Abstract:In order to efficiently plan the flight path of the UAV in the 3D coverage detection task, a path planning model that meets the coverage requirements is established, which can be divided into two steps: The first step is to determine the viewpoint and line of sight of the UAV inspection, and the second step calculate the collision-free access sequence of viewpoints. First, starting from the 3D point cloud of the inspection target, a method of generating candidate viewpoints based on K-means clustering is proposed, and an incomplete graph model of candidate viewpoint interconnections is constructed. Secondly, a sorting-oriented hybrid ant colony algorithm ( sortingoriented hybrid ant colony algorithm, S-HACO) finds the UAV inspection path, and the optimization goal takes into account the length of the path, the number of viewpoints, the number of sharp turns, etc. The simulation results show that the viewpoint obtained by this method compared with the offset method and random sampling method, the number is reduced by 96. 25% and 42. 10%, respectively, and the performance of the S-HACO algorithm is better than that of the traditional algorithm, the objective function is reduced by 19. 14%, and the running time of the algorithm is reduced by 25. 27%. The effectiveness of the model and the feasibility of the algorithm.

Abstract:Aiming at the fast and accurate requirements of the image processing for the feasible domain and obstacle segmentation system of unmanned surface vessels (USV), an algorithm for fast segmentation of images on water according to the on-board vision sensor of unmanned surface vessels (USV) is studied. Firstly, the experimental images were collected through multiple experiments, and the original database was constructed through data cleaning, image de-duplication, and manual screening. The feasible region and obstacle segmentation data set of the unmanned ship were constructed using the Human-in-the-loop annotation method, with a total of 5 620 images and 25 875 tags. Secondly, it practices the mainstream semantic segmentation methods based on deep learning, including FCN, DeeplabV3 Plus, U-Net. Finally, a fast segmentation network DeeplabV3-CSPNet based on improved DeeplabV3 Plus is proposed according to the characteristics of water images and the requirements of fast segmentation. The results of the network learning experiment, offline navigation experiment, and model deployment results show that the DeeplabV3-CSPNet algorithm achieves a fast and accurate segmentation with an average accuracy of 84. 17% and an operation speed of 49. 26 fps, which can reach 45. 45 fps on the edge computing platform.

Abstract:Aiming at the problem that the reinforcement learning algorithm is not ideal to detect moving obstacles in a dynamic environment, which affects the optimal obstacle avoidance strategy. A state predict error-intrinsic curiosity module (SPE-ICM) with state prediction error as intrinsic motivation is proposed to improve the ability of policy functions to explore the environment of Agents. First, the internal reward mechanism is introduced to provide multiple reward (reward) structure for Agent. Secondly, according to the internal and external reward structure optimization, the Agent’s perception of environmental information is improved, the collection and detection method of moving obstacles on the data structure is improved, and the optimal obstacle avoidance strategy function is optimized and improved by relying on the new detection method. Finally, the network model is combined with the deep deterministic strategy gradient algorithm (DDPG), and the comparative experiment is carried out in the path planning simulation environment built by ROS to verify the feasibility of the proposed algorithm. The results show that the proposed algorithm has significantly better effects in detection ability and decision-making ability.

Abstract:Aiming at the problems of aimless search, slow convergence and unsmooth path planning of ant colony algorithm in global path planning, this paper proposes a smooth path planning method that combines A ∗ ant colony and dynamic window algorithm. First, for the traditional ant colony algorithm, the improved A ∗ algorithm is used to distribute initial pheromones unevenly to solve the aimless problem of initial search of the algorithm. The self-defined moving step size and searching method are given to improve the efficiency of path optimization. The heuristic function value in the transition probability function is modified and the obstacle influence factor is added to avoid deadlock and speed up the convergence. The secondary path optimization strategy is adopted to make the path shorter and smoother. Secondly, the dynamic obstacle avoidance evaluation sub function is introduced into the evaluation function of the dynamic window method to improve the path safety. The simulation results show that the improved A ∗ ant colony algorithm can reduce the path length by 8. 75% and the turning points by 59% compared with the traditional ant colony algorithm. After the dynamic window method is integrated and optimized, the mobile robot not only ensures the global optimal path planning in the static environment, but also realizes the path planning in the dynamic environment, effectively avoids dynamic obstacles in the environment.

