Abstract:In order to improve the feature extraction ability of the fault signal of diesel engine water pump cover and diagnose the fault type quickly and effectively, a fault diagnosis method combining robust local mean decomposition algorithm (RLMD) with BP neural network optimized by BAS algorithm is proposed. Firstly, the collected signal sequence is denoised by wavelet threshold and RLMD, and then the signal components (PF) with high similarity with the original signal are screened out according to the Spearman correlation coefficient. Then, the wavelet energy entropy and wavelet singular value entropy of each component are calculated as fault features. Finally, the BP neural network optimized by BAS is used for fault diagnosis and fault pattern recognition. At the same time, compared with neural networks optimized by GA-BP and PSO-BP. The results show that BAS-BP is superior to PSO-BP and GA-BP neural network in all aspects, and the fault classification accuracy of BAS-BP can reach 98.90%.

Abstract:In order to find out the near-field radiation characteristics of phased arrays, the near field mathematical model for triangular grid arrays is established according to the beam forming principle. The simulation is carried out to verify the model. It shows that the calculation is in good agreement with the simulation because the maximum deviation for the near field intensity from phased arrays is about 1dB on the main beam. Then the model is used to analyze the near field distribution characteristics of phased array. In the region where the main beam is not formed in the near field, the beam shape has a few peaks and troughs. The transmission direction hovers near the direction pointing to each observation point when the field intensity on the normal line in the region reaches maximum value. The peaks and troughs converge and the beam width gradually narrows as the distance increasing. Also the hovering ranges of the transmission direction decreases slowly and finally it is consistent with the actual beam direction after the main beam is formed. The conclusion is also available for the rectangular grid phased arrays. The research provides a theoretical basis and guidance for electromagnetic compatibility design of systems and radiation near-field measurement for phased arrays.

Abstract:In the treatment of thrombotic diseases, extracorporeal ultrasound thrombolysis breaks through the limitations of conventional treatment and in vivo interventional thrombolysis and shows great advantages.However, the ultrasonic transducer used for extracorporeal thrombolysis has low transmitting power and is easily influenced by the environment, which makes it difficult for the ultrasonic signal to reach the lesion area through human tissues and to achieve the destruction of the thrombus structure.In order to solve this problem, a new model of extracorporeal ultrasonic thrombolysis was designed and simulated based on the idea of defective state structure and finite element analysis in this paper.The experimental results show that this structural model can have good acoustic field localization effect and acoustic intensity enhancement effect in the low frequency band part of 10~20 kHz ultrasonic waves, and when the acoustic frequency is rate 14.9 kHz, the structural model has good body penetration effect and has certain possibility to produce structural damage to thrombus in vitro and achieve the effect of thrombus cleaning.These studies are of great importance for the research of efficient and safe in vitro thrombus clearance methods.

Abstract:In order to accurately calculate the discharge energy of high frequency inverter resistance welding power supply， the energy loss in the power supply circuit was quantified.  On the basis of precise modeling， the effectiveness of PSpice physical device simulation model is verified by circuit simulation.  Based on the parametric quantified energy consumption model， the main loss of power circuit is analyzed by fitting integral method， which provides reference for the design of power prototype.  The simulation results show that the finite bipolar soft switching technology can effectively reduce the switching loss of the high-frequency inverter circuit， and the synchronous rectifier circuit loss accounts for 69% of the total power loss. Therefore， the structure and thermal loop of the high-frequency transformer and synchronous rectifier integrated design of the power supply should be considered.

Abstract:Magnetic flux leakage (MFL) technology is suitable for weld crack, porosity and other defects testing. Due to the particularity of weld, magnetization paths and MFL signals under different magnetization directions are different, and adopting appropriate magnetization direction can stimulate stronger MFL signals, thus improving the detection rate of small size defects. Ansys FEM simulation software was used to calculate the MFL signals of weld defects, and the strength and components of the MFL signals of circular holes and longitudinal rectangular slots defects when the magnetization direction was perpendicular or parallel to the weld seams were respectively analyzed and compared quantitatively. On this basis, the variation rules of the MFL signals in different magnetization directions were discussed. A MFL testing platform for welding seams was established for experimental verification. The experimental results show that the magnetic flux leakage signal amplitude of circular flaw is only 18.6% of that of parallel magnetization, and the magnetic flux leakage signal amplitude of longitudinal flaw is only 9.2% to 29.3% of that of vertical magnetization.

