Abstract:For the problem of poor objective evaluation of online education courses, a prediction model based on Light Gradient Boosting Machine - Field-aware Factorization Machine - Logistic regression is designed. The model collects online course viewing history data, extracts users' generic features, temporal features and other feature values, and focuses on the relationship between high-dimensional features and low-dimensional features of feature values to achieve multi-dimensional feature combinations and improve data sparsity, so as to improve rating prediction performance. Based on masked data test on an online course website, the determination coefficient between predicted grading and actual one in this model is 0.87, with 0.42 of average mean square error, improved model generalization capability, this model serves more objective and realistic predicted grading to online course.

Abstract:In order to enhance the high efficiency, real-time performance and stability in the process of digital image acquisition and recognition, a system based on Altera Cyclone IV E series EP4CE10F17C8 FPGA as the main control chip and SCCB transmission protocol is designed.The system uses the OV5640 camera as the front end of image acquisition. The collected digital image data is cached to the external SDRAM, and then the data is read from the SDRAM to carry out color space conversion, binarization processing, projection segmentation and feature matching recognition by LCD driver.Finally, the identification results are displayed by digital tube and RGB LCD screen.For printed numbers, the system can realize fast and accurate identification, suitable for low delay and high demand on identification accuracy.

Abstract:The access of unmeasurable load branch had an impact on the protection of distribution network. So this paper designed a differential protection scheme based on the similarity of bilateral currents. The current characteristic of active distribution network with unmeasurable load branches in different fault places was analyzed. On this basis, Euclidean distance was used to measure the similarity of the currents on both sides, so that the braking coefficient can be modified adaptively with the fault location. It used PSCAD to verify the scheme, and the results show that when different unmeasured load branches accessed, the braking coefficient would decrease to 0 in case of in zone fault or increase to 0.87 in case of out of zone fault, so as to realize that the current differential protection could operate correctly. Therefore, the influence of unmeasured load branches on distribution network protection is avoided.

Abstract:In order to solve the problem that the wireless magnetic induction communication system in underground environment has large path loss and can not achieve long-distance communication, this paper constructs the magnetic induction channel models based on waveguide technology and collaborative technology. Through waveguide technology, passive relay coils are introduced between the transceiver coils, and the transmission of signals is completed through the magnetic coupling of a series of resonant coils; through cooperative technology, the transmitted magnetic induction wave reaches the receiver by multiple different paths to complete the transmission of information.Through the simulation analysis of path loss, BER and other characteristics, the results show that: when the path loss is less than 100dB, the maximum transmission distance of the ordinary magnetic induction system is about 20m, while the maximum distance of the magnetic induction collaborative system and the waveguide system is extended to 45m and 150m respectively; the path loss of the magnetic induction waveguide system is greater than that of the cooperative system when the transmission distance is close, and its path loss can be further reduced by reducing the spacing and resistance of the relay coils. The path loss of magnetic induction cooperative system is less affected by the location, spacing and number of coils, thus reducing the difficulty of deployment. So, both communication models can effectively reduce the path loss of the channel and increase the communication distance, and the waveguide technology is more effective.

Abstract:In view of the phenomenon that the distributed static synchronous compensator (DSTATCOM) is insufficient in compensating the reactive current of the unbalanced load and the compensation speed and accuracy of the unbalanced current in the three-phase four-wire distribution network. Firstly, the four-wire DSTATCOM mathematical model is established, then the design of a new linear active disturbance rejection controller (LADRC) is completed, and the new controller is applied to the current inner loop to quickly track the command current signal. The disturbance immunity, tracking and stability were analyzed by classical control theory and mathematical theory. Finally, the three-phase four-wire DSTSTACOM system model under the traditional and new LADRC is built through the MATLAB /Simulink simulation platform for simulation experiments.The results show that under the action of the new controller, the power factor of the system is closer to 1, the midline current is closer to 0 and the response curve is faster, indicating that the four-wire DSTATCOM under the new LADRC has better compensation speed and accuracy.