Abstract:The unmanned target vehicle is the target of all kinds of guided weapons in the experiment and test stage. In order to test the hitting accuracy of guided weapons under the conditions of different driving speeds, the unmanned target vehicle should be able to complete path selection and path planning under the conditions of different speeds. In order to solve this problem, this paper proposes a path planning algorithm for unmanned target vehicle which can be applied to different speed conditions. This algorithm is improved on the basis of the traditional A ∗ search algorithm. Combining the heading angle constraint and arc search method, and establishing the corresponding relationship between vehicle speed characteristics and search step length. This improved A ∗ algorithm is more in line with the characteristics of vehicle driving. In order to verify the effectiveness of the algorithm optimization, MATLAB was used for simulation verification. The results showed that: The improved A ∗ algorithm in this paper meets the steering limit of unmanned target vehicle, it can solve the problem of path planning at different speeds (30, 45, 60 km/ h). The experiment on remote sensing map proves that it can provide a new solution for the path planning problem of unmanned military target vehicle at different speed.

Abstract:Aiming at the problem that the inaccurate pose calculation of the lidar odometry when mobile robots build the map in the outdoor open environment, which will make the accuracy of the simultaneous SLAM algorithm drops, an optimized SLAM algorithm based on multi-sensor fusion is designed. In terms of algorithm, the reliability of the SLAM algorithm is improved by optimizing the front-end odometry, the data of the lidar odometry is integrated with the data of several sensors which are suitable for outdoor use, such as GNSS, we achieve the lightweight of the extended Kalman filter and embed it in the LOAM algorithm technically, and improve the lidar odometry without increasing computing resource as much as possible. Based on the optimization algorithm, an actual mobile robot platform is built and the algorithm has been transplanted on it, the hardware solution of multi-sensor fusion and the method of processing extended Kalman filter in practical engineering are realized. The experimental results in real scenes show that the algorithm can be stably maintained at 10 Hz outdoor mapping after increasing the odometry calculation, and it is reliable and feasible in complex open environment and lowcost conditions in real scenes.

Abstract:In view of the slow detection speed of the current algorithm of the target detection system for Chinese college students’ driverless formula racing cars, the low detection accuracy and serious missing and false detection in different scenarios are easy to occur. In the recognition module, first of all, in order to improve the detection speed and recognition accuracy of the original YOLOv5 basic model, the paper uses CIoU as the boundary box regression loss function. To solve the problems of slow convergence speed and low recognition accuracy of the algorithm during training, the original weighted nonmax suppression method is changed to DIoU_NMS in this paper, the test accuracy is 0. 963, which is 2. 1% higher than the original algorithm. The results show that the improved algorithm is more suitable for cone color recognition in the competition scene. Secondly, in the tracking module, the depth apparent feature cone color recognition model is trained, and the single target tracking algorithm is changed to be able to track multiple types of targets. Compared with a single target detection algorithm, the phenomenon of missing detection is effectively reduced. Finally, the ranging module is added to use the height information of the detection frame to distance the vehicle camera to the cone barrel. The average error within 90 meters is less than 9%. The frame rate of the whole system reaches 20 frames/ second, realizing cone color recognition and effective distance measurement, and providing more data support for the game.

Abstract:Multi-sensor measurement technology is considered to be a very effective solution in surface metrology. Aiming at the problem of modeling and fusion of multi-scale complex data sets, this paper proposes a multi-source point cloud data fusion framework based on Gaussian process. Firstly, a robust point cloud registration method with adaptive distance is proposed to unify coordinate systems of different measurement datasets. Then, by introducing adjustment theory, the residuals between multiple independent data sets from different sensors are approximated, and a Gaussian process model based on Matern kernel function is constructed. Finally, the method is verified by simulation verification and practical application, and a series of comparative experiments with existing methods are carried out to verify the effectiveness of the method. The results show that the method can fuse multi-sensor datasets with higher fusion accuracy and faster computational efficiency.

Abstract:In the working process of MEMS capacitive devices, the capacitor plates are usually not in the very opposite position but in an inclined position, which will induce the fringing field effect and cannot be ignored. Based on conformal mapping transformation and complex function correlation theory, the fringing effect model is proposed by inheriting the traditional models and modifying them. Compared with the traditional Heerens’ and Huang’s model, the proposed model is better because the error between the proposed model and the finite element simulation, which is from 10% to 20% when the capacitor plate is moved from perfectly aligned to completely misaligned, is less than that between the traditional Heerens’ s and Huang’ s model. Furthermore, according to the proposed model, when the plate coincidence degree is lower than 40%, the fringing effect increases rapidly. As a result, the proposed model should be applied to correct the capacitance. All above are verified by a digital experiment concerned with MEMS differential capacitor array. The research is helpful for the design and performance analysis of capacitive MEMS devices.

