Abstract:The electromagnetic torque sensor can realize contactless torque measurement. The sensor has many advantages such as high accuracy, strong anti-interference ability, no additional power supply and easy installation. It is very suitable for torque measurement in extreme environment with high temperature, high humidity and dust. This article introduces and revolves around the working principle of electromagnetic torque sensors The coverage categories of electromagnetic torque sensors is clearly summarized and several representative electromagnetic torque sensors are introduced. The development of electromagnetic torque sensor in the past 30 years at home and abroad is reviewed. According to the technical characteristics of measuring torque range, the application and study status of electromagnetic torque sensor are discussed in detail, including traditional automobile, ship, aviation and other industries. In addition, the application and study status of emerging industries are elaborated, including biomedicine, medical equipment and robot industry. The problems that restrict the development of electromagnetic torque sensors and the direction of improvement are analyzed. Finally, the development trend of such electromagnetic torque sensors is summarized and explained.

Abstract:With the continuous development of integrated circuits, low power consumption and small area have gradually become the important specifications in chip design, which promotes the continuous reduction in the size of the devices that constitute the circuit. In the semiconductor chip manufacturing process, smaller device size has higher requirements for temperature control accuracy in the process. The slight deviation of the wafer temperature and the temperature non-uniformity higher than 1% will directly affect the yield of the final product. In order to achieve high-precision control of temperature and temperature field distribution, more accurate predetection and real-time acquisition are necessary, which pushes the wafer temperature monitoring technology in the integrated circuit manufacturing come into being. Focuses on the two main directions of contact and non-contact temperature measurement technologies, introduces the principles of the temperature monitoring technologies applied in the temperature monitoring range of 0℃ ~ 1 300℃ , based on the principles analyzes the advantages and drawbacks of various technologies in detail, keeps track the development status of various each temperature measurement technologies both at home and abroad, and looks forward to the future development of wafer temperature monitoring technology.

Abstract:Magnetic fluid has both the fluidity of liquid materials and the magnetism of solid materials, which can exist stably for a long time under the action of gravitational field and magnetic field. The unique first-order buoyancy principle of magnetic fluid makes it could suspend non-magnetic objects, whose density is higher than itself, under the action of magnetic field gradient. Based on the first-order buoyancy principle of magnetic fluid, a novel magnetic fluid tactile sensor is designed. When the contact pressure acts on the nonmagnetic point of the suspended rod, the movement of the suspended rod will cause the change of magnetic field at the Hall element. Then, the output voltage signal will change accordingly. The structure can be used for simultaneous measurement of contact pressure, surface profile and micro displacement. Compared with the traditional silicon wafer, the tactile sensor has small size and low cost. In addition, the magnetic liquid can absorb energy in the system, which can help to improve the impact resistance of the sensor system. Within the contact pressure measurement range of 0~ 0. 09 N, the measurement accuracy can reach in order of 10 -2 N. The sensitivity is 3. 34 V/ N, the linearity error is 3. 4% , the hysteresis error is 1. 4% , and the resolution is 1. 1% F. S.

Abstract:In the simultaneous localization and mapping (SLAM) problem, the solution accuracy of the odometry part plays a vital role in the subsequent mapping. The inertial measurement unit ( IMU) can provide valuable assistance for odometry in SLAM. Based on the consideration of the movement characteristics of the planar mobile robot and the indoor environment characteristics, proposes a laser odometry solution method based on IMU loose coupling to realize the precise positioning of the odometry part. In the first stage, the point cloud information is processed in real time during the robot movement. The ground points are segmented and key points are extracted. In the second stage, the IMU information is introduced into the Kalman filter to provide the pose prior for the inter-frame matching. In the third stage, after the filter outputs the pose estimation value, the non-linear optimization method is used to match the point cloud frames to realize the solution of the odometer movement. Experimental results show that the proposed method has good stability and accuracy in laser point cloud processing and motion solving. The offset error can be controlled within 0. 4% . This method provides powerful data guarantee for subsequent mapping.

