Abstract:The rapid development of industries such as the Internet of Things, automobile manufacturing, and smart medical care has accelerated the promotion and application of endpointdevice chips, and subsequent chip security issues have also been exposed. Traditional micro control unit（MCU）or ARMA series CPU chips can no longer meet the increasingly complex application requirements.In order to solve the problems of insufficient chip security protection, slow transmission speed, high power consumption, and insufficient computing resources in current end devices, combined with the SoC design concept, this paper proposes a cryptographic SoC design scheme based on highspeed bus.This scheme realizes the acquisition of the dynamic status of the sensors, chips, and hardware of the enddevice, receiving multiple highspeed protocol interface data, encrypted storage, and backup to the cloud.The solution uses an opensource processor to complete a lowpower encryption monitoring chip that combines a processor, a highspeed bus, hardware peripherals, and an encryption unit.Synthesis and power analysis and experimental results show that highspeed and reliable data transmission and encryption are realized to meet the needs of fast encryption and decryption of largecapacity data; low power consumption design is adopted, performance is not affected, and power consumption is reduced by about 20%.

Abstract:With the development of 3D printing technology, its application areas have been extended to clinical medicine, and 3D biotechnologybased additive printing of skin tissue, cellular scaffolds and other tissues and organs has been realized.3D bioprinter generally use permanent magnet synchronous motors as their mobile platform drive motors. Traditional methods usually use multiple algorithms such as genetic particle swarm optimization fuzzy rules to adjust PI parameters to achieve control. However, the mechanical superposition algorithm increases the complexity of the algorithm and seriously affects the performance of the motor control effect. Therefore, this paper adopts fractionalorder PI control instead of traditional PI control, and uses particle swarm optimization to optimize the gain, the number of fractional orders and the adaptive mechanism in the model reference adaptive system in fractionalorder PI to finally obtain the optimal solution. Simulink simulation shows that compared with traditional PI control and methods such as genetic particle swarm optimized fuzzy PI control, the particle swarm optimized fractional order PI improves the motor response speed by 106% and 56%, and the stability by 813% and 60%, respectively, and is suitable for 3D bioprinter mobile platform with high control accuracy.

Abstract:The complex indoor environment is easily affected by the multipath effect and nonlineofsight, which leads to the unreliable RSSI value and affects the prediction performance of SVR model and positioning accuracy of the system. To solve the problem, a SVRPSO algorithm based on error correction and adaptive operator is proposed. This algorithm proposes to use the prediction error of the nearest neighbor reference labels to correct the prediction distance of the measured label, so as to make up for the inaccurate prediction of SVR model due to the unreliable RSSI value. Then, the nonlinear equations of the measured label’s position coordinates is constructed and solved iteratively by PSO algorithm. Aiming at the problem that the standard PSO algorithm is easy to fall into local optimum and the convergence speed is slow, an adaptive operator is designed to improve the inertia weight and learning factor of PSO algorithm respectively. The simulation results show that both error correction and adaptive operator have certain effects on improving the indoor positioning accuracy. Compared with SVRPSO, the average positioning accuracy of the system is improved by 316%. With the same positioning accuracy, the algorithm uses fewer reference tags.

Abstract:Aiming at the problem that the traditional path planning method can not plan the optimal path according to the timevarying characteristics of urban road network weight, a timevarying road network path planning method based on double deep Qnetwork was proposed. Firstly, the urban road network model with timevarying weights is constructed, in which the weights at each time period of the road segment are generated by random functions. Then, the state features, interaction actions and reward functions are designed to model the timevarying weight network path planning problem, and DDQN algorithm is used to train the agent to learn the timevarying weight characteristics of the road network. Finally, the path is planned according to the modeled state features to realize the effective path planning of the timevarying weight network. The experimental results show that the agent trained by DDQN algorithm has better global optimization ability in the timevarying weight road network. Compared with the rolling path planning algorithm, the proposed method can plan the optimal path under different circumstances, which provides a new idea for the path planning of the road network with timevarying weights.