Abstract:High In today's highly information-based, highly automated and highly integrated technology era, multi-core cables are increasingly used in large-scale electrical equipment. Multi-core cable will be affected by various external environments, so it is very easy to fail. The performance of multi-core cable has become an important factor affecting the overall reliability of the system. At present, cable detection is always difficult. At present, the main detection means is manual detection. In view of the above problems, this paper designs and studies a circuit equipment conduction test device. In this device, ATmega64 single chip microcomputer is taken as its core control component, CD4051 chip is used to select the conduction test channel, TLP521-4 chip is used to isolate the digital signal, and then the measured conduction result data is transmitted to the single chip microcomputer and transmitted with serial port assistant through MAX232 chip. Finally, the test system is tested to obtain the test results and analyze the results.

Abstract:This paper designs a deep-sea attitude system based on the MPU9250 nine-axis data sensor to solve the problem of deep-sea instrument attitude monitoring, and conducts experimental tests to verify it. The system uses STM32F103 as the main control chip to collect MPU9250 nine-axis data, and quickly calculate the current real-time motion posture through software algorithms. And use the three-axis acceleration data to calculate the inclination angle with the natural coordinate system Z axis. The system is designed with a storage module and a real-time clock module, which can record the movement posture and tilt angle of the equipment in the storage module in real time. And according to the environmental requirements of deep-sea instruments, combined with the finite element design method, design the deep-sea pressure tank. Experimental results show that the attitude system can detect the attitude and inclination of deep-sea instruments with an accuracy of 0.5°, and the pressure resistance can reach a depth of 4000 meters.

Abstract:The path planned by the traditional Grey Wolf Optimization(GWO) algorithm better in the world, but it had the defects of low solution efficiency and easy to fall into the local optimum. Although the path planned by the Artificial Potential Field(APF) algorithm is smooth, there are turbulences and fluctuations in the planned path. The defect of unreachable target. Aiming at the different shortcomings of the two algorithms, the two algorithms are improved respectively, and the two algorithms are merged, and an algorithm that takes into account the global and local characteristics is proposed—Grey Wolf Potential Field Algorithm (GWPFA). First, a new method for establishing feature grid maps is proposed to speed up the determination of feature grids and the establishment of feature grid maps; secondly, by setting the relative distance d of the grey wolf individual and the adjustment factor λ, the parameter a is improved to non-Linear attenuation; again, the concept of node priority is proposed, and the path planning problem is modeled based on this concept; finally, the node that improves the global path planning of the GWO algorithm is used as the temporary target point of the APF algorithm, and the temporary target point is improved as temporary Boundaries, and then local path planning. The simulation results show that in the global static environment, the running time, optimal path length, and turning angle of the GWPAF algorithm are optimized by 224.5s､16.3m and 38.9°C respectively compared with the GWO algorithm; in a local dynamic environment, the GWPFA algorithm is guaranteed The path is optimal while avoiding obstacles successfully. The simulation results verify the effectiveness､feasibility and superiority of the GWPFA algorithm.

Abstract:Content resource popularity prediction is one of the main basis for content delivery network to improve the efficiency of caching and scheduling. In view of the poor feature representation ability, adaptability and low accuracy of current popularity prediction algorithms, this paper proposes a content resource popularity prediction algorithm based on deep learning. The algorithm is better based on a two-way GRU model of the fusion attention mechanism, which can better mine the information contained in the resource access history and its correlation, improve the efficiency and quality of feature extraction, and has a more tolerant generalization ability. The experimental results on different data sets show that the various indicators of the algorithm are better than the existing mainstream algorithms, and the accuracy rates are as high as 96.20% and 98.03%.