Abstract:The design and optimization of electromagnetic ultrasonic transducer (EMAT) is a multi-objective optimization problem, which has many variables, complicated analysis and great difficulty in optimization. By formulating the electromagnetic force acoustic finite element model of the electromagnetic ultrasonic transducer, the sample sets of the optimal targets such as Lorentz force, magnetostrictive effect, induced current density and vibration displacement can be achieved. In this way, the agent model of the multi-support vector machine is established. An improved NSGA-Ⅲ multi-objective optimization method based on the combination of reference point and congestion degree is proposed. The optimization design is implemented on the optimization target, and the most satisfactory optimization scheme is selected from Pareto solution set through the multi-index weighted grey target decision model. Compared with other optimization methods, the improved NSGA-Ⅲ algorithm is more effective for solving complex multi-objective problems. The rationality of the optimization process and the accuracy of the optimization results are evaluated by experiments. Results show that the detection signal of the electromagnetic ultrasonic transducer is increased by about 25% after optimization, which effectively improves the energy exchange efficiency. It provides a new way for the parameter optimization of the electromagnetic ultrasonic transducer.

Abstract:In view of the special characteristics of metal materials in electrostatic field, a novel electrode capacitive sensor is designed to detect the flow velocity of metal particles in the pipeline. Firstly, based on the multi-electrode theory, a mathematical model for measuring metallic substance with capacitive sensor is formulated. On this basis, a 3D simulation model of the sensor is established by utilizing the COMSOL Multiphysics finite element analysis tool. The spatial sensitivity distribution characteristics of the sensor is analyzed. Then, based on the spatial filtering effect of the sensor, the principle of metal particle velocity measurement is introduced. The relationship between the cut-off frequency of the sensor output signal power spectrum and the metal particle velocity is established. Finally, a specific measurement system based on the triangular electrode capacitance sensor and the capacitance digital conversion chip Pcap01 is further developed. Experiments are verified and analyzed on a gravity transmission device. Theoretical and experimental results show that the measurement method has good repeatability. In the range of 2. 44 ~ 5. 34 m/ s, the repeatability error of the measurement velocity is less than 4% .

Abstract:In the replacement from old version to new version, there is a conflict between the new double “H” silicon resonant pressure sensor and the traditional pressure sensor, such as vibration cylinder and single beam silicon resonant. To solve the problem, develops a dual-frequency acquisition and synthesis system, which is based on the GD32 single-chip timer counting and interrupt timing technology. The system builds a multiple timer synchronous counting circuit by using the tiny single-chip microcomputer ( GD32F405RGT6) to realize dual frequency 20 ms synchronous acquisition. On this basis, a new digital frequency conversion method is proposed, which can control the I/ O port flipping level by the timer interrupt of the single-chip microcomputer to outputs square wave. And a dual-frequency synthesis error compensation method is designed to decrease the maximum theoretical error of the dual-frequency synthesis from 51. 15 Pa to 0. 127 9 Pa. Experimental results show that the system can work under -45℃ ~ +85℃ and 2~ 266 kPa. The single frequency range of the system output is 4~ 10 kHz. The dual frequency acquisition and synthesis error is less than 0. 005% of the full pressure scale. The digital circuit board is small and can be integrated into the new double “H” type silicon resonant pressure sensor circuit, which well satisfies the requirement of the dual frequency acquisition and synthesis output for the silicon resonant pressure sensor in atmospheric pressure detection and monitoring.

Abstract:The signal detection method of stress-strain sensing system based on the principle of Brillouin backscattering is studied. At present, the signal detection method in this field generally has many problems, such as large amount of calculation, low accuracy, and poor consistency. This paper proposes an adaptive gradient descent algorithm (Adam algorithm), which can be used to fit the Brillouin scattering signal. Meanwhile, a Brillouin strain measurement system for heterodyne coherent detection is established. Experimental results show that the average strain error demodulated after fitting the scattering spectrum curve with the Adam algorithm is 24. 89 με, which is 51. 96% of the maximum strain error of the Gauss-Newton algorithm fitting curve. This value is Levenberg-Marquardt LM algorithm fits 57. 42% of the maximum strain error of the curve. When Adam algorithm is used to fit the scattering spectrum curve, the good fit is 0. 998 9 and small root mean square error is 0. 110 5. These results indicate high precision of fitting measurement and signal processing. In addition, the signal processing time is only 18. 5 ms. Research results of this study provide theoretical and experimental basis for the high-precision data feature extraction of Brillouin scattering sensing technology.