Abstract:Aiming at the problem of low estimation accuracy of Liion battery state of health (SOH), a method based on principal component analysis (PCA) and improved LevenbergMarquardt algorithmdouble Gaussian kernel RBF (ILMDGRBF) neural network was proposed, which realized the accurate estimation of SOH. Firstly, the health indicator (HI) highly related to the capacity decline was extracted, and PCA method was used for dimensional reduction processing to reduce the redundancy between HI. Secondly, a double Gaussian kernel RBF neural network was created, and improved LM algorithm was used to realize the online learning of neural network parameters to establish ILMDGRBF neural network. Thirdly, ILMDGRBF was trained with the enhanced battery test data to realize SOH estimation. The verification shows that the principal component 1 obtained by PCA dimensionality reduction can effectively reflect the aging trend of Liion battery, and can be used for SOH estimation; Compared with other models, the established ILMDGRBF model has higher estimation accuracy and better robustness, and the error of the estimation results is controlled within 15%. Finally, based on this method, a new SOH intelligent estimation system was constructed to provide a reference basis for battery safety management.

Abstract:Aiming at the fact that the traditional pedestrian dead reckoning (PDR) algorithm can only be used in a single state of normal walking, which is difficult to meet the practical application requirements, an improved PDR algorithm based on adaptive peak detection is proposed. The algorithm divides the pedestrian motion mode into walking and running states, fully considers the relationship between the peak acceleration and the motion state during the pedestrian movement, obtains the peak acceleration under different motion states through experiments, and sets dynamic thresholds to achieve step detection and step size estimation under different states. The improved PDR algorithm is applied to pedestrian positioning: using the pedestrian motion data obtained by the inertial measurement unit (IMU), the improved peak detection method is used to detect the pedestrian steps and identify the pedestrian status, and the adaptive step size estimation formula is used to estimate the step size according to the pedestrian motion status. Finally, the pedestrian position information is obtained by combining the calculated heading. The experimental results show that the improved PDR algorithm has good robustness and high gait recognition rate. Compared with the traditional PDR algorithm, the closedloop error is reduced by 142%, which effectively improves the accuracy of pedestrian positioning results.

Abstract:The random multiple access protocol is of great significance to the quality of service（QoS）of the vehicular ad hoc networks. Due to the highspeed mobility of the vehicle nodes in the network, the network topology changes frequently. The fixed media access control protocol will limit the overall QoS of the highdynamic ad hoc network. In order to solve this problem, based on the CSMA/CA series protocol backoff algorithm, a competitive window adaptive backoff algorithm (NCWCOCT) is proposed, which is based on vehicle node density, channel occupancy factor and collision threshold. Firstly, in order to minimize the collision probability, a onedimensional Markov model is established based on the vehicle node density and the competition window value, and then the objective function is constructed. Then, the channel occupation factor and the optimal conflict threshold are proposed to realize adaptive backoff of the competition window aiming at optimizing the communication quality of vehicular ad hoc networks. Simulation results show that NCWCOCT is compared with similar DCW algorithm and IMBEB algorithm, the throughput performance is improved by an average of 1164% and 677%; the packet loss rate performance is reduced by an average of 1946% and 1329% respectively.

Abstract:Active reflector technology plays an important role in the development of radio telescopes. In order to improve the realtime surface detection in active reflector technology and reduce the detection errors caused by environmental factors, an active reflector surface monitoring research platform is established. A calculation and detection method of active reflector surface using distance measuring sensor is proposed. First, the active reflector datum surface is determined, and secondly, the panel motion is regarded as rotating motion. Rodrigo rotation formula is used to get the rotation matrix of the panel. Finally, the current position and attitude of the panel are calculated by using the rotation matrix of the panel. The simulation results show that the method is feasible, and the timeconsuming calculation of panel position and attitude at 1 625 nodes is 5 070 ms, which can improve the realtime performance of surface detection, provide reference for improving the operation accuracy of active reflector and achieve better performance index of radio telescope using active reflector technology.

