized by super parameter and control parameter self-tuning, Finally, the simulation results verify the superiority and reliability of the proposed improved grasshopper algorithm to optimize the fuzzy neural network PID algorithm.

Abstract:In order to solve the problem of poor anti-interference ability of the permanent magnet synchronization motor in the direct torque control system, the speed loop adopts the active disturbance rejection controller (ADRC) to replace the traditional PI controller, removes the tracking differentiator in the ADRC to improve the system signal tracking rate, and reduces the burden of the ADRC through introducing the load observation to feed-forward compensation on the speed loop. Meanwhile, considering the interference caused by DC offset in current sampling, the traditional voltage model is filtered by a second-order high-pass filter and the second-order high-pass filter is discretely processed by the trapezoidal discretization method, which improves the observation accuracy of the observer. The filter cutoff frequency is set to follow the change of the electrical angular velocity so that the dynamic performance of the system is improved. Finally, simulation results suggest that the system output speed is super-adjustment, the output magnetic chain, high torque accuracy, and good anti-interference ability.

Abstract:To improve the accuracy and applicability of the monocular depth estimation network with supervised learning for actual scene ranging tasks, a distance calculation method based on monocular depth estimation and calibration mechanism is proposed. Firstly, by introducing multivariate attention blocks and optimizing the design network structure, a network integrating global context and spatial attention mechanism (GSNet) is constructed. Then calibration parameters are formulated to establish the proportional relationship between the predicted distance of the scene and the actual distance of the scene, to obtain the calibrated distance value. Experimental results show that the fusion network GSNet and calibration mechanism can effectively reduce the error of the monocular depth estimation method in the actual measured distance. Compared with other methods, the average absolute error is less than 0.15m, and the average relative error of the measured distance in this method is less than 10%, which has good feasibility and accuracy.

Abstract:In order to improve the control accuracy and response speed of the temperature control system of the herbal medicine roaster, an improved fuzzy sliding mode control method is proposed. The improvements include the design of sliding mode surface and reaching law. A sliding mode surface with exponential function is designed. When the error is large, the control heating system maintains the maximum power output and shortens the system adjustment time; In order to suppress the chattering phenomenon caused by fixed switching gain, an exponential reaching law with switching gain of power exponential function is designed. The weight of power exponential function is adjusted by fuzzy controller. When the system error is small, the chattering is effectively suppressed by using the advantage of rapid attenuation of power function. MABLAB/simulation results show that compared with the traditional fuzzy sliding mode control, the regulation time of this control strategy is reduced by 54.94%; It has obvious inhibitory effect on buffeting. The engineering application results show that when the herbal medicine roaster with this control method works stably, the temperature fluctuation range is±2.12℃, which meets the requirements of traditional Chinese medicine frying process.

Abstract:As a promising distributed machine learning paradigm, Federated Learning brings huge privacy-preserving potentials, and has become a hot topic of research in recent years. To tackle client drift induced by statistically heterogeneous user data, this paper first presents an intermediate feature generation method based on Generative Adversarial Networks for the aim of classifier correction. Secondly, to deal with the particular problem of concept shift, a personalized model aggregation approach is proposed on the basis of classifier clustering. Finally, the two strategies mentioned above are integrated and tested on the CIFAR-10 image classification dataset. Various empirical results show that the proposed integrated strategy, compared to the classic Federated Averaging algorithm, helps realize both better generalization of the single-center global model, and better personalization of the multi-center cluster models.

Abstract:Dangerous rocks and falling rocks around the railway intrude into the railway boundary, which will seriously endanger the life and property of passengers and the safety of railway traffic Aiming at the problems that traditional detection methods have high false detection in complex dynamic environment and low accuracy of small target recognition, a video-based deep learning method for railway rockfall intrusion detection is proposed. First, a hybrid attention module is incorporated into the network structure, which can enhance the network's ability to detect rockfalls similar to the background. Secondly, part of the network structure of YoloX is improved to a bidirectional feature pyramid network, which strengthens the mutual exchange of features at different levels and improves the ability to identify small targets. Simultaneously collect a large number of simulated rockfall data from different scenarios, build a simulated rockfall data set, and use the Mosaic data enhancement method in training to enhance the generalization ability of the method. The experimental results show that with the addition of improved modules, the identification accuracy of the method in this paper is continuously improved. Compared with various mainstream target detection methods, the highest identification accuracy is achieved, and the identification of different sizes targets is stable, which proves the application value of the algorithm in this paper in the actual railway scene.