Abstract:Focusing on the problem that traditional LVDT measurement circuit requires many high-cost modules and that the performance is relatively low, a LVDT measurement system based on oscillation circuit is proposed and designed. LVDT coil is connected to a highperformance LC oscillator as a resonant inductance. Displacement is obtained by using the frequency shift of the resonance signal. System linearity is improved with the help of the resonator mathematical model and theoretical analysis. The experimental results of the prototype show that under the full displacement state, the frequency range span can reach 73 ~ 122 kHz. The sampling rate and resolution can be adjusted according to the actual needs. A submicronic measuring resolution of 16. 7 bits is achieved when the sampling rate is 200 Sps and the nonlinearity error is 0. 14% FS. The measurement system outperforms the traditional bridge circuit in most aspects, and does not need low distortion excitation and high-precision demodulation module. Besides, it has significant advantages in circuit cost and size.

Abstract:To address the problem that the time-series characteristics of vibration data are lost in the process of feature extraction in the prototype network, and the distribution of samples in the metric space is not corrected which results in low model accuracy under few-shot task, this paper proposes a new boundary-enhanced prototype network with time-series attention for gearbox fault diagnosis. First, the time-series fusion features of the channels are obtained by building a time-series attention module to establish the time-series feature dependencies between channels. Then, after calculating the class prototypes, the near-neighbor boundary loss is added to correct the intra- and inter-class distributions of the fault features in the metric space to clarify the representation boundaries of the class prototypes. Finally, the fault diagnosis results are output by calculating the Euclidean distance between the test sample and the class prototype. The experiments show that the proposed method in this paper has higher fault diagnosis accuracy compared with other methods under small sample conditions.

Abstract:In using generative adversarial networks to generate oil well production parameter data, this method causes the inconsistency between partially generated data characteristics and characteristics of oil well production process, which leads to the low quality of soft sensor modeling of dynamic liquid level. This paper presents an expansion method of soft sensor modeling data of oil well dynamic liquid level based on expert diagnosis-wasserstein generative adversarial networks. After the discriminator obtains the original loss value based on the real data and generated data, the rationality of the generated data is diagnosed by the expert diagnosis module in combination with the mechanism process of oil well production, and the discriminator judgment results are detected. The error results are compensated and added to the loss functions of the generator and discriminator for subsequent confrontation training, thus the better soft sensor modeling sample data of dynamic liquid level which consistent with the characteristics of oil well production process is generated. Through simulation experiments, the prediction accuracy of the dynamic liquid level improved by adding the generated data to the training data of soft sensor modeling, and the root mean square error is reduced by 5. 99%. It shows that the data generated by the generator after adding the expert diagnosis module has higher quality and can better meet the production needs of the oilfield.

Abstract:Underwater images usually have problems such as low contrast and color imbalance, which lead to unclear image texture information. Aiming at finding a solution, an underwater optical image sharpening method based on fusion of channel quantization and red prior is proposed. First, two input image versions are designed. Image 1 adjusts the image contrast by quantifying the color channel histogram and redistributes the pixel value. In image 2, in order to achieve color equalization, the red channel prior is substituted into the underwater imaging model to estimate the background light, direct component transmittance and backscattered transmittance. Then, three weight maps are designed for each input image, including brightness map, saturation map and saliency map. Finally, the multiscale fusion strategy is used to fuse the image after local contrast enhancement and color correction with its normalized weight map. The experimental evaluation carried out on multiple databases by subjective and objective indicators shows that the proposed algorithm can recover more color and detail information while presenting high contrast, effectively improve the quality of underwater images, and has advantages over other classical and novel algorithms.

Abstract:In response to the problems of information loss and slow computation in previous studies of single-channel EEG signal EOG artifact removal algorithms, a method for removing EOG artifacts based on empirical wavelet transform (EWT), wavelet transform (WT) and approximate entropy is proposed. Firstly, the empirical wavelet transform (EWT) is used to adaptively segment the EEG signal, and the appropriate wavelet filter banks are constructed in the segmentation interval to extract the tightly supported modal components. Then, the WT decomposition is performed for each modal component, and the approximate entropy of the decomposition is calculated, while the approximate entropy threshold is set for automatic identification and removal of EOG artifacts. Finally, the signal is reconstructed using the inverse transform of wavelet transform (WT) and empirical wavelet transform (EWT). The algorithm was tested using the publicly available Klados dataset and Mohit Agarwal’ s EEG-VR dataset, and the experimental results showed that the mean value of the computation time of the method was 0. 199 5 s, and the mean value and variance of the power distortion of the Alpha wave were 0. 128 4 and 0. 151 1, the mean value and variance of the power distortion of the Beta wave were 0. 097 7 and 0. 158 0. Compared with EMDICA, CEEMDAN-ICA and WT algorithms, the proposed algorithm has faster computation speed, better artifact removal ability, and can retain more useful information of EEG signals.