Abstract:Subsynchronous oscillation is a kind of abnormal electromagnetic and mechanical oscillation which occurs when the equilibrium point of power system is disturbed. Aiming at the problems of noise interference and mode aliasing in the extraction of subsynchronous oscillation components by the Hilbert Huang transform, a method combining multisynchrosqueezing transform (MSST) and Hilbert transform is proposed to identify subsynchronous oscillation parameters. Based on Fourier synchronous compression transform, the frequency spectrum of subsynchronous oscillating signal is compressed synchronously for several times, so as to improve the reconstruction accuracy of signal timefrequency distribution and the degree of energy aggregation. Through simulation and verification combined with actual engineering recording data, firstly, the signal timefrequency analysis was carried out using the multisynchronous compression transform method to obtain the signal timefrequency diagram, and then the multisynchronous compression transform transform inverse transformation decomposition was used to reconstruct each modal component, and finally the extracted single modal component parameter identification was carried out using the Hilbert transform. Identify its frequency, damping ratio, attenuation factor and other major parameters. The simulation results show that compared with shorttime Fourier transform (STFT) and synchroextracting transform(SET) and Fourierbased synchrosqueezing transform (FSST). MSST can improve the energy concentration degree and reconstruction accuracy of signal timefrequency distribution, and realize multicomponent subsynchronous oscillation mode decomposition. The actual data show that the method can overcome the noise interference and mode aliasing effectively, identify the subsynchronous oscillation parameters accurately, and has certain reference significance for the safe and stable operation of power system.

Abstract:Water supply pumping station is a variable frequency and constant pressure system to ensure water pressure stability, which can respond quickly and stably under the disturbance. In order to improve the stability and efficiency, this article proposes an improved PSOBPNN adaptive PID control method. Firstly, the variable frequency and constant pressure control model of water supply pump station is established. Then, PSO weight iterative strategy based on BPNN is proposed to improve optimization efficiency of PID parameters, so as to preferably satisfy the control requirements. The results show that the proposed algorithm responds quickly without overshoot, and shows stable tracking capability for nonlinear signal. Compared with the PID control algorithms of BPNN and PSOBPNN, the proposed algorithm can shorten the regulation time by 296% and 28% in the constant voltage experiment, and the overshoot is reduced by 346% and 199%, and the stabilization time is shortened by 73% and 16% in the disturbance experiment. The algorithm can improve the stability and efficiency of water supply pumping station system.

Abstract:Cognitive radio (CR) nonorthogonal multiple access (NOMA) network (recorded as CRNOMA) is one of the research hotspots in the field of wireless communication. Due to the openness of wireless channel, the security communication of wireless network becomes an urgent problem. For the spectrum sharing multiantenna CRNOMA network, First, the secrecy outage probability (SOP) of the optimal antenna selection (OAS) scheme is analyzed, and its exact closedform expression is derived. Secondly, the SOP of spacetime transmission (STT) scheme is analyzed, and its exact closedform expression is derived. Then, the asymptotic secrecy performance of secondary user SOP is analyzed to reveal the influence of system parameters on the secrecy communication performance. Finally, the Monte Carlo simulation is used to verify the correctness of theoretical analysis. It is found that there is an optimal transmit power in the base station under two schemes, which makes the secrecy outage performance of secondary user the best. The simulation results show that increase the peak transmission power of the base station, the secrecy outage performance of close secondary user under OAS scheme is always better than that of STT scheme, the secrecy outage performance of remote secondary user will be inferior to that of STT scheme.

Abstract:Pedestrian detection is an important branch of deep learning object detection field, but there are serious occlusion problems in dense scenes, which brings great challenges to pedestrian detection. To alleviate this problem, a task alignment method for target detection and attitude key point detection was proposed on the CenterNet multitask learning model, and the improved model was Center_tood. Firstly, the separation module is proposed. This module separates the original features into the features that pay more attention to each task. On this basis, a task alignment method is proposed: the alignment measurement is designed to constrain the loss, so that the model can optimize towards the direction of multitask alignment to a greater extent on the gradient. At the same time, the consistency constraint is used to make the model learn the common information between different tasks, so as to align the features of different tasks. In the experiment part, CrowdPose data set was used for training and testing. The AP value of the proposed algorithm is 743%, which increases by 115%. The key point AP value of human posture was 558%, which increased by 96%. Experimental results verify the effectiveness of the proposed multitask learning algorithm in pedestrian detection in dense scenes.