Abstract:To improve the measurement accuracy of the spindle rotation error of precision machine tools, the multi-turn coincidence method is used to process the sensor data in the spindle rotation error measurement, which is based on the four-point matrix algorithm. The convergence speed of the traditional genetic algorithm is improved and the dependence of the optimization results on the initial value is reduced. The crossover and mutation probability factor formulas are updated, and the improved genetic algorithm is used to optimize the sensor installation angle and output weight coefficients. By using the improved genetic algorithm, the convergence rate is about 50% higher than the value of the traditional genetic algorithm. The spindle radial rotation error measurement and separation experiment are implemented by a multifunctional inclined rail CNC lathe. Compared with the calibration value, the deviation of the standard mandrel shape after separation is within 5% , and the error repeatability is less than 5% . Results show that the accuracy of the separation results is high. Thus, the correctness and feasibility of the proposed algorithm are verified.

Abstract:The scanning probe is the core part of the high precision 3D thread synthetical measuring machine, its dynamic characteristics seriously affect the precision of the machine. In order to improve the precision of the measuring machine, the dynamic characteristics of the linear scanning probe for 3D size measurement of high precision thread are studied. Firstly, the measuring principle of the scanning probe used in the 3D thread synthetical measuring machine is analyzed, then the dynamic characteristic model is established and the factors affecting the dynamic measurement result are put forward. Finally, the dynamic characteristic of the probe structure and the effectiveness of the measuring processing method are verified through experiments. The experiment results verify the correctness of the influencing factors. Through optimizing the sampling spacing factor that has the greatest influence, 80% of invalid data points can be filtered, so that the mean residual sum of squares error of the fitting in this section is reduced by 91. 0% and the linear error is reduced by 67. 4% . The measurement accuracy of 3D thread measuring machine is improved.

Abstract:In engineering, the inclined plane model is generally used to calculate the transmission efficiency of the ball screw pair, which cannot reflect the influence of load on the transmission efficiency, and there is lack of experiment verification. Therefore, it is necessary to study the transmission efficiency of the ball screw pair under different loads. Starting from the force analysis of the screw, considering the contact deformation, merging the structural parameters and the actual working conditions, proposes a transmission efficiency calculation model. In order to verify the effectiveness of the transmission efficiency calculation model, a transmission efficiency test bench for the screw pair was developed, and the transmission efficiency test of a 4010 type double-nut ball screw pair was carried out. The results show that: The calculated value of the transmission efficiency model and the measured value of the test both increase with the increase of the load, and finally tend to be stable. The transmission efficiency calculation model can more accurately calculate the efficiency value of the ball screw pair under different loads. Compared with currently used calculation formula for engineering type selection, the proposed model possesses stronger scientificalness. In actual engineering type selection design, the transmission efficiency calculation model can be used to effectively estimate the efficiency according to the load conditions.

Abstract:The service robot industry is developing rapidly. Laser ranging module is a core component of the robot, and its reliability directly affect the performance of the whole machine. Based on the working environment of service robot, this paper studied the operating characteristics of laser ranging modules, and established the universal environmental reliability model and test evaluation method of ranging modules. Taking ranging accuracy as the research object and the reliability evaluation factor Rs as the evaluation merit, the general expression of the reliability function for ranging modules was proposed. The environmental reliability test and quantitative evaluation of the ranging accuracy of five types laser ranging module samples were carried out with the self-built ranging accuracy test platform. Evaluation results showed that the Rs value of only one sample was less than 1, which indicated that its environmental reliability level was high, the Rs values of the other four samples were greater than 1 after the waterproof, dustproof and salt spray tests, which showed the phenomenon of inaccuracy or failure and exposed the weak parts in the design and application of the products. The test and evaluation results further verified the universality and feasibility of the environmental reliability model and its quantitative evaluation method.

