Abstract:A data transmission system based on PXIe bus and FPGA was designed to realize the acquisition, uploading and analysis of image data in the focusing and leveling system based on linear CCD imaging principle. The system uses FPGA Mezzanine Card (FMC) architecture, composed of mezzanine card and carrier card. The mezzanine card is equipped with four high-speed optical fiber communication interfaces to realize interface interconnection with high-speed image output equipment. The carrier card adopts FPGA as the main controller to receive the image data transmitted by the sandwich card in real time, cache it to the onboard DDR SDRAM, and then transmit it to PC through PXIe interface. After the development of the system, the data transmission performance of the optical fiber interface at the transmission rate of 10Gbps was tested, and the correct and clear track eye diagram was obtained. The batch data transmission test of PXIe interface is carried out, and the results show that the data sent by FPGA is completely consistent with the data received by the host computer, which verifies the correctness and rationality of the scheme.

Abstract:Firmware Over-The-Air (FOTA) is a technology which uses wireless communication to upgrade the software of electronic control unit. FOTA has been applied in a large scale in automobile electronic control devices, as a result, the vehicle control system is faced with increasing security threats from the public network. However, the present mainstream FOTA schemes merely focus on the remote transmission from the server side to the vehicle side, and the security handling of the in-vehicle firmware is still a weak link. This thesis proposes a multi-check scheme of firmware security, the server signs the firmware through the Elliptic Curve Cryptography-based digital signature algorithm, so as to get two check codes, which are respectively used for remote transmission and check of firmware integrity and identity in-vehicle handling process, thereby ensuring full-process security of FOTA. The experimental results show that the scheme proposed in this thesis can well identify the risk of tampering during remote transmission and storage of firmware, and the time cost increased is only about 5%. Meanwhile, compared with the scheme realized by using RSA, this scheme is advantaged by fast speed of signature verification and small occupancy rate of resources.

Abstract:With the trend of multifunctionality, miniaturization and light weight, this paper, based on former researches, presents a new design of planar ultrawideband modular antenna (PUMA) array with lower profile. By replacing the substrate materials and adding metal plate at edges of antenna, the antenna can achieve lower profile and good performance within operating band of 5.5～21.35GHz. The simulation results show that the Active VSWR < 2.0, 3.0, 3.5 while scanning to broadside, 45°, 55°, respectively, and both port isolation and cross polarization can be lower than -15dB. With good performance, lower profile leads to much less cost on engineering application.

Abstract:In the traditional deep reinforcement learning, the mobile robot can get a positive reward only when it reaches the target position within the specified time step in the sparse reward environment. Each step of the intermediate process is a path planning problem with negative reward. A path planning method based on improved depth Q network is proposed. In the process of exploration, the mobile robot samples the trajectory conditional on the real target, and replaces the real target with the state that the mobile robot has reached in the process of experience playback, so that the mobile robot can obtain enough positive reward signals to start learning. Through the deep convolution neural network model, the original RGB image is used as the input, trained through the end-to-end method, trained the neural network parameters by using the upper bound exploration strategy of confidence interval and the method of small batch samples, and finally obtained the Q values of up, down, left and right actions. In the same simulation environment, the results show that the algorithm improves the sampling efficiency, the training iteration is faster, and it is easier to converge. The success rate of avoiding obstacles to reach the end point increases by about 40%, which solves the problem caused by sparse reward to a certain extent.

Abstract:Predicting latent faults of transformers is essential to evaluate their health status. This paper proposes a new transformer fault prediction method. First of all, a prediction framework of temporal attention mechanism is built based on LSTM network, and the IALO algorithm is used to optimize the hyperparameters of LSTM. Afterwards, use the optimized model to predict the dissolved gas in transformer oil. Then, the SVM model optimized by the MPA algorithm is used to diagnose the gas prediction results. Finally, the fault diagnosis results are counted, and the model is verified by comparing with the actual operation state. The experimental results show that the abnormal operation status is up to 29 times form the 42th to 58th day, and the abnormal operation probability is 86.89% in the next two months, among which the proportion of medium temperature overheating fault is highest, 88.67%, and the errors from the actual situations are only 2.46% and 1.29%. The predicted results are in good agreement with the actual operating situations of transformers, which proves the feasibility of the proposed method in accurately predicting the time point and fault type of abnormal operation states of transformers.