Abstract:Standard metal gauge is a common measuring instrument in the measurement industry, which has been widely used in the measurement industry. This paper aims at the problems of low efficiency and low experimental accuracy in the manual verification process of standard metal gauge. A liquid level control method is proposed. Firstly, the standard metal gauge tank is modeled and analyzed, and the switching points of large and small flow valves are calculated. Then, the liquid level sensor is used to read the liquid level height at the glass pipe of the tank neck in real time, and the duty cycle of the small flow valve is controlled by the programmable logic controller, So as to achieve the purpose of high-precision liquid level control. Finally, based on the simulation platform, the liquid level control at the glass tube of the standard metal gauge is simulated, and the liquid level control error is ± 0.05 mm. The results show that the method proposed in this paper can improve the liquid level control accuracy of standard metal gauge and has good popularization value.

Abstract:The modular design method is adopted for the hardware circuit of electronic control unit on the high pressure common rail aero piston engine, which is helpful for the electronic control hardware system to adapt to the development trend of functional integrated chip. As the core of Infineon 32-bit microcontroller chip, it was modularly designed from the whole to the details by including the 3.3 V and 5 V power supply, MCU crystal oscillator power supply, 0 ~ 5 V sensor signal conditioning, high and low side power driving of the actuators, PCB layout and so on. After PCB printed circuit board and manual welding of electronic components, the circuit hardware system was verified on the engine bench. In the experimental results, it was analyzed that the 0 ~ 3.3 V voltage limiting and variable frequency square wave conversion of the signal conditioning module of crankshaft and camshaft sensor, the maximum 5 A power driving of the high-side MOS tube of the high pressure pump metering solenoid valve and the driving Peak-Hold transient current of the 24 ~ 48 V hybrid half-bridge MOS tube of the fuel injectors.

Abstract:Aiming at the problem that the rapid expansion of random tree algorithm (RRT) has no directionality and long and tortuous paths, this paper proposes an improved RRT algorithm and applies it to the path planning of manipulators. The principle of probability target bias and double sampling point selection are used to optimize the sampling points to enhance the heuristic of the algorithm; then the idea of target gravity is introduced to change the expansion direction of the new node, and on this basis, a variable step idea is proposed to enhance the effect of obstacle avoidance. The improved algorithm, RRT algorithm and the probability-biased RRT algorithm are simulated in Matlab, the results show that the execution time of the improved algorithm is 54.7% and 33.6% less than that of RRT and P-RRT in a three-dimensional environment, respectively, and the path length is also greatly improved.

Abstract:A comprehensive power quality evaluation method based on the game theory and the improved technique for order performance by similarity to ideal solution (TOPSIS) is proposed to solve the problems caused by using a traditional and single evaluation method in the comprehensive evaluation of power quality, such as too strong subjective consciousness, incomplete evaluation of objective raw data and uneven weight distribution. Firstly, consider the actual engineering situation and establish the valuation index system, and the Best Worst Method (BWM) and criteria importance through intercriteria correlation (CRITIC) methods are used to obtain subjective weights and objective weights, respectively. Then, the game theory idea is introduced to combine the subjective and objective weights to avoid the defects of the single assignment method. Secondly, improve the fitting degree flaws of TOPSIS through relative distances, which makes it more applicable, and the evaluation objects are ranked in order of merit and gaven level determination by using the improved TOPSIS. Finally, the data collected from four monitoring points in a power supply area is used as an example to carry out a comprehensive evaluation of the power quality and a comparison with other evaluation methods. The results of the comparison analysis show that the relative distances of the improved TOPSIS method at monitoring points 2 and 3 are 0.232 and 0.256 respectively, which overcomes the shortcomings of the traditional method where the fit is similar and difficult to distinguish, indicating that the proposed method can reasonably evaluate and rank the power quality of monitoring points, verifying the effectiveness of the method.