Abstract:Monocular visual-inertial SLAM (simultaneous localization and mapping) systems recover poses by tracking the hand-crafted point features, such as Shi-Tomas, FAST, and so on. However, the robustness of hand-crafted features is limited in some challenging scenes, such as severe illumination or perspective changes, which may lead to poor localization accuracy. Inspired by the excellent performance of SuperPoint network in feature extraction, a monocular VINS (i. e. , CNN-VINS) is proposed, which is based on the selfsupervised network and works robustly in challenging scenes. Our main contributions are summarized in three terms. An improved SuperPoint-based feature extraction network is proposed. The dynamical detection threshold adjustment algorithm is used to detect and describe feature points uniformly, which can establish accurate feature correspondence. The improved SuperPoint network is efficiently integrated into a complete monocular visual-inertial SLAM including nonlinear optimization and loop detection modules. In addition, to evaluate the performance of the feature extraction network encoder layer in terms of the localization accuracy of the VINS system, learn and optimize the intermediate shared encoder layer and loss function of the network. Experimental results on the public benchmark EuRoc dataset show that the localization accuracy of our method is increased 15% more than that of VINS-Mono in challenging scenes. In simple illumination change scenes, the mean absolute trajectory error is between 0. 067~ 0. 069 m.

Abstract:In order to improve the tracking accuracy of moving targets in video when various interference factors such as deformation, scale variation and similar targets occur, a siamese network model with combined attention is proposed. Firstly, a lightweight network, i. e. , MobileNetV3, is adopted as the backbone network to extract object feature. Then, in order to improve the attention of the model to the key features of the target, a model structure combining channel combined spatial attention and siamese network is proposed. Finally, through weighting and fusing the cross-correlation results of the feature vectors of attention module and non-attention module, the response map can be obtained, which can be used to obtain the tracking result. Experiment results show that the proposed algorithm can achieve good tracking effect on the OTB50 and OTB100 datasets, the average accuracy and success rate for the two datasets reach 78. 5% and 58. 3% , respectively. In addition, when multiple uncooperative factors, such as deformation, scale variation and similar targets exist, the proposed algorithm can still achieve good tracking effect, which shows that the proposed algorithm has good robustness.

Abstract:The real-time operating system - LabVIEW real-time is used to deploy high-speed image processing algorithm, and combining with FPGA ( field programmable gate array) technology, a high-speed real-time detection tracking and stimulation system facing to animal behavioristics experiments is established. The definitive output of continuous frame image is achieved, and the motion target highspeed detection tracking and stimulation in different animal behavioristics experiment scenarios are realized. The application experiments show that the system has good real-time and stability under the acquisition rate of 100 fps, the processing time of the whole process from target detecting to stimulus feedback signal output is 10 ms stably, the average relative error of the execution time is reduced by 15% compared with that of the traditional system. With online analysis the functions of target detection, motion track, mouse behavioral characteristic analysis in the dividing area are completed, and with offline analysis the motion track heat map is generated, which greatly improves the experiment efficiency.

Abstract:The blade flaps during the rotation reflect the performance of the helicopter rotor. The measurement and analysis of the blade flaps in the whole rotation field is a difficult problem in the helicopter rotor test. To address this issue, a new method is proposed, which is based on stereo vision with large field of view ( FOV). Firstly, a stereo vision system with large FOV is constructed and the 3D coordinates of the markers on the blades are measured. Secondly, the blade flaps of the markers are resolved in the hub coordinate system. Finally, the flapping modes of the blade at each azimuth are obtained by using the fourth-order polynomial regression model. The flapping law of the blade during the rotation is achieved by utilizing the regression model of compound sine functions. The proposed method is evaluated by the measurement experiments on the blade flaps of the hovering helicopter. In the 4. 6m × 4. 6m scene, the root mean square error of the blade flap measurement is less than 1mm, and the regression models of the flap mode fit the blade flap data with the root mean square error less than 1mm.

Abstract:The offshore DC microgrid system has requirements of high reliability, low power capacity, low line impedance, and high fault-current rising rate for its detection. To address these issues, a kind of combined short-circuit-fault detection method is proposed, which consists of the differential current method and the directional current method. Short-circuit fault of bus and branch is detected by using the basic law of current, current direction characteristics and variation magnitude of each branch fault. The specific implementation method is given. A scaled-down experiment platform is established. By setting short-circuit fault through controlling IGBT ( insulated gate bipolar transistor), experiments on short-circuit fault of bus and branch are implemented. Experimental results show that the removal time of bus short-circuit fault is 0. 09 ms, and the removal time of branch short-circuit fault is 0. 22 ms. The detection speed is faster than that of the current DC ring micro-grid system short-circuit fault, and it also has good selectivity and wide coverage.