Abstract:Part recognition is an important basis for the assembly and packing of mechanical components. The efficiency of manual recognition is low. The traditional machine vision inspection requires high and the scene is single. This paper proposes a machine part recognition method based on deep learning machine vision. The detection accuracy of the original Mask R-CNN instance segmentation model is improved by adding the PointRend module; the category subdivision method is designed for parts with high similarity, through size estimation and feature matching, It can better solve the problem of size feature loss caused by data-enhanced image scaling. Collecting 25 different parts for recognition experiments, the results show that the method in this paper can effectively improve the recognition accuracy of mechanical parts, and the algorithm can recognize similar parts with an accuracy of 100%, which is 11.51% higher than the original Mask R-CNN method. And the method in this paper can be extended to other recognition tasks with similar features.

Abstract:Aiming at the problem that it is difficult to fully extract the characteristics of small leakage signals of natural gas pipelines on a single scale, a method for identifying small leakage signals of pipelines based on the combination of variational modal decomposition (VMD) and multi-scale fuzzy entropy (MFE) is proposed. First, the VMD algorithm is used to denoise the pipeline negative pressure wave signal, and the effective mode of the VMD decomposition is determined and reconstructed by the Euclidean distance (ED) method to determine the VMD decomposition based on the principle of the highest signal-to-noise ratio of the reconstructed signal The number of modes; the multi-scale fuzzy entropy is used as the fault eigenvalue vector, and finally the support vector machine is used to classify and recognize the eigenvalue vector. The experimental results show that the overall recognition rate of the pipeline signal state by this method is 99.33%, which proves that the overall recognition effect of the method is good, and it can realize the accurate identification of small pipeline leakage signals.

Abstract:In response to the shortcomings of the Rapidly-exploring Random Tree connect (RRT-connect) algorithm in terms of time consumption and randomness of node sampling during path planning, the RRT-connect algorithm based on gravitational field guidance is proposed. The algorithm sets a third node as a new extension node between the start and end of the path so that it alternates between the three nodes to expand the random tree, while superimposing a gravitational field on each node to guide the direction of node generation in order to reduce the search range of the invalid space. The algorithm was studied in simulation experiments in three scenarios with few obstacles, many obstacles and the presence of narrow passages, and the results showed that the improved algorithm reduced the average number of iterations by 47.1% and the average path planning time by 43.4% compared to the base algorithm. This demonstrates that the improved algorithm can effectively reduce the planning time and has higher planning efficiency than the RRT-connect algorithm.

Abstract:Coiling temperature control precision is the main elements influencing the presentation of strip steel items, and further developing curling temperature control exactness and guaranteeing winding hit rate is a main point of contention in the field of hot rolling. To resolve the issue of low hit pace of individual steel grades in the current coiling temperature setting model of a steel mill, a new modeling idea based on Gray Wolf Optimized Extreme Learning Machine is proposed combining data mining and field master insight, and Henon mapping, small-hole imaging strategy and weight factor strategy are introduced to improve the Gray Wolf algorithm, and a hot-rolled based on Improved Gray Wolf Optimized Extreme Learning Machine (IGWO-ELM) is established. strip coiling temperature prediction model based on the Improved Gray Wolf Optimized Extreme Learning Machine (IGWO-ELM) and contrasted and ELM model, GA-ELM model and GWO-ELM model. The model results show that the established IGWO-ELM model has a hit rate of 91.1% for predicting the coiling temperature within ±3°C and 96.7% for predicting the coiling temperature within ±4°C, both of which are better than the comparison models and have a wide range of pragmatic application prospects.