Abstract:Aiming at the slow detection speed of traditional retinal blood vessel segmentation algorithm, it is difficult to apply to real-time medical aided diagnosis system, a lightweight retinal blood vessel segmentation model based on Shuffle-Unet is proposed. In order to simplify the model structure, the lightweight model ShuffleNetV2 is structurally pruned, and the last convolutional layer, global pooling layer and fully connected layer on the ShuffleNetV2 structure are pruned; In order to reduce the computational complexity of the model and improve the running speed of the model, the pruned ShuffleNetV2 is used as the backbone extraction network of the model; Use the random channel separation operation module to build an upsampling model structure to enhance the network feature transfer capability; The attention mechanism module is used to fuse the output of the first feature layer and the upsampling layer of the model to enhance the extraction of effective features from the two dimensions of the channel and the space. By comparing the two public datasets DRIVE and CHASE_DB1 with other retinal blood vessel segmentation algorithms, it effectively proves that the Shuffle-Unet model has the characteristics of high segmentation accuracy and high detection speed.

Abstract:Due to the large number of pictures of electric power fittings taken in the power inspection, the inspection workload is large. In order to improve the automatic detection effect of electric power fittings, this paper proposes a UNet-DB_ECA (UNet Dimensionality Reduction, BN, and ECANet, UNet-DB_ECA) network detection method based on UNet network. First reduce the width of the UNet network, then embed the efficient channel attention mechanism module ECANet (Efficient Channel Attention Networks, ECANet) and Batch Normalization (BN) in the network, and finally introduce the Hard-Swish activation function, thus constructing UNet- DB_ECA network. This paper uses the electric power fittings detection dataset to conduct experiments. The experimental results show that the method proposed in this paper has a good detection effect. Compared with the UNet network detection effect, it improves the detection effect and also takes into account the algorithm performance. In addition, the power fittings detection dataset contains seven types of fittings with different shapes, which shows that the method proposed in this paper has good generalization ability, so the method has certain application prospects in the automatic detection of power fittings.

Abstract:The state assessment of overhead transmission line fittings is crucial to the reliable operation of the line, and the detection of the fittings is an important part of the assessment work. In response to the heavy workload of manual labeling of datasets in identification and detection of fittings, as well as the difficulty of balancing high precision and rapidity, an improved transmission line fittings detection method based on YOLOX network is proposed. The fitting images captured by UAV are augmented with preprocessing to enrich the datasets. The backbone network adopts the enhancement methods of online Mosaic and Mixup. The transfer learning based on feature extraction is introduced and the cosine annealing learning rate is used for two-stage model training. The experimental results show that the mean average precision of the improved method for the detection of all types of fittings is improved by 18.32%. Compared with five mainstream detection models such as Faster R-CNN algorithm, the mean average precision of proposed method is the highest, and its detection speed is lower than YOLOv3’s, which can identify various types of fittings more quickly and accurately, and reduce the workload of manual labeling to a certain extent.

Abstract:Aiming at the problems of low accuracy of traditional orb-slam2 algorithm, easy loss of picture frame tracking, and the lack of dense point cloud map and octomap, the originally constructed sparse point cloud map can not be directly applied to the three-dimensional path planning of robot. Based on the traditional orb-slam2 algorithm, this paper improves the selection of key frames. Firstly, based on the traditional orb-slam2 algorithm, the comprehensive transformation factor of relative motion is added between adjacent image frames, and the inter frame feature point tracking is added to improve the accuracy of key frame selection; Then, the selected key frame is used to construct the dense point cloud map and octomap; Finally, the verification is carried out on the tum data set, and the physical test is carried out based on the real environment. The experimental results show that the improved key frame selection method can increase the positioning accuracy of orb-slam2 algorithm on the premise of ensuring the accuracy and rapidity of key frame selection, effectively alleviate the problem of easy loss of picture frame tracking, and the octree map can be directly used for robot 3D path planning.