Abstract:To effectively detect the latent trace evidence on porous paper, the ultraviolet fluorescence characteristics of commonly used paper are analyzed, which provides the basis for filtering the strong background fluorescence from paper. A detection device based on ultraviolet laser-induced fluorescence is developed. To effectively stimulate the fluorescence of various trace substances, a deep ultraviolet laser is used as the excitation light source. To obtain more accurate trace information, the device receives multispectral fluorescence and compares the results. Meanwhile, the standard deviation of background gray value is utilized to measure the uniformity of paper fluorescence. The trace contrast of image is used to evaluate the trace detection effect. Five kinds of latent trace on the common paper are detected. Experimental results show that the proposed method can clearly display latent fingerprints on paper, and the contrast of fingerprint ridges is inversely proportional to the standard deviation of paper background gray. At the same time, the trace information of 10% alum solution on students work paper, 2. 5% sodium glutamate solution on copy paper, 2. 5% sodium sulfonate solution on sticky note and 0. 004% copper sulfate solution on sticky note can be clearly displayed. The proposed method has the advantages of sensibility, non-destructiveness, non-contact and multispectral. By utilizing this method, clear images of various kinds of latent trace on paper can be obtained quickly, which can provide powerful help for criminal investigation.

Abstract:A three-dimensional heat conduction model of wind turbine blade with defect is established based on equivalent source theory, which solves the problem that one-dimensional heat conduction model cannot effectively predict the three-dimensional heat flow around the anisotropic material defect, and the accurate evaluation of the defect depth of the large-scale wind turbine blade is realized with pulsed infrared thermal imaging technology. The three-dimensional heat conduction equation of the blade with defect is simplified to isotropic problem with linear coordinate transformation, and then the analytical solution of the surface excess temperature in the defect area under the third boundary condition is obtained using separating variable method. Finally, the quantitative relationship among geometry dimensions of the defect, the peak time of excess temperature and the depth of defect is established. The on-site detection of 1. 5 MW wind turbine blade made of GFRP composite material was carried out, which proves the feasibility of the proposed method, and the detection results show that the detection range reaches 7. 8 mm, the detection error is less than 10% , and the detection accuracy is improved by 10% ~ 31. 4% compared with the one-dimensional model method. In addition, when the defect depth exceeds 3 mm, the boundary heat transfer cannot be ignored, otherwise more than 10. 0% detection error will be caused. The method proposed in this paper can provide a reference for the defect detection of other anisotropic materials.

Abstract:The reconstruction of the region of interest (ROI) by ultrasonic imaging is meaningful to obtain defect features, e. g. , shape, dimension and orientation. In this study, the full-mode total focusing method (FTFM) is proposed by using the strongest response signal among 21 views with different ray paths at each reconstructed point. The profile reconstruction and quantitative detection of unknown cracks are realized by signal acquisition with a set of phased array (PA) probe and wedge. A number of 4 mm length cracks with 25 mm central depth and different angles, e. g. , 0°, ±20°, ±50° and ±80°, are designed in the 40 mm thickness carbon steel specimens. Full matrix capture (FMC) is performed by using 45° wedge and 64-element linear array probe with a central frequency of 5 MHz to obtain FTFM images. The crack profiles are reconstructed intuitively by simulation and practical experiments. Measurement errors of crack length, orientation angle and central depth are no more than 0. 60 mm, 2. 39° and 0. 73 mm, respectively. Finally, appropriate parameters and position for PA probe are beneficial to achieve optimal FTFM image.

Abstract:Aiming at the problems of low recognition rate of surface defects in industrial aluminum product and inaccurate location of small defects, etc. of the traditional detection algorithm, an improved deep learning network called Faster RCNN is proposed to detect 10 kinds of aluminum product surface defects. Firstly, after the data is enhanced, the feature pyramid network (FPN) structure is added to the backbone network to enhance the feature extraction ability for small defects, and then the ROI Align algorithm is used to replace ROI Pooling algorithm to obtain more accurate defect location information. Finally, the K-means algorithm is added to cluster the defect data to obtain an anchor more suitable for aluminum product defects. The experiment shows that the mean of the average precision (mAP50) of the improved network for aluminum product surface defect detection is 91. 20% , which is 16% higher than that of the original Faster RCNN network, and the detection ability of aluminum product small defects is also improved obviously.