Abstract:In order to solve the problems of differential evolution algorithm, such as easily falling into the local optimum and the identification accuracy needs to be optimized, the improved algorithm introduces the random walk strategy based on the original selection, mutation and cross operation, which enhances the local search ability of the algorithm and improves the diversity of the population. In this paper, based on the measured and predicted values of equivalent series resistance (ESR) and equivalent impedance (Z), the objective function is constructed, and the improved algorithm is used to optimize the objective function. The parameters of two electrolytic capacitor models with different complexity are identified, and the results of parameter identification and the predicted values of ESR and impedance Z are obtained. Simulation results show that the improved algorithm is effective, and the prediction accuracy of the improved algorithm under the classical model is always 5% better than that of the traditional algorithm, which can improve the monitoring accuracy of the electrolytic capacitor.

Abstract:With the development of technology in the defense industry, the application of missile navigation and guidance technology has become more and more extensive. It’s the most critical step to obtain attitude information to realize the flight control of the high spinner. However, to measure attitude, a disadvantage of traditional methods (e.g., IMU, geomagnetism) is that there is a large error in roll angle results and insufficient sensitivity. In this paper, a method of using the compensated three-axis geomagnetic data to measure the attitude is proposed. The method of least squares and ellipsoid fitting is used to calibrate the temperature and the number of missiles of the magnetic sensor, which compensates for the error of the geomagnetic data. On this basis, the method of using the geomagnetic information to measure the missile attitude is studied. The experimental results show that while the method proposed in this paper is easy to implement in engineering, the error of collected geomagnetic data is reduced by nearly 20% compared with the international geomagnetic standard ,and the error of attitude calculation is reduced by 14.21%. The accuracy of the corner calculation is improved by 30.72%.

Abstract:Aiming at the problem of precision loss caused by errors in dynamic testing system, a dynamic error tracing method based on ensemble empirical mode decomposition and BP neural network was proposed. Based on the theory of the whole system dynamic precision, the method decomposed the total output error of the dynamic test system through the EEMD, and performed Hilbert transform on the single error obtained by the decomposition, analyzed the amplitude-frequency characteristics of the error signals, and then used BP neural network to fitting and tracing. The simulation results show that the method can trace back the error module in the dynamic test system effectively, and the accuracy of deviation is up to 10-2, compared with empirical mode decomposition, the method has better traceability effect and avoids mode aliasing problems existing in EMD, it is feasible and applicable.

Abstract:In view of the weak magnetic control system of permanent magnetic synchronous motor (PMSM), the basic structure of the system is analyzed, but the system under the conventional fuzzy PID is still insufficient.In this paper, a control strategy combining fuzzy control with moth flame optimization (MFO) algorithm, constructs the current loop and speed loop in the control system and self-rectify the control parameters in real time.This paper tests the traditional PID control algorithm, fuzzy PID control algorithm, obtains the corresponding speed and torque simulation curve and analyzes the fluctuation error and recovery time. The final results show that the system recovery time under the control of the algorithm has improved 0.04s and 0.02s, compared with the traditional PI control algorithm and fuzzy PID control algorithm and anti-interference capability.

Abstract:EMG interference is the noise interference caused by slight muscle vibration in the process of ECG acquisition,so it is necessary to de-noise the acquired ECG signal further.Aiming at the problem of Oscillation and distortion of ECG signal after de-noising by traditional wavelet threshold, an improved wavelet threshold function with two dynamic parameters is proposed,the improved threshold function can be adjusted in whole and in part by the dynamic parameters, so that the noise can be denoised in different degree.In order to verify the de-noising effect of the new threshold function, the signal-to-noise ratio (SNR) , mean square deviation (MSE) and the time-frequency characteristics of ECG were used to evaluate the de-noising effect, the improved threshold function is compared with the traditional EMG denoising algorithm.The results show that the improved threshold function can preserve the characteristic waveform of ECG signal after de-noising, and overcome the shortcomings of soft and hard threshold function.Compared with the traditional EMG interference denoising algorithm, the signal-to-noise ratio of denoising slight noise and serious noise is 38.7948 and 36.7212 respectively, and the mean square deviation is 0.0013 and 0.0018 respectively,the denoising evaluation result is the best.