Abstract:Nighttime vehicle detection is of great significance to the safety of unmanned vehicles. At night, low light intensity makes the geometric characteristics of a vehicle inconspicuous. Moreover, a remote vehicle is even difficult to be recognized due to its small size, thus resulting in a significant increase of difficulty in its detection. In this context, this paper proposes an algorithm on nighttime target detection for unmanned vehicles based on an improved YOLOv5s model. To begin with, some night scenes concerning roads in Yulin City are collected for dataset construction. The data is then enhanced by Retinex algorithm. On this basis, the following three measures are made to improve the traditional YOLOv5s network: introducing depthwise separable convolution into the Backbone structure to reduce the number of network parameters; combining multiple attention mechanisms with the FPN structure to improve the ability of feature extraction of the network; embedding dilated convolution into the PAN structure to reduce the number of network parameters, as well as the loss of feature information, while keeping the receptive field unchanged at the same time. The final experimental results demonstrate that the average accuracy of nighttime vehicle detection reaches 848%, which is 52% higher than before. The corresponding detection speed is up to 48 frames per second, an increase of 91%. The research results can lay a theoretical foundation for improving the driving safety of unmanned vehicles during accidentprone nights.

Abstract:JDE algorithm in multiobject tracking jointly learns target detection and reidentification for the first time, which greatly improves the tracking speed.However, the tracking accuracy is reduced due to the poor tracking effect caused by complex background interference and occlusion processing.In order to balance the tracking speed and accuracy, SAMJDE is proposed in this paper. This model integrates SimAM attention mechanism, multiscale fusion and other ideas to improve the accuracy of target tracking by enhancing the ability of feature extraction. CIoU_Loss is used as the regression loss function to improve the positioning accuracy by accurately building the position relationship between the target box and the prediction box.In the association matching part, Kalman filtering is used to predict the motion information, and the Hungarian matching algorithm completes the target association in the time series dimension. Testing on MOT16test dataset shows that MOTA reaches 664% and tracking speed is 206 FPS. On the basis of ensuring realtime performance, tracking accuracy is 23% higher than JDE algorithm, which better optimizes the balance between accuracy and speed.

Abstract:Quantitative precipitation prediction based on radar echo extrapolation has broad prospects.It’s important to get accurate nowcasting.To this end,we propose GLnet,an efficient neural networksbased on Unet and SwinTransformer architecture equipped with two different attention modules CBAM and Nonlocal. The model has an asymmetric twoway feature extractor. In this way,the GLnet model extracts local and global features of radar echo images through convolution and windows selfattention mechanisms respectively.We create two datasets, NL20 and NL50, in Netherlands Precipitation Dataset by filtering the original precipitation dataset and choosing only the images with at least 20% and 50% of pixels containing any amount of rain respectively. We evaluate our approaches in NL20 and NL50. The experimental results show that compared with the classical model Unet，the mean square error is reduced by 144% and 106% respectively.

Abstract:Most of the existing image thresholding methods are only suitable for processing the images with a specific gray level distribution. To deal with the issue of threshold selection in different gray level distribution within a unified framework, an automatic thresholding segmentation method guided by maximizing Pearson correlation is proposed. This method first performs edge detection on the original image to generate a reference template image; then it performs contour extraction on the binary images obtained by different thresholds to generate the corresponding contour images; it finally utilizes Pearson correlation coefficient to measure the similarities between different contour images and reference template images, and the threshold corresponding to the maximal similarity is selected as the final segmentation threshold. The proposed method is compared with 3 newly proposed thresholding methods and 4 nonthresholding methods. The experimental results on 4 synthetic images and 50 realworld images with different gray level distribution show that, compared with the second best method in segmentation accuracy, the proposed method is reduced by 0140 3 and 0121 5 in terms of the average misclassification error on the synthetic images and the realworld images, respectively. The proposed method has no advantage in computational efficiency, but it has more flexible segmentation adaptability to images with different gray level distribution patterns, and can obtain segmentation result images with higher accuracy.