Abstract:Flow measurement in shallow water is of great significance for hydrological observation in intracontinental aera, which can promote the development of flood control irrigation and other water conservancy projects. The broadband acoustic Doppler current profiler (ADCP) can accurately monitor the information of the flow velocity in the water area. The broadband ADCP used for shallow water flow measurement is not easy to achieve better velocity accuracy in the narrow flow layer and is vulnerable to local turbulence interference in the narrow flow layer. To solve the above problems, this paper proposes a retrieval method of the optimal value of the solution results based on η criterion, optimizes the local correlation evaluation method, and completes the development of the ADCP system based on system on chip (SoC). The effect of the algorithm is further verified by the radial velocity generated by the beam-velocity angle. The outdoor sailing experiment is carried out. When the radial layer thickness is 6. 3 cm with turbulence interference, the velocity precision and degree of accuracy can be improved by 37% and 77% respectively after the algorithm is applied, 66% and 72% respectively in the case of low SNR. The error between the angle calculated according to the speed measurement value and the set value is 3. 06°, which is reduced to 2. 60°after algorithm processing. The angle error obtained by the error relation of measured speed can reach 0. 6°. In the boat experiment in a calm lake, the standard deviation of the system can reach 1 mm/ s, and the error of interlayer velocity can reach 1 mm/ s, which is close to the theoretical velocity distribution.

Abstract:Due to wireless monitoring node rotates with shaft to monitor the film pressure of water-lubricated bearing, the power supply for wireless node is difficult. The continuity with stability of wireless monitoring node cannot be realized by common power supply methods, such as battery-powered, slipring-powered, self-powered, etc. For this reason, a contactless power supply method is proposed, and the coupling mechanism of coupler is studied. First, in order to obtain the relationship between coupler and electrical parameters, the equivalent magnetic circuit model of coupler was established by magnetic flux path. Second, the influence of core numbers, gap length, radial offset, and rotation to coupling performance were analyzed by FEA, the capacitor is used to compensate the coupler circuit, and the output power before and after compensation was compared by co-simulation. Finally, the coupler was installed on water-lubricated bearing test-rig, and the dynamic test was conducted. The research results show that the simulation and test have favorable consistency, the coupler has well coupled and offset resistance performances when the gap length is 5 mm, meanwhile, it is not affected by rotation condition. After compensation, the average output power of coupler is 12. 496 W with a transmission efficiency of 71. 96%, which can meet the needs of wireless monitoring node.

Abstract:In order to enhance the safety as well as comfort of multiple high-speed trains in cooperative operation, a cooperative cruise control strategy of multiple high-speed trains based on improved distributed consensus algorithm is proposed. Firstly, the safety distance interval is designed that can change with the train speed in real time. Secondly, the distributed cooperative control strategy is given by combining weighted hyperbolic tangent function. Meanwhile, the error model of the closed-loop train dynamics is transformed into a typical second-order multi-agent system model by coordinate transformation, and the asymptotic stability of the closed-loop system under the error state is proved by using Lyapunov’ s stability theorem. Finally, the proposed method is verified by simulation, the proposed method can dynamically adjust the safety distance between trains and always keep it within [-0. 7,0. 7]m/ s 2 . The results show that the safety and comfort of multiple trains tracking operation can be guaranteed.

Abstract:Aiming at the problems of low positioning accuracy and high cost of traditional LiDAR in the process of loading and unloading of automated container terminals, a vision-based three-dimensional measurement system for container attitude is proposed. Firstly, through a small - scale deep learning network for rapid coarse positioning container corner, secondly, the traditional image processing algorithm is used to reposition the container corner pieces to obtain the precise position of the container keyhole, and the threedimensional measurement of the container posture is carried out in combination with the physical movement of the container during the loading and unloading process. The experimental results show that compared with the deep learning network before improvement, the measurement accuracy is higher and the measurement speed is faster, the measurement accuracy of the overall algorithm is 93. 71%, about 12. 45 frames/ s, and the average measurement error of container attitude measurement is about 4. 95%, which meets the requirements of automatic loading and unloading.