Abstract:In the process of oil recovery,the depth monitoring of the oil well's fluid level is very important to ensure the safe production of the oil well. When the acoustic method is used to measure the dynamic liquid level depth of oil well, the coupling wave and the liquid level echo of the acoustic signal received by the wellhead are easily interfered by the noise due to the influence of the complex structure in the casing. As a result, it is difficult for the traditional signal processing method to calculate the velocity and travel time of the acoustic wave, so the dynamic liquid level depth of oil well can not be obtained. In order to solve this problem, this paper firstly conducts Butterworth low-pass filter for coupling wave, and uses short-time mean amplitude difference function to obtain the average sampling times of coupling wave, so as to calculate the propagation velocity of acoustic wave in oil well. On this basis, the liquid level echo was denoised by wavelet, and the liquid level echo position was obtained by using the wavelet singular value detection method, so as to calculate the time difference between the detonation wave and the liquid level echo position, and then realize the detection of the dynamic liquid level depth of oil well. In order to compare the processing effect more intuitively and conveniently, this paper designs a visual data processing software based on Matlab. This paper selects multiple sets of acoustic signals obtained from oil production sites for testing. The results show that the absolute measurement error is controlled within 1 meter, and the relative error does not exceed 0.075%. Compared with other signal processing methods (relative error range 0.3% ~ 0.5%), the proposed method has less error and can better meet the needs of actual engineering.

Abstract:When the rolling bearing fails, the fault characteristic signal will be mixed in the vibration signal, resulting in unsatisfactory extraction effect of the fault characteristic signal. To solve this problem, a fault extraction method for rolling bearings based on complementary integration of empirical mode decomposition and multi-point optimal minimum entropy (CEEMD-MOMEDA) is proposed. At first, the collected vibration signals are processed by CEEMD algorithm, and then the non-fault impact components are s[基金项目：国家自然科学基金项目（61761049,61261022）]creened out by kurtosis criterion. finally, the recombined signals are processed by MOMEDA algorithm to suppress the influence of noise and extract fault features. And compared with the single MOMEDA algorithm. The results show that the fault extraction ability and anti-interference ability of the proposed CEEMD-MOMEDA algorithm are greatly improved.

Abstract:Nuclear power plants have gradually increasing demand for the use of plate heat exchangers. Existing fouling thermal resistance prediction models have low generalization capabilities and few design options from the time series angle. Through the principal component analysis of the experimental data of the RRI/SEC heat exchanger of Unit 1 of Ling'ao Nuclear Power Plant, the long- and short-term memory neural network design model was optimized to predict the instantaneous fouling thermal resistance, covering variables such as the temperature of 12 pipelines and the flow rate of 4 pipelines. The model can accurately predict the demand for dirt cleaning in the next 25 days with an accuracy of 99.35%. In actual use, it can reduce the labor cost of heat exchanger monitoring, so as to stop and clean some units of plate heat exchangers in advance, extend the life cycle and improve heat exchange efficiency.

Abstract:To address the problems of large similar block matching errors and insufficient protection of image details in the classical BM3D denoising algorithm, an improved BM3D image denoising algorithm based on rotated blocks is proposed. The new algorithm firstly rotates the reference block at different angles to obtain the rotating block, and then performs the similar block matching process through the rotating block; then uses low-rank regularization to replace the hard threshold filtering in the traditional algorithm; finally, the BM3D algorithm combining rotated block matching and low-rank regularization is adaptively adjusted to improve the denoising effect in uniform image regions. Experimental results show that the new algorithm has a higher matching degree of similar blocks, and the peak signal to noise ratio (PSNR) is improved by 0.5dB on average compared with the classical algorithm, effectively preserving image edges and texture details.