Abstract:Due to the complexity and uncertainty of noise in acoustic environment, the traditional multichannel speech enhancement algorithm has insufficient noise suppression effect, resulting in a poor auditory experience. To solve this problem, an improved TFGSC and improved post filter speech enhancement algorithm was proposed in this paper. The algorithm used the maximum likelihood method to obtain the power spectral density of the target speech signal and noise signal, and then proposed an improved TFGSC which used the variable step normalized least mean square algorithm obtained by the signal power spectral density ratio. An improved optimally modified log spectral spectrum estimator was also proposed using the estimated speech presence probability by combining the signal power spectral density ratio and a priori signal to noise ratio. The simulation experiments in different SNR environments show that the algorithm proposed in this paper can effectively filter coherent noise and incoherent noise. Compared with other algorithms, the enhanced speech signal has higher SNR and speech quality.

Abstract:Aiming at the problem that the radio frequency fingerprint(RFF) extracted by convolutional neural network(CNN) is easily interfered by the channel fingerprint, resulting in a sharp decrease in recognition accuracy. An IEEE80211a signal radiation source identification method with channel fingerprint removal was proposed. Firstly, extract the timedomain training sequence of the frame head of the signal to be recognized, and the timedomain training sequence is used as the reference signal. Then use the LMS adaptive filter and timedomain training sequence for channel equalization and compensation. Finally, IQCNet model is used to extract the RFF from the time domain signal for device identification. The experimental results show that the recognition rate of 6 wireless routers based on IEEE80211a protocol reaches up to 96% in different wireless channel environments. The proposed method can effectively remove the negative influence of channel fingerprint on RFF identification.

Abstract:Partial discharge (PD) is a hidden danger to the stable operation of the power grid, and it is necessary to carry out realtime and accurate distributed online monitoring of PD of cables and electrical equipment. In order to solve the problems of poor noise reduction effect, high consumption of arithmetic resources, slow noise reduction speed and poor adaptivity in traditional PD signal noise reduction algorithms, a noise reduction algorithm for PD signals based on the gray wolf algorithm optimized variational modal decomposition (GWOVMD) is proposed. The algorithm firstly uses GWO to adaptively select the VMD decomposition parameters k and α to obtain the decomposed modal components; then selects and reconstructs the modal components according to the minimum envelope entropy; finally uses the adaptive threshold wavelet function to process the decomposed and reconstructed PD signal, achieving fast and effective adaptive noise reduction of PD signal. In this paper, the theoretical PD signal and the measured PD signal are simulated and processed for noise reduction. The experimental results show that the proposed GWOVMD algorithm has significantly improved the noise reduction effect, arithmetic resource utilization and noise reduction speed, which can provide a useful reference for the optimal design of edge computing of partial discharge online monitoring system based on power IOT technology.