Abstract:In the detection of massive multiple-input multiple-output systems, the minimum mean square error algorithm can obtain approximately optimal detection performance, its complexity is very high and cannot guarantee the real-time detection of the signal. An improved Richardson signal detection method is proposed, which uses the steepest descent and the whole-correction method to improve the performance of the Richardson algorithm. The steepest descent can provide more efficient search paths and obtain different approximate solutions, in order to improve the accuracy of the algorithm, the whole-correction method is used to modify the different approximate solutions, so that the convergence speed is faster, and the complexity of the algorithm is reduced from the order of magnitude O(K 3 ) to O(K 2 ) . Simulation results show that the proposed algorithm can approach MMSE with only 3 iterations, which reduces the complexity and improves the BER performance.

Abstract:Laser interference can cause interference such as saturation and glare to thermal imager, it seriously affects the image quality. Vanadium dioxide is a phase change material, which has high transmission of infrared radiation at low temperature semiconductor state. In the high temperature metallic state, vanadium dioxide has highly reflectance of infrared radiation. This characteristic can be used to protect infrared thermal imager from laser interference. Self-motivated phase transition vanadium dioxide thin films were prepared by molecular beam epitaxial method, and the transmittance of thin films to mid-infrared laser was measured by an experimental device. The transmittance of the film in semiconductor state is 0. 693 and 0. 069 in metallic state at 3. 525 μm wavelength. An experiment using the vanadium dioxide thin film to protect the thermal image from the middle infrared laser interfering was done. The experimental results show that when the vanadium dioxide film is semiconductor state, most of the incident laser energy can transmit the vanadium dioxide film, which will cause serious interference to the thermal imager. When vanadium dioxide film is in metallic state, most of the incident laser energy is attenuated, and the interference degree of laser to the thermal imager is greatly reduced. Vanadium dioxide thin film can be used to protect thermal imager from laser interference.

Abstract:Contactless temperature measurement is an effective means to conduct mass epidemic prevention screening in response to the “COVID-19”, influenza and other infectious diseases, could avoid the risk of cross infection and realize human body temperature monitoring in public places, and can realize human body temperature monitoring in public places. In this paper, a portrait recognition and temperature measurement system are designed based on YOLOv5 and infrared camera. Infrared images are used for face target detection, and a set of forehead region assisted location algorithm relying on face and occluding objects is proposed. For portrait and glasses, masks, hats and other condition to construct the data set, as recognition of infrared image training and forecasting, implements for facial interested area (forehead) precise positioning and temperature measurement, and realize human body temperature measurement by this location, using C # development of the software interface, realize the visualization display and management of infrared image and its temperature. After the experimental test, the average accuracy of the prediction based on YOLOv5 is 94%, and the auxiliary positioning accuracy of the forehead region reaches 97. 3%. The influence of the algorithm on the infrared temperature measurement effect is within ±0. 15℃ . The system can run for a long time and has good applicability to multiple application scenarios.

Abstract:Aiming at the problem that existing object detection algorithms have low accuracy in full-size object detection, this paper proposes an improved full-size object detection algorithm based on the YOLOv3 model. In the method, a new adaptive recursive FPN network architecture is designed, and a recursive pyramid model based on channel attention is proposed to improve the feature extraction ability of YOLOv3 and the detection ability of objects at different scales. At the same time, loss function transformation is introduced in the training process to solve the problem of dynamic parameters that is not being optimized in the training process. Compared with other mainstream object detection algorithms, the accuracy of small-size objects, large-size objects and multi-size objects with complex backgrounds respectively improved by 5. 6%, 2. 6%, and 1. 6%. Experimental results show that the detection accuracy of the proposed method is significantly improved.

Abstract:To speed up the optimization of antenna modeling, an improved one-dimensional convolutional neural network ( 1D-MCNN ) model is proposed. The convolution kernel size of this one-dimensional neural network is 2, and the ReLU function is used as the activation function to reduce the gradient dispersion. The Adam optimizer is combined with dropout technology to improve the feature learning ability and nonlinear function approximation ability of the model. In this paper, the 1D-MCNN model is used to model the geometric parameters of the ultra-wideband microstrip monopole antenna. The eight geometric parameters of the antenna are used as feature inputs to predict the return loss value of the antenna. Experiments show that compared with the deep MLP network model, MLP network model, and RBF neural network model, the average error of the return loss value of the 1D-MCNN model proposed in this paper is reduced by 1. 95%, 120. 27%, and 125. 71% respectively. It has higher accuracy and stronger prediction ability. It is feasible to optimize the modeling of ultra-wideband antennas and has certain advantages.