Abstract:Beamforming is a particular method for sound source localization. However, in most of the relating researches, the intensity of the source is often ignored. Therefore, this research proposes a method of sound source localization and intensity estimation based on the residual network based on the conventional beamforming map, aiming to accurately predict the position and intensity of the point sound source. Acoular software is adopted to simulate time signals, the neural network is trained by the simulated signals, and the prediction model can be obtained then. It is verified by computer simulation whether the deep neural network can obtain an accurate description of a single point source from the microphone array data. The results show that the proposed method predicts the location and the intensity of the sound source fast and effectively. Moreover, the proposed method behaves better prediction effect for higher frequencies.

Abstract:Because the infrared thermometer needs to measure the body temperature of the subject at close range in the Covid-19 epidemic, and the measurement accuracy is relatively low, the safety of the body temperature measurement is not high and the risk of misjudgment is caused. In order to improve the accuracy of human body temperature detection by infrared temperature measuring equipment, this paper studies the distance temperature compensation method to improve infrared temperature accuracy, and designs an infrared temperature measuring gun system with long distance measurement, accurate measurement and wide application. The system is designed with STM32F103 as the main controller, and AMG8833 infrared thermal imager sensor module, OLED display module, laser ranging module and buzzer are used as peripheral equipment. In order to improve the accuracy of temperature measurement, the temperature value of the infrared dot matrix collected by the sensor module of the infrared thermal imager is compensated for light perception and distance temperature, that is, the collected temperature points at different distances are processed by fast descending sorting method firstly, and then the temperature points at different distances are fitted by multiple linear functions, so that the temperature compensation after passivation of different distance sorting is realized. After many experiments and debugging, the error is great when the distance temperature compensation is not carried out, and the error of the temperature value after distance temperature compensation is very small compared with the reference temperature value, which can meet the accuracy requirements of human body temperature measurement and improve the accuracy of long-distance temperature measurement.

Abstract:The current detection of the power loop and the signal loop is an important test item for the Solar Array Driving Mechanism (SADA). Existing detection methods perform contact detection on the SADA current loop. If the detection equipment fails or the personnel operate improperly, it will cause damage to the SADA. In order to avoid damage, a non-contact multi-channel current tester is designed based on the Hall Effect, which can effectively measure the current in the SADA power loop and signal loop without touching the target under test, and the detection accuracy is better than 0.2 %. The comparison test with the original detection method shows that the relative errors of the two methods for the SADA signal loop and power loop are both within 0.1%. It can be seen that the measurement accuracy of this instrument is sufficient to effectively meet the test requirements of the SADA signal loop and power loop, and its advantage is to reduce the risk of product damage during the detection process.

Abstract:The reducer is one of the core components of the motion of industrial robots. There are a large number of trigonometric function calculations in its real-time load calculation, which requires a large amount of calculation, and the calculation using traditional methods is complex and has a high delay. This paper proposes a CORDIC calculation algorithm for robot reducer load, which can effectively reduce the load calculation delay. The focus is on the implementation of the algorithm in FPGA and structural improvements. Finally, a joint simulation is carried out in Quartus Ⅱ and ModelSim software. The results show: The absolute error of the load calculation of the robot reducer based on the CORDIC algorithm is less than , and the calculation delay is 15ns. Compared with the traditional load calculation method, it has the characteristics of high accuracy, good real-time performance, and less internal resources of the FPGA.

Abstract:Aiming at the problem that the incipient fault characteristics of rolling bearing are weak, and the vibration signal is a group of time-varying sequence, which has a certain time-series correlation, leading to the difficulty of incipient fault detection of rolling bearing, a dynamic independent component analysis fault detection method based on deep decomposition principle (Deep DICA) is proposed in this paper. The main idea is to first increase the observation data matrix in order to take the dynamic process into account. Then, in order to better dig out the weak incipient fault information, the principle of deep decomposition is proposed to extract the features of incipient faults. Finally, a fault detection model is established for online fault detection and the proposed method is verified by bearing experiments. Experimental results show that the proposed fault detection method based on Deep DICA has good accuracy and applicability.