Abstract:In order to improve the autonomous navigation accuracy of the detector in complex deep space environment, an improved particle filter algorithm based on second-order central difference method and PCA model is proposed. Firstly, the optimal importance density function is obtained by the second-order central difference filtering method, and the sample points are sampled by the symmetrical proportional sampling algorithm; Then, the principal component analysis method is introduced to preprocess the collected sample set, and the scaling factor is introduced to resample the particle set. Through simulation experiments, this method can greatly improve the particle degradation problem in the particle filter algorithm, make the average error of the tracking system reach 0.429, improve the tracking accuracy and stability of the algorithm, and realize high-precision autonomous navigation in the complex environment of the detector.

Abstract:The voice coil motor (VCM) is mainly used in small-range linear positioning scenarios. It is necessary to ensure that it has high precision and high robustness, and due to the influence of motor parameters and friction, the VCM is non-linear and time-varying. This article analyzes and establishes the corresponding second-order system model from the working principle of VCM. The control strategy adopts the combination of sliding mode control and state observer, and proposes corresponding solutions from the sliding mode surface and the approaching law direction to the chattering problem of sliding mode control. The sliding mode surface adopts complementary integral terminal sliding mode control (CISMC). On the basis of reducing chattering, the integral terminal sliding mode designed by homogeneous theory achieves finite time convergence; the reaching law adopts the power reaching law Combining with the fal nonlinear function, it can reach the sliding mode surface faster; outside the sliding mode control, the state observer (ESO) is used to perform feedforward compensation for system disturbances, etc., and a complementary integral terminal based on the reaching law is proposed. Sliding Mode Control (CISMC). According to the deduced mathematical model, the system simulation model is established in MATLAB/simulink. The simulation experiment shows that the CISMC+ESO scheme can reduce the stability error by up to 75% compared with the complementary sliding mode control (CSMC)+ESO scheme in tracking sinusoidal signals. The full tracking time is reduced by 28.5% at the maximum; when tracking the ramp signal, the steady-state error is significantly reduced, and the full tracking time is reduced by 84% at the maximum. Experiments show that it has a significant improvement in system control accuracy and robustness.

Abstract:In the process of studying the dynamic calibration of the force sensor, it is found that the dynamic response performance of the sensor is obviously affected due to the non-rigid coupling between the force sensor and the base of the test device. Therefore, based on the existing dynamic calibration device of force sensor, the dynamic response performance and optimal mathematical model of force sensor are studied, and three optimized mathematical models based on the original model of force sensor are proposed. Then this paper analyzes and compares the three models through the experimental data, and finds that model 4 is the best of the three mathematical models. Finally, through 27 repeated impact tests and model 4, the characteristic parameters of the force sensor are calculated. These parameters are very important to improve the dynamic response ability of the force sensor.

Abstract:In view of the abnormal and short-term missing of UWB positioning information in UWB/INS integrated location, UWB/INS integrated location method based on improved robust kalman filter and SVR is proposed. In this method, the robust kalman filter (RKF) is improved. The improved IGG3 weight function is used to modify the innovation piecewise in order to reduce the influence of abnormal measurement information on the filtering result. When the UWB signal is normal, the position error is estimated with the improved RKF. When the UWB signal is missing, the position error is estimated with the online training SVR model, and the carrier position information is corrected according to the estimated or predicted position error. The experimental results show that the proposed method can not only reduce the integrated positioning error by 33.33% when the UWB signal is normal, but also online training SVR model can improve the performance of localization algorithm significantly compared with fixed SVR model and can continue to effectively locate when the UWB signal is short-term missing, reducing the integrated positioning error by 29.63%.

Abstract:To reduce the negative impact of inner heat gain of buildings caused by holidays on the accuracy of load forecasting in large central air-conditioning systems, an office building complex in Shanghai Expo Park is taken as the research area, a cooling load forecasting method for large central air-conditioning system based on improved Deep Deterministic Policy Gradient is proposed. New date-related feature named Days from Previous Holiday is introduced, the fully connected neural network in Deep Deterministic Policy Gradient structure is replaced by Long Short-Term Memory neural network, and a load forecasting model based on Recurrent Deterministic Policy Gradient is constructed. The experimental results show that the improved prediction model within Days from Previous Holiday can timely capture the load change trend caused by holidays, effectively improve the prediction accuracy which is 0.951 and the error value is 7.08%.