Abstract:The internal hardware and software of coherent optical transceivers are relatively complicated, and it is hard to carry out a large-scale test with the conventional manual method. The testing efficiency and productivity of coherent optical modules are also seriously restricted. To this end, an automated test system for 400 and 800 Gb / s coherent optical modules is designed and implemented. This system utilizes the programmable manipulation components to efficiently obtain the firmware, state sequence and optical parameters of coherent transceivers. Moreover, an experimental test setup is built to verify the systemic effectiveness, in which multiple 400ZR and 400ZR+ modules from different vendors are evaluated. The test results show that compared to the manual method, the measuring time can be lowered by nearly an order of magnitude with the proposed method, which significantly improves the testing efficiency.

Abstract:In the research of DOA estimation of coherent signals from logarithmic spiral arrays, this paper proposes a VA-MMUSIC algorithm that virtualizes the logarithmic spiral array into a uniform line array, derives the covariance matrix of the virtualized array, then applies the covariance matrix to the MMUSIC algorithm to perform DOA estimation of coherent signals. Simulation results show that the VA-MMUSIC algorithm, is able to achieve DOA estimation of coherent signals, and at a signal-to-noise ratio of 10 dB and signal intervals within 5°, the VA-MMUSIC algorithm is still able to accurately estimate the azimuth of the coherent signal source, with errors always remaining within 0. 5°, verifying the effectiveness of this method. The effectiveness of the VA-MMUSIC algorithm in a practical environment is also verified by using a logarithmic spiral array to receive coherent signal source data under real experimental conditions.

Abstract:In order to improve the state of charge (SOC) estimation accuracy of lead-acid battery under random conditions, reduce the influence of error variation on estimation accuracy. Aiming at the limitation of fixed length selection of error innovation sequence in adaptive extended Kalman filter, an improved adaptive extended Kalman filter algorithm is proposed to estimate SOC. The likelihood estimation is used to monitor the distribution change time of the error innovation sequence in the covariance matching algorithm, and the length of the innovation sequence is adaptively adjusted according to the distribution change of the error innovation, thereby reducing the error when estimating SOC. Firstly, the equivalent model parameters are identified by the recursive least squares method with forgetting factor ( FFRLS ), the average error voltage of the model is 13. 63 mV. Then, in the random condition experiment, it is found that the improved algorithm improves the accuracy of RMSE and MAE performance by 14. 44% and 17. 26% respectively when estimating SOC. The results show that the improved algorithm has better stability and accuracy.

Abstract:Aiming at the problems of poor robustness and are difficult to monitor the settlement value of tunnel timely and effectively. An adaptive recognition and measurement algorithm of tunnel settlement is proposed by combining coordinate attention with object detection algorithm. Using industrial camera to get object images in different environment to build datasets, then training object detection model with coordinate attention, prediction accuracy of the model is 97. 9% in the test sets. Using the target figures and LED lights of the pixel coordinates in images to do the camera calibration and calculate settlement value. The results show that the measurement error of tunnel surrounding rocks is less than 1 centimeter within 25 meters, less than 5 millimeters within 10 meters.

Abstract:A voltage-force-strain conversion novel high voltage sensor is designed by combining the piezoelectric ceramic actuator with a cantilever beam strain transducer through a longitudinal-bending conversion mechanical structure. The piezoelectric ceramic actuator is T-connected with the elastic cantilever beam structure to form the longitudinal-bending conversion structure, which converts the longitudinal stress/ strain generated by the piezoelectric actuator under the action of voltage into the bending stress/ strain of the cantilever beam, and the longitudinal-bending conversion of stress/ strain is realized. Four resistive strain gauges are placed on the upper and lower sides of the cantilever beam to form a Wheatstone bridge circuit and the bending stress/ strain of cantilever beam is measured. The mechanical conversion design between the strong electricity side of the high voltage and the weak electricity side of the force sensitive ensures a high degree of electrical isolation. The physical model of voltage-force-strain conversion of the sensor is established and verified by finite element simulation. The prototype device is fabricated, and the test results show that the measurement range of the sensor is 0~ 5 000 V, the sensitivity is 0. 01 μV/ V, the nonlinearity error is 1. 83%, and the resolution is 30 V.