Abstract:Aiming at the problem that the fixed fusion algorithm can not adapt to the difference between video frames in dynamic scene, which results in poor fusion effect and even failure of fusion, a fusion method of infrared and visible video mimicry based on differential feature driver is proposed. Firstly, three different features of infrared and visible video sequences are extracted respectively. Secondly, the fusion validity of different fusion algorithms is calculated by using the improved fusion validity formula, the fusion validity is weighted by entropy weight method, and then the decision score of the fusion algorithm is obtained, and the optimal fusion algorithm on different video sequences is determined. The experimental results show that the proposed method is better than the optimal fusion algorithms RP and MOD in the whole video sequence, compared with the fixed algorithm mentioned above, the comprehensive objective evaluation index is improved by 59.9249% and 2.7608% , which provides a new idea for infrared and visible light video fusion.

Abstract:In order to solve the problems of SIFT algorithm in stereo vision, such as long time consuming, high degree of mismatch and feature points clustering, a fast feature matching algorithm based on quadtree fusion of sift and k-d tree is proposed. This method uses a fast feature point extraction algorithm combined with adaptive threshold to extract the key points. Because the extracted key points have the phenomenon of clustering, a quadtree structure is proposed and applied to image matching. The improved k-d tree and random consistency algorithm are used to rough match and purify the key points. Experimental results show that the average matching rate of the improved algorithm is 3.35 times higher than that of SIFT algorithm, and the matching accuracy is improved from 86.18% to 97.53%. At the same time, the improved algorithm has more advantages than SIFT algorithm in view angle, blur, illumination and scale change, so the algorithm can meet the requirements of high matching degree, good real-time performance and uniform feature points.

Abstract:A crack extraction method combining improved MASK dodging and K-means clustering algorithms is proposed to address the problem of stains, shadows and uneven illumination in the captured bridge crack images, which makes it difficult to extract crack features at a later stage. The method firstly improves the MASK dodging algorithm, improves the adaptive capability of the algorithm, uses contrast stretching to enhance the image contrast, then uses the K-means clustering algorithm to segment the image according to the difference between the grey value of cracks and background pixels, and finally combines morphological methods and connected domain detection to bridge and denoise the cracks. The experimental results show that, compared with other methods, this method can effectively reduce the influence of image brightness uneven interference on the crack extraction results, and the crack extraction accuracy reaches 95%, ensuring the accuracy of later crack size measurement and bridge damage degree assessment.

Abstract:Aiming at the problems of long time consuming and high error rate of  large multi-repeating unit PCB image stitching, A fast robust image stitching method was proposed. The collected high-resolution PCB images were sampled down, and image units containing overlapping areas were accurately obtained based on manual points as registration areas; The Shi-Tomasi corner detection algorithm was improved by introducing the suppression radius method to make the extracted regional feature points more evenly distributed; The violence matching method was used to carry out the rough matching of the regional feature points respectively and the registration coefficient matrix was obtained after removing the mismatched point pairs by RANSAC algorithm. Combined with affine transformation formula, the registration coefficient matrix of the original image was deduced and calculated. According to the registration coefficient matrix, the stitched images were fused to obtain a complete PCB stitched image. Experimental verification shows: the proposed image matching method accelerates the speed of large PCB image stitching, and also significantly improve image registration precision at the same time, When the image is downsampled 8 times, Compared with the traditional Shi-Tomsi algorithm and Harris algorithm, the improved Shi-Tomasi algorithm improves the matching accuracy by 7.8% and 4.0%, respectively, which proves the feasibility of the proposed method.