Abstract:To study the diffusion process of liquefied natural gas tank gas phase leakage, we can get the laws of wind speed, leakage height and obstacle factors on the leakage and diffusion of LNG tank. In this paper, FLACS is used to build the tank leakage diffusion model. With the help of Flowvis, the post-processor of FLACS, the design of visual gas diffusion model is realized. Two-dimensional and THREE-DIMENSIONAL graphics output operation of various variables is performed, and automatic video generation and data analysis are completed. The results are as follows: when the wind speed increases from 2m/s to 8m/s, the maximum downwind diffusion distance of the 1/2LFL cloud increases; when the wind speed is larger than 8m/s, the maximum downwind diffusion distance of the 1/2LFL cloud decreases with the increase of the wind speed. With the increase of leakage height to 8m, the farthest downwind diffusion distance of gas cloud decreases gradually. After 60 s of leakage, the maximum diffusion distance of 1/2LFL gas cloud is 116 m, which is 14 m shorter than that of single tank leakage. The results show that the wind speed will increase or decrease the diffusion of gas phase leakage due to the change of actual data. The higher the leakage height is, the lower the gas phase leakage distance is. The existence of obstacles to a certain extent obstructs the diffusion of gas cloud downward wind direction, reducing the explosion damage range.

Abstract:To solve the problem of excessive cumulative error and long heading divergence in the position projection (PDR) algorithm, an adaptive extended Kalman filter (EKF) fusion algorithm based on UWB/PDR is proposed. The adaptive calibration factor can be achieved from the UWB positioning value and the PDR real-time position solution, and the position error is calibrated by dynamically adjusting the weight of the UWB observation with the adaptive calibration factor based on the conventional EKF algorithm. And the real-time ranging of the UWB is used to periodically correct the heading divergence of the PDR. The experimental results show the heading dispersion error is reduced by 63.9% with the adaptive EKF fusion algorithm compared to the pure PDR and 31.1% compared to the general EKF fusion algorithm, Moreover, the positioning 100m error is reduced to 0.33m.

Abstract:The new generation of video compression standard (H.266/VVC) provides 67 prediction modes in intra-frame prediction, which greatly improves the coding efficiency, but also brings extremely high computational complexity. This paper proposes a fast algorithm for intra-mode decision-making based on deep learning. First, for the problem of the size and shape of the block after the size of the coding block is divided, the extracted brightness block is preprocessed, and the block size and quality are guaranteed through random cropping, resampling, and convolutional neural network (CNN) upsampling. . Then the CNN architecture is carefully designed to reduce the complexity of intra prediction, and it is proposed to use the current coding block, the adjacent reference block and the residual block as the input of the network to convert the rate-distortion decision-making process into a classification problem and reduce unnecessary Pattern traversal. In order to train the proposed deep learning network, this paper establishes a model decision data set based on the characteristics of H.266. Experimental results show that compared with VTM10.0, the algorithm proposed in the article reduces the coding time by 39.56%~43.45% on average, which effectively reduces the computational complexity of coding, while the rate-distortion performance remains basically unchanged, which is comparable to the latest references. The overall performance has also been improved.

Abstract:In order to improve the automation degree of non-contact tonometer and speed up the efficiency of human eye pupil location, a multi template matching algorithm based on XLD contour is proposed, which can realize the rapid and accurate location of human eye pupil, increase the detection range of non-contact tonometer pupil location and improve the automation level. At the same time, for the false detection image, bilateral filtering, threshold segmentation and corrosion expansion processing are adopted, and then the pupil region is obtained by setting parameters. Finally, Hough transform is used to reposition the false detection image. The experimental results show that the positioning success rate of the proposed method is more than 99%, the error is within 5 pixels, and the average positioning is 0.121s. It can accurately and quickly obtain the central coordinates of the human eye pupil, and can meet the requirements of the non-contact tonometer for accurate and rapid pupil positioning. At the same time, it has strong robustness and generalization to interference factors such as individual pupil difference, posture difference, partial occlusion of pupil, influence of edge spot and so on.