Abstract:Rail corrugation with multiple wavelengths is often mixed on subway lines, and the current rail corrugation identification method is mainly suitable for rail corrugation with a single wavelength. Aiming at the problem of rail corrugation identification of mixed wavelengths, this paper proposes a multiwavelength rail corrugation identification algorithm based on ensemble empirical mode decompositionindependent component analysis. Firstly, the vehicletrack coupling dynamics model and the rail corrugation excitation model are established, and the vibration acceleration signal of the axle box under the action of the mixed wavelength rail corrugation is obtained through dynamic calculation. The ensemble empirical mode decomposition is performed on the calculated vibration acceleration signal of the axle box. Introduce the correlation coefficient to screen the qualified eigenmode components, calculate the energy average value of the selected eigenmode components, determine whether there is rail corrugation by setting the energy threshold, and finally selected eigenmode components and the source signal are reconstructed into a multidimensional signal, and the reconstructed multidimensional signal is used as the input matrix of the independent component analysis to solve the underdetermined problem of the independent component analysis, The center frequency of the positioning separation results determines the rail corrugation wavelength. In order to better verify the algorithm in this paper, the vertical vibration acceleration signal of axle box and the line irregularity level under the wave and wear excitation were collected on a subway line in Guangzhou, and the experimental data were analyzed using the algorithm of this paper. The results prove that under the mixed excitation of two different corrugation wavelengths of 16 and 315 mm, the method can still identify two different corrugation wavelengths, while the traditional wavelet packet energy entropy method and EEMD energy entropyWVD method can only identify corrugation with a wavelength of 16 mm with obvious vibration characteristics, in other words, these two methods cannot be applied to the problem of mixed wavelength corrugation identification. The research results of this paper provide theoretical support for the identification of rail corrugation with mixed wavelengths in subways.

Abstract:With the infrastructure construction project to the sea, the shipborne concrete batching plant has been widely used, but its compound movement of rising and sinking, horizontal rocking, horizontal and vertical under wave excitation will occur, which will lead to deviation of the metering system. Based on this, an error suppression algorithm of shipborne load cell based on periodic sliding average Kalman filtering is proposed. Firstly, the original data is processed by traditional Kalman filtering to eliminate random errors. Then, the spectrum analysis of the data is carried out by shorttime Fourier transform to obtain the frequency characteristics of periodic error. Finally, periodic errors in the system are eliminated by sliding window mean filtering. Through the sixdegreeoffreedom experimental platform, the movement of the ship in the presence of wave excitation is simulated and weighed by measuring and weighing through the threepoint scale, and the weighing data processed by different algorithms are recorded separately. The experimental results show that the maximum error of the original weighing data is 96%. The maximum error of the weighing data processed by the Kalman filter is 21%. In this paper, the maximum error of the weighing data processed by the algorithm is 03%, which can effectively eliminate the periodic error caused by periodic wave excitation and the random error generated by the sensor itself, and improve the measurement accuracy of the shipborne concrete batching plant.

Abstract:In the field of EMC testing, there is a lack of magnetic field immunity evaluation methods suitable for digital magnetic isolators. To solve this problem, the following studies were carried out. Firstly, the working principle of the digital magnetic isolator is studied. Based on the principle of electromagnetic compatibility and IEC standards, a set of evaluation methods for the magnetic field immunity of the digital magnetic isolator based on the GTEM small chamber method is established. The test system was built, the circuit board was designed, the test seat method was used to solve the problems of incorrect direction and insufficient strength of the magnetic field coupling in the disturbance rejection test, two failure modes and criteria of level fluctuation, and fixed 0/1 were defined, and the comparison table for structural analysis was designed. Second, the case of the GL1200P digital magnetic isolator was verified to verify the feasibility of the test method, and the device’s antidisturbance failure sensitive frequency was determined to be 113 MHz, and the rating was 400 V/m—C. The case proves that this method can evaluate the immunity of digital magnetic isolators.

Abstract:Switch machine is an important equipment to realize turnout conversion on the railway. Its operation and maintenance takes a long time, its fault identification accuracy is not high, and there are many problems such as misjudgment, omission and soon. To solve the above problems, this paper proposes a new fault recognition method for S700K switch machine based on artificial intelligence, deep learning and other new technologies. Compared with the traditional Harr or Mexicanhat wavelet decomposition, in this paper, the power curve data sampled by the microcomputer monitoring system is decomposed and composed by an orthogonal wavelet Daubechies wave with tight support, and the feature vectors of eight common types of faults are extracted, which are normalized as the input of the improved wavelet neural network. Then, the IPSOWNN fault recognition model is constructed by using the classification learning particle swarm optimization algorithm to optimize the weights and thresholds in the network. Finally, the action power curve in the station monitor data base is selected for network training and testing of the fault identification model. The algorithm proposes in this paper has a fault identification accuracy of more than 95% and takes only about 21 seconds on the 8 common fault of switch machine. It can be effectively applied to the fault identification of S700K type switch machine and improve its accuracy and speed, providing theoretical support for the prediction of fault identification of switch machine.