Abstract:Aiming at the problem that traditional body size measurement methods cannot achieve remote and convenient measurement, this paper proposes a body size measurement system based on WeChat applet to be applied to remote clothing customization. The design and development of WeChat applet set specific contour acquisition contains three types The posture of the human body image data set, using the u-net network framework for training, extracting the human body contour feature map, positioning the key points of the human joints through the openpose human body pose estimation algorithm, extracting the key parts of the human body to extract the feature points, using the height ratio method and the double ellipse model The fitting body size information is fed back to the user through the WeChat applet. With manual measurement as the standard, the accuracy rate of two-dimensional size measurement in the body size measurement is 97.8%, the accuracy rate of three-dimensional size is 89.5%, and the final average total error is 3.19%. Experiment It shows that the body size measurement system based on WeChat applet can realize convenient body size measurement and remote clothing customization.

Abstract:In this paper in the natural disasters such as earthquakes and shelter space range sensor volume restricted and parameters on the surrounding environment perception problem, a gas sensor for a three-dimensional micro assembly tester, combining with the design of GPSO - BP algorithm, the tester can realize real time under high temperature and high humidity environment to the concentration of toxic gases and pressure of multi-parameter measurement, And the temperature and humidity compensation, with high performance, miniaturization, high accuracy characteristics. The test results show that in the humidity of 80%RH, temperature of 50℃ environment measurement of different gases, after the temperature and humidity compensation, the accuracy of the tester is improved to at least 2.2% above, strong stability, for earthquake and other harsh environment of environmental parameters perception is of great significance.

Abstract:In order to realize the transverse flux induction heating (TFIH) of aluminum tubular oil cooler, and to solve the problem of the erosion of pipe end and low pass rate of the conventional plane coil, a new conical coil is designed to replace the conventional plane coil. The electromagnetic-thermal coupling simulation was carried out by COMSOL software based on variable physical properties, the temperature distribution, heating curve, current density distribution, joule heat distribution of aluminum tubular oil cooler and influences of conical angle on heating effect are analyzed. The results show that aluminum tubular oil cooler has a good temperature distribution under the new conical coil, avoiding the erosion of pipe end and the pass rate is above 95%. The pipe end temperature is monitored by temperature sensor, and the relative error of coupling simulation is less than 4%. The results show that the coupling simulation has a certain accuracy. This study can guide industrial design and production, and provide TFIH coil solutions for aluminum tubular oil cooler.

Abstract:In order to fully explore the potential connection between the rolling bearing fault types and the vibration signal to improve the diagnosis accuracy, a fault diagnosis method based on the hybrid model of scale adaptive convolutional neural network (SACNN) and modified gated recurrent unit (MGRU) is proposed. To begin with, a scale adaptive factor is proposed to obtain appropriate CNN window size for extracting local fault information from the raw signal more effectively, and scaled exponential liner unit (SELU) is introduced into CNN to improve the robustness of its training process. Subsequently, SELU is embedded into GRU to further enhance the network stability and the network structure of GRU is ameliorated to enhance the temporal feature extraction ability, thereby extracting temporal feature from the local fault information more fully. Finally, the softmax function is applied for recognizing fault types. The experimental comparison and analysis reveal that the proposed method achieves better convergence and stability, can effectively mine the fault information contained in the vibration signal for accurately recognizing the rolling bearing fault types at different speeds with the recognition accuracies higher than 99.5%, which has certain application value.

Abstract:The existing high voltage direct current (HVDC) fault detection methods have low sensitivity and are difficult to identify high resistance grounding faults. This paper proposes a HVDC transmission system fault detection method based on Improved Grey Wolf optimizer (GWO) optimized time convolutional network (TCN), The fault current signal collected by the rectifier side detection device is directly used as the input data of TCN, which overcomes the cumbersome process of fault signal processing. The ± 500 kV HVDC transmission line model is established by using Simulink simulation software, and the simulation experiments are carried out for different fault areas and fault types. The fault detection methods based on LSTM model, bilstm model and CNN model are compared. The test results show that gwo-tcn network can reliably and accurately select the fault pole and selection of HVDC transmission line when the transition resistance is up to 800 Ω.