Abstract:The local convex connected Patches algorithm (LCCP) suffers from the defects of super voxels crossing object boundaries and failing to utilize the regionally implicit concave-convex information. In order to improve the problems of low segmentation accuracy and object adhesion caused by the above defects, an improved algorithm combining connected domain segmentation is proposed. Firstly, the depth-adaptive superpixel segmentation (DASP) method is used to divide the image into superpixels based on depth information and normal vector angle; secondly, the concave-convexity of neighboring superpixels is determined based on the normal vector angle of superpixels, and all convex connected superpixels are combined to form the preliminary result; finally, the distance transformation and the watershed growth segmentation method based on superpixels are used to quickly segment the concave connected domain with large area into multiple convexregions. The segmentation is validated on the IC-BIN dataset, and the results show that the average segmentation accuracy (AP) is improved by 25% and 35% compared to LCCP and constrained plane cut (CPC), respectively, which significantly improves the under-segmentation problem.

Abstract:In this paper, an unsupervised deep learning algorithm is presented to solve the problem of multi-focus fusion of rock slice images collected under a microscope. In order to extract the deep features of images, a codec neural network is trained with unsupervised method to extract the depth features of different focused images and get the feature map. Then, a binary decision graph is calculated using the spatial frequency of the feature graph. Due to subtle decision bias, there may be holes and burrs in the binary decision maps, so the decision maps are morphologically processed and filtered. Finally, the fused image is obtained from the processed decision map. The experimental results show that the data evaluation index 、、 of this method is 0.7477、0.9874、0.7969.At the same time, the subjective effect is better than other methods. Therefore, Experiments show that this method can achieve good results in the application of multi-focus fusion of microscopic rock slice images and general images.

Abstract:In order to achieve high-precision measurement of surface defects of precision optical elements, a surface defect detection method of optical elements based on multispectral image fusion is proposed. By using light sources of different wavelengths incident on the optical surface, the detection results of optical surface defects illuminated by single wavelength light sources of 450nm, 532nm and 650nm are obtained in the micro dark field imaging system. After the collected images are fused by three methods: weighted average method, Laplace pyramid transform method and wavelet transform method, then image processing is performed to obtain the defect size information. By comparing the experimental results, compared with the defect detection results of single wavelength original image, the detection results of multispectral fusion image are more accurate, and by comparing the results of three fusion methods, the Laplace pyramid transform fusion has the best effect.

Abstract:To measure the moment of inertia of rigid body under damping condition, a new method combining machine vision and torsional pendulum method is proposed. According to the measuring principle of the torsional pendulum method, the mathematical model of the pendulum swing angle θ with respect to time t, the damping ratio ζ and the undamped natural frequency ωn is established. Finally, the calculation method of the moment of inertia under the linear damping condition is deduced. A high-resolution industrial camera and an imaging system were used to photograph the torsion pendulum. The curve of the pendulum swing angle θ with respect to time t was obtained by capturing the image edge, extracting the feature points and calculating the pendulum swing angle. According to the curve, the damping ratio ζ, the period T and the moment of inertia I were calculated. A set of experimental system is set up to measure the moment of inertia of the measured object whose moment of inertia is less than 4.00×10-3 kg·m2. The results show that the absolute value of the measurement error is less than 4.55×10-5 kg·m2. Repeated measurement experiments were carried out for four kinds of tested objects, and the standard deviation was better than 1.2×10-4 kg·m2. This method can be used to measure the rigid body inertia in engineering. The Z-axis rotation inertia of an unmanned aerial vehicle is measured, and the validity of the method described in this paper is further verified.

Abstract:Hyperspectral image contains rich geographical location information and spectral information. Hyperspectral image classification is a basic and important research direction in the field of remote sensing. However, the insufficient number of hyperspectral image samples is still the main problem that restricts the further improvement of classification accuracy.In generative adversarial network, generator and discriminator are constantly learning against each other. In the final ideal state, the pseudo sample discriminator generated by generator cannot be discriminated and pseudo data samples very similar to real samples are generated.  This paper uses generative adversarial network to generate new pseudo-samples based on a small number of original samples, so as to solve the problems of sample acquisition difficulty and insufficient sample quantity.  In the experiment, 200 and 400 sample points were selected from two hyperspectral image data sets, and new pseudo-samples were generated in the generative adversarial network for classification training.  Compared with SVM, 3DCNN and other classification methods with insufficient samples, the average accuracy of the whole classification has been significantly improved. Experimental results show that the classification performance of the proposed method is better than that of other classification methods.