Abstract:Vertebra CT image segmentation is the key to the visualization of vertebra 3D reconstruction. Aiming at the problems of blurred vertebra edge, complex structure and changeable shape in vertebra CT images, a dual-decoder network is proposed based on deep learning method. The network adds a parallel decoding branch with the same structure on the basis of the U-Net structure of the encoding and decoding network, and the two decoding branches can extract image features complementary. Moreover, a dual feature fusion module is added between encoding and decoding to solve the problem of semantic information loss caused by the network downsampling and upsampling. At the same time, the original convolution module is replaced by the densely connected hybrid convolution module to improve the network's ability to extract multi-scale features. In addition, an efficient attention module is added to make the network focus on learning regions of interest in space and suppress irrelevant information in channels. Tested on the CSI2014 public dataset, the Dice coefficient reaches 0.970, the Jaccard coefficient reaches 0.945, and the Recall rate reaches 0.962. The experimental results show that the network can improve the accuracy of vertebra segmentation, has better generalization ability, and can meet the needs of clinical vertebra CT image segmentation.

Abstract:The research goal of this paper is to redesign a feature pattern for extracting feature points to meet the needs of long-distance visual positioning. Similar to the two-dimensional code, the feature pattern in this paper is also composed of black blocks, but the number of arrangements is relatively small. the black blocks have continuity, which greatly reduces the resolution requirements, and the template matching algorithm is used to identify them, and the similarity rate is 80% to identify the definition in the text, and the comparison experiment is used in the experiment. In the same environment and shooting distance, the effective recognition distance of the characteristic pattern is 225% higher than that of the QR code. Under the size of 30*30cm, the longest recognition distance reaches 90m, and in the robustness experiment A variety of interference processing is carried out on the characteristic pattern and all of them can be identified, showing sufficient anti-interference ability, and suitable for outdoor applications. Therefore, the characteristic pattern proposed in this paper can basically meet the needs of long-distance visual positioning, and is very suitable for application in vehicles. Visual positioning.

Abstract:To solve the problem of weak signal of shear horizontal guided wave detection across weld, a time reversal detection method based on SH guided wave mode is proposed in this paper. The low frequency SH0 guided wave mode is selected for this detection. A 3D equivalent model for the defect detection of cross weld on a 3mm thick plate was established, and the time reversal detection was simulated numerically and demonstrated by experiment. Firstly, it is verified that time reversal can improve the sensitivity of post-weld defect detection from the perspective of simulation. Then, the PPM EMAT system was designed and built to generate SH0 mode wave for conventional and time-reversal detection experiments. Experiments verify the validity of a time reversal detection method. Experimental and simulation results show that the time reversal method has a strong focusing effect on the low defect echo signal in conventional detection. This method can increase the conventional signal amplitude to 3.22 times, and can be used to detect 1mm longitudinal crack defects effectively. This paper provides a new idea for conducting wave detection of defects across welding seams.

Abstract:A metal contaminants detection system based on FPGA is proposed. The system solves the problem of low sensitivity and high false positive rate under the influence of product effects. In this paper, the principle of the balanced coil sensor is analyzed, and a balanced compensation algorithm is proposed to achieve the automatic balancing of the sensor. The metal signal is modulated by high-frequency excitation, and a lock amplifier is implemented in the FPGA to demodulate the received signal, which improves the signal-to-noise ratio. The output signal of the sensor is processed by cross-correlation algorithm, the characteristic phase of the product is obtained by fitting the product learning data, and the product signal and metal signal are distinguished. The test results show that the system has high sensitivity and can detect metal contaminants effectively under strong product effects.