Abstract:In order to meet the miniaturization requirements of a certain filter in satellite communications, a C-band interdigital filter with a ceramic substrate was designed based on thin film technology. The filter is composed of a 5-order parallel coupled resonant unit, and the input and output form adopted a tap structure, which is more conducive to the miniaturization of the filter than the parallel coupling input and output. In order to meet the high rejection index, the filter adopted cross-coupling and parallel 1/4 wavelength open route to introduce transmission zero points, which effectively improved the filter’s out-of-band rejection.Besides,the standing wave of the filter was adjusted by means of impedance compensation.The test results showed that the center frequency of the filter is 3.7GHz, the relative bandwidth is 27%, the center insertion loss is -1dB, the stopband rejection is 30dB at 2.7GHz and 4.8GHz, and the stopband rejection is 40dB at 2.5GHz and 5GHz. The filter size is 9.25mm *6.72mm.

Abstract:In the process of predicting radio wave propagation characteristics in mountain scenes by ray tracing technology, in order to improve the computational efficiency of ray tracing algorithm, an improved acceleration method based on region division and considering ray directivity is proposed. This method connects the emission point with the four vertices of the terrain triangulation model, divides the terrain surface into four regions, then calculates the intersection of each group of rays emitted by the emission point and the four edges of the terrain surface, and calculates the mesh passing through the connecting line between the starting point and each intersection according to the triangle index in the terrain model. Finally, according to the ray directivity in the vertical direction, the triangular meshes where the rays cannot intersect are further screened out to reduce the number of intersection. In this paper, the same mountain terrain is selected, and the acceleration method proposed in this paper and the traditional octree acceleration technology are used to simulate and compare the simulation results. The numerical results show that when the number of emitted rays is set to 120 × 181, compared with the traditional octree algorithm, the calculation speed of this acceleration method is improved by 50.302%, and the error is 9.547%. Therefore, compared with the octree algorithm, the algorithm implementation of the proposed method is simpler. It can improve the simulation speed and maintain small error, which has a certain reference value.

Abstract:In order to realize the long-time acquisition and storage of underwater acoustic transducer on small underwater working platform, a miniaturized, low-power, large-capacity acquisition and storage system based on STM32H743 series as the core controller is designed. The program design mainly uses STM32CubeMX to complete the overall framework construction and peripheral controller configuration, and completes the application control logic design based on the HAL library. In particular, the necessary exception handling and information feedback are added to the storage type peripheral control, which not only improves the reliability of the system, but also provides a variety of emergency handling solutions for the application platform. After the test, the system performance is stable, and the overall power consumption can be controlled within 1.5W. After data upload and analysis, the Ethernet upload speed can reach 3MB/s, the upload function is stable, and the data storage is normal. At present, the system has been produced in small batches and applied to the small underwater working platform.

Abstract:Aiming at the Existing problems of ultrashort wave radio that detection process cannot be detected in different places, destroy the current connection state radio, not to much radio detection at the same time, developing a quick wireless detection system, through the beidou satellite navigation system and radio test adapter device、the testing platform, make double wireless channel including the RF signal channel and Beidou satellite channel between stations and testing platform. Short message The Beidou satellite system is used to measure the distance between the stations and testing platform, calculate the attenuation of the radio frequency signal, use a certain algorithm to compensate, and through the Beidou short message to carry out the transmission and correction of the relevant information in the detection process, to achieve the detection of the station's main transmission and reception Performance indicators. By using the testing system to test the 60 km away from the radio, compared with the field measurement, the frequency error is within 10-6, transmission power, sensitivity, modulation error within 5%, within 20 minutes to complete the two different locations radio detection and meet the needs of radio index rapid detection, improve the efficiency of different radio detection.

Abstract:For the issue of workpiece size measurement, most of them use manual method to measure the size of workpiece, which is not only labor-intensive, but also prone to errors. In order to avoid this kind of situation, machine vision technology is designed to measure the size of the workpiece. Here, this paper uses Halcon image processing software to measure the height of the workpiece, uses binocular camera to collect the image of the workpiece, analyzes the working principle of binocular stereo vision, calibrates the camera, and puts forward a method of extracting saddle point to locate the workpiece, and uses plane vector to determine the height dimension of the workpiece, Results the height dimension of the workpiece was successfully measured to be 50.3mm. The experimental results show that after a series of binocular stereo vision processing methods, the height and size of the workpiece can be obtained, which not only improves the efficiency, but also has great practical value for the automatic production of the assembly line.