Abstract:The state of health (SOH) of the lithiumion battery is evaluated to provide an important reference for battery safety, charge and discharge control, heat management and other functions. Regional capacity analysis (RCA) based on capacity incremental analysis (ICA) is proposed, and the concept of regional voltage and regional capacity is introduced. The ICA analysis of the lithium iron phosphate (LFP) battery module charged voltage data at different magnifications, extracted the highest peak value of the IC curve and the regional capacity of RCA as health factors, and established mathematics model between health factors and SOH. The results show that the goodness of fit (R2) of the linear relationship between the highest peak and SOH is 0815 4 in the charging stage and 0874 1 in the discharging stage when the chargedischarge rate is 1 C, while the R2 of the linear relationship between the regional capacity and SOH is 0984 2 in the charging stage and 0957 6 in the discharging stage; When the chargedischarge rate is 2 C, the fitting degree of the highest peak as a health factor of SOH in the charging stage is only 0188 4, and the fitting degree of the discharging stage is 0576 7, while the fitting degree of the regional capacity to SOH in the charging stage is 0894 2, and the R2 in the discharging stage is 0988 2. It can be seen that when the chargedischarge rate is 1 C or 2 C, the regional capacity is better as a health factor to evaluate the SOH of the battery. The research results have important reference value for the evaluation of battery SOH at large current rates.

Abstract:In order to solve the problem of large inclination error of Piezoresistive micro differential pressure sensor at different attitude positions, this paper designs four range micro differential pressure sensors that meet the sensitivity requirements. The results show that the single island membrane structure forms a stress concentration at the edge of the silicon island, and the double island membrane structure forms a stress concentration at the center between the two islands, both cases help to improve sensitivity. The influence on inclination error is reduced through single isolation diaphragm and micro oil filled package design. The results show that the zero point output of sensor is approximately linearly related to the inclination angle, and the smaller the differential pressure range is, the greater the inclination error is, and the inclination error at 2 kPa shall not exceed 094%. This study provides a basis for the design of micro differential pressure sensor and the analysis of its inclination error.

Abstract:Aiming at the problems of low efficiency, high cost, insufficient automatic detection label samples and high missed detection rate in the surface defect detection of ceramic tiles, a selfsupervised learning model is proposed, without a large number of defect samples, the detection and location of common defects on the surface of ceramic tiles can be realized. Selfsupervised learning generates negative samples through sample expansion, and uses distributionaugmented contrastive learning to improve data irregularity and expand sample distribution, thereby reducing the consistency of comparative representation and making the representation feature distribution consistent with the classification target. Based on selfsupervised learning representation, a class of classifiers is constructed to achieve accurate anomaly detection and localization. The experimental results show that compared with the other two advanced methods, under the standard evaluation criterion(AUROC) of anomaly detection, the anomaly detection rate is increased by 371% and 274% respectively; the abnormal location rate increased by 122% and 401% respectively, with more reliable detection performance.

Abstract:n the process of helicopter design finalization flight test, the reducer is a very important component in the helicopter transmission system, and its performance has a great impact on the safety of the helicopter. The gear transmission system of the reducer usually works in high speed, high temperature, heavy load and other harsh environments. At the same time, due to the complexity of the transmission gear itself, it is easy to cause gear tooth structure to break, peel and other failures, which seriously affects the stability and reliability of the helicopter transmission system. In this paper, through the identification of typical faults of the transmission gear of the reducer, combined with the theory of finite element analysis and gear meshing, the finite element model of the reducer gear transmission in the helicopter is established, and the stress distribution and change under the typical fault of the transmission gear are analyzed, which provides a new idea for the fault monitoring and testing technology of the helicopter transmission system.