Abstract:Currently, CCTV pipeline closed-circuit television inspection systems are generally used to detect drainage pipes, but the system has problems such as high manual participation in the defect judgment process, low detection efficiency, and large subjective errors. This paper proposes a method based on instance segmentation algorithm combined with CCTV drainage pipeline defect detection method, using CCTV inspection system to collect pipeline images, based on Mask R-CNN convolutional neural network drainage pipeline defect instance segmentation detection, cracking, deformation, corrosion, deposition, obstacles The defects of objects and tree roots shall be inspected and classified, and the cracking defects shall be inspected and rated. The experimental results show that the on-site test defect detection accuracy rate reaches 100%, and high-level classification accuracy of cracked defect detection in field experiment reaches 100%, showing good detection performance.

Abstract:As electronic equipment becomes more sophisticated and complex, ATS has gradually exposed the problem of poor versatility. The industry has produced signal-oriented tests and proposed relevant standards to achieve generalization. This method is relatively mature in terms of signal and virtual resource description, but in practice, it needs to complete the virtual-real mapping and other tasks. In response to this problem, a signal-oriented ATS runtime system solution is proposed. Based on ATML standards, a runtime system with initialization channels, signal capability matching, path selection and processing functions is designed to complete the virtual and real mapping from virtual resources to physical instruments. The program is simulated and verified, and the matching results and channel output are in line with expectations. For the development and use of ATS, it is of great significance in reducing special equipment, achieving generalization, and ease of development and maintenance.

Abstract:Early and accurate detection of cardiovascular disease is of significant importance to reduce the mortality caused by it. In view of the digital and synchronous combined detection technology bottlenecks in the common auscultation and ECG detection methods used in the early detection of cardiovascular diseases, a highly sensitive heart sound probe based on the MEMS vector hydrophone chip was developed and integrated a 5-road heart sounds and standard 12-lead ECG simultaneous acquisition instrument system. The test results show that the developed heart sound sensor probe has a signal-to-noise ratio of 40dB, which is better than the 3M stethoscope; and the QRS wave ECG signal collected by the system is synchronized with the characteristic peaks of the S1 and S2 heart sound signals, and the waveform characteristics are accurate and clear. The test has realized the digital detection of heart sounds and the intuitive identification of abnormal characteristics of the disease. In addition, the system is optimized to have intelligent diagnosis capabilities. Therefore, a cloud server and database have also been built, and an intelligent diagnosis algorithm for heart sounds has been designed and developed. Through testing on the collected clinical case, the results show that the accuracy of the five-category heart sound intelligent recognition is over 90%. Therefore, the detection system has important technical value and clinical guiding significance for assisting physicians in the early digital detection and joint diagnosis of cardiovascular diseases.

Abstract:Currently, pixel-level edge detection algorithms are mostly used in the dimension vision measurement for shaft parts, which can hardly be adapted to the high-precision measurement requirements of industrial automation. In order to improve the measurement accuracy of shaft diameter of shaft parts, a dimensional measurement method for shaft parts based on improved Zernike moment is proposed. Firstly, Canny edge detection algorithm is used for rough positioning in shaft images to obtain pixel-level edges. Then, in accordance with the grayscale difference between the target and the background information of the image , a method for obtaining the optimal decision threshold of Zernike moment based on the multi-threshold Otsu method is proposed to obtain sub-pixel edges. Finally, an improved least square method to fit the edge line of shaft images based on edge point search is proposed to obtain the measurements of the shaft diameter. The experimental results show that, taking the measurement of bicycle rear axle diameter as an example, the proposed algorithm has better stability than the traditional Zernike moment method, and the relative error with the manual measurement is less than 0.011%. The measurement accuracy of the proposed method can meet the accuracy requirement of 6-level of bicycle axle diameter in industrial parts dimension measurement.