Abstract:The liquid level measurement of the connecting pipe is an important way to detect the deflection of the bridge. Abstract: Aiming at the problems of insufficient real-time performance and measurement accuracy of traditional liquid level measurement methods for connected tubes, a liquid level measurement system for connected tubes was studied based on machine vision, which was composed of liquid level observation tube, light source, parallel light plate, industrial camera, microprocessor and router. The edge contour of the liquid level image was obtained by image preprocessing, and the sub-pixel edge detection algorithm was used to accurately locate the bottom coordinates of the edge contour of the liquid level image. A method for calibrating and measuring the liquid level height based on the principle of relative motion was proposed. The function relationship between the bottom coordinates of the edge contour of the liquid level image and the liquid level height was obtained by fitting with the least square method. Finally, the feasibility, accuracy and dynamic response of the liquid level measurement method are verified by static measurement test and pseudo-bridge vibration test in the laboratory. The maximum absolute measurement error is less than 0.5mm, which can meet the engineering application needs of deflection measurement of connecting pipe bridges.

Abstract:In order to solve the issue of poor generalization ability of traditional classification models, an innovative meta-learning method was proposed in this paper. Firstly, a dynamic task-driven module was employed, which enabled the classification model to mine external task information; Secondly, a collaborative training method was adopted to overcome the increased complexity of calculation caused by the introduction of task-driven module; Finally, in order to exploit the relevant information between the samples, the graph neural network was used to be the classification module to achieve the purpose of further improving the classification accuracy. Compared with baselines, the accuracy of this method has increased by 6.81% and 6.20% on the MiniImageNet and CIFAR-FS datasets respectively. The experimental results show that the method in this paper can effectively improve the generalization ability of the classification model.

Abstract:In order to solve the three dimensional block matching algorithm on the processing image edge of high-frequency pseudo like ladder and inhibition effect, this paper proposes a adaptive BM3D algorithm based on generalized variational all. First, after a preliminary classification get two blocks with different structure information subset. In a complex change of subset, considering the adaptive algorithm can significantly reduce the BM3D algorithm used to match the traversal, to adopt adaptive reference block matching the area; In a uniform variation of subset, application of the original fixed reference block size. The staircase effect in view of the image processing, this paper puts forward a new kind of generalized variational (CTGV), second order total generalized variational can automatic balance of first and second order partial derivative of the image features, this paper introduced the variation coefficient, the adjustable generalized variational diffusion effect. By comparison with the experimental data, the algorithm compared with BM3D algorithm and other algorithms, image PSNR improvement more than 1～2 db, SSIM values also have a significant boost, contrast visual effects, this algorithm is effective in addition to the high frequency artifacts and the effect of the inhibition of ladder.

Abstract:Aiming at the common problems of mixed noise and fuzzy edge contour in digital images, an image preprocessing algorithm combining modified alpha mean filtering and edge enhancement algorithm is proposed. The modified alpha mean filter algorithm first statistically sorts the pixel values in the pixel matrix, then removes the quantitative maximum and minimum values, and then calculates the average value of the remaining pixels in the pixel matrix.Finally, the average value obtained is used to replace the pixel value of the original center point to achieve image denoising. The edge enhancement algorithm first uses the Sobel operator to locate the edge information of the image, and after extracting the edge of the image, the edge is superimposed on the denoised image, and finally the edge composite enhancement of the image is realized. It has been verified that with the continuous increase of noise density, the algorithm has always maintained a smaller mean square error value than traditional algorithms, and the mean square error growth rate is about 31%.The results show that the proposed algorithm effectively filters out the mixed noise of digital images, enhances the edge information, and satisfies the real-time requirements of the image noise reduction system.

Abstract:Aiming at the problem that during the process of large-size panel-level fan-out chip packaging, the chip position drift in the plastic packaging process will affect the subsequent process, the precision positioning and detection method of the 600mmx600mm large-board fan-out package chip is studied. The automatic positioning of large-format small targets is realized by the CCD camera mounted on the marble air-floating gantry platform. Measure and analyze the CCD image edge positioning accuracy, camera installation horizontal deflection error, positioning accuracy of the marble floating platform, straightness and verticality errors, establish an error model, reduce the cumulative error generated by the motion platform through error correction, and improve the motion platform The positioning accuracy. Experiments show that the sub-pixel edge detection algorithm combining Canny operator and Zernike moment has strong anti-interference ability, and the edge positioning accuracy reaches 0.3 pixel. After error compensation, the positioning accuracy of the moving platform is reduced from 55.3um to 2um, and the verticality of the X and Y axes is reduced from 14.40arcsec to 0.96arcsec. The positioning accuracy of the large-slab fan-out package chip positioning detection method integrated with the marble air-floating gantry platform and the CCD camera is 7um, which can meet the accuracy requirements of most chip positioning in the chip packaging process.

Abstract:The most popular electrolytic capacitor detection method can be divided into two stages：inner circle detection and polarity detection. However, when affected by the light, the luminance of the inner circle can be very close to that of the outside, the small difference is difficult to be captured by GHT-based circle detection, thus making the inner circle mislocated. In addition, the polarity detection method based on the sliding window smoothness does not consider the three cases of low luminance or high noise in the polar region and the existence of high luminance pollution in the non-polar region, so the application of this method is limited. In this paper, we propose a novel method of inner circle detection based on the clustering segmentation of super pixels, which takes the relationship of luminance and position among super pixels into consideration simultaneously and can obtain the correct position even under poor illumination conditions. Furthermore, we propose a polarity detection method based on the selection among peaks and valleys that calculated by gray-level mean and variance of the sliding windows. The method has the ability to handle the general cases as well as the three complex cases mentioned above. It is worth noting that with our approach it is possible to provide direction directly, which was not possible in any previous approach, and facilitate full automation. The experimental results show that the test accuracy of the method reaches 98.6%, and the average detection time of a single image is 192ms±23ms. The solution has been accepted by many manufacturers and performs well.

Abstract:Because of the great difference of sensor imaging, the registration of high-resolution remote sensing image is faced with more prominent local deformation problem.In particular, the relative parallax offset of high-rise objects in the city is more significant, which leads to serious nonlinear error in spatial transformation.Therefore, this paper proposes a registration method for multi-source high resolution remote-sensing image based on high-rise objects.Firstly, through shadow detection and image segmentation, the high-rise objects are screened.On this basis, a threshold adaptive feature point extraction strategy based on phase consistency is proposed, in order to improve the number of feature points in high-rise objects and their overall distribution uniformity.Secondly, the distance of feature vector weighted by shadow area is introduced to eliminate the interference of shadow on feature point matching.Finally, an adaptive penalty factor of transformation error is designed to reduce the influence of high-rise objects’ spatial variation on affine equation.Through the registration experiments on groups of multi-source high resolution remote sensing images, it is found that the registration accuracy and root mean square error of the proposed method can reach 88.9% and 1.481 respectively.

Abstract:In order to solve the problem of robot sorting in the disorderly parts box in the complex industrial production environment, it is necessary to complete the spatial positioning of the workpiece, the recognition of different types of the workpiece and the grasping operation of the robot. The existing vision technology can not meet the random sorting task.Therefore, an intelligent robot sorting system combining binocular stereo vision, deep learning and UR5 robot is proposed.A three-dimensional vision system combining stereoscopic vision and projection structured light was proposed. The energy function of stereoscopic matching was reconstructed to complete the spatial positioning of the workpiece.An instance segmentation method based on deep learning was used to accurately identify the workpiece.A random sorting system based on binocular structured light and deep learning was realized by combining robot hand-eye calibration technology with workpiece spatial positioning and recognition results. By analyzing the success rate of random sorting of screw workpiece in the process of random sorting, the average success rate of all single sorting of screw workpiece of different quantity is 92.8%, and the success rate of random sorting according to the total number of workpiece is 98.8%, which verifies the feasibility of the system.