Abstract:Aiming at the problems of low aerospace seal ring surface defect detection efficiency of manual inspection and poor versatility of traditional image processing detection algorithms, two kinds of deep learning based surface defect detection algorithms for aerospace sealing rings are proposed. Firstly, aiming at the characteristic that most of the defects are small targets, the RetinaNet network that is more sensitive to small targets is selected as the basic architecture of the defect detection algorithm, and the MoGaA-RetinaNet algorithm is constructed by introducing the lightweight network MoGaA into the RetinaNet network. Secondly, in order to improve the detection accuracy, on the basis of MoGaA-RetinaNet, the newMoGaA backbone network is constructed using the decomposition convolution module to replace the depthwise convolution in the MoGaA backbone network, and the newMoGaA-RetinaNet algorithm is designed. Finally, the experiment results on the test set show that the MoGaA-RetinaNet algorithm has faster detection speed but slightly lower detection accuracy compared with the RetinaNet algorithm; the newMoGaA-RetinaNet algorithm achieves a good balance of detection accuracy and detection speed, Compared with those of RetinaNet algorithm, the detection accuracy rate increases by 4. 5% , reaches to 92% ; the detection speed increases by 55% , reaches to 31 frame / s; and the number of network parameters is reduced by 50% . The newly designed newMoGaA-RetinaNet algorithm can achieve fast and accurate detection of the seal ring surface defects.

Abstract:The complex radar emitter signal recognition methods have problems of poor anti-noise performance, low recognition rate, etc. To address these issues, we propose a new recognition method based on ambiguity function contour lines and stacked denoising autoencoders. First, the ambiguity function is processed by the Gaussian smoothing and the contour lines are calculated by linear interpolation. Then, principal component analysis is used to reduce its feature dimension. The main ambiguity energy information is remained. Finally, deep learning stacked denoising auto-encoders are established to learn and extract the deep and more ubiquitous features of contour lines. The Softmax classifier is used to classify them. Simulation experiments show that the overall average recognition rates of six types of typical radar signals are all above 99. 83% when the signal-noise ratio is 0 dB. The recognition rate can also reach 83. 67% when the signal-noise ratio is -6 dB. Results prove that this method has good performance and feasibility under the extremely low signal-noise ratio conditions.

Abstract:The peak point location is susceptible to many factors of the fluorescence immunochromatographic quantitative image, which can cause the problem of low substance quantification accuracy. To address this issue, a cascaded convolutional neural network (CNN) algorithm for fusion target detection is proposed. The improved AlexNet is utilized in the first-level cascade algorithm to detect and extract the regions containing the quality control (C) peak and test (T) peak in the fluorescence immunochromatographic quantitative image. The extracted image area is sent to the second-level cascaded convolutional neural network to locate C peak and T peak quickly. Then, the location results are taken as the input of the third-level cascaded convolutional neural network. The fine-tune the location results of the C peak and T peak can be realized from the previous layer. Finally, the accurate location information of the C peak and T peak is achieved. Experimental results show that the proposed cascaded convolutional neural network algorithm can locate the peak points of fluorescence immunochromatography images with the accuracy of more than 96% , and the location accuracy of peak points is enhanced.

Abstract:The asymmetric system has stronger signal amplification ability. A new two-dimensional asymmetric bi-stable stochastic resonance (NTABSR) system is proposed. First, the theoretical analysis of the output signal-to-noise ratio (SNR) is implemented under the adiabatic approximation theory. The influence of each system parameter on the SNR of the system is analyzed. Experimental results show that the system can obtain higher output SNR by changing the asymmetry factor while other parameters remain fixed. Then, the system is applied to diagnose fault signals of two different bearings. The system parameters are optimized through the adaptive genetic algorithm, and the detection results are obtained. The final results show that the performance of the NTABSR system is better than that of the TSBSR system. This provides good theoretical support and application value for subsequent theoretical analysis and practical engineering.

Abstract:Temperature control is significant to building energy conservation, and the accurate prediction of indoor temperature is the prerequisite for precise control of building temperature. Proposes a global indoor temperature prediction model based on computational fluid dynamics (CFD) and LightGBM algorithms to realize global temperature simulation and global temperature change prediction over time. The simplified CFD model is based on the space building structure, sensor accuracy range, and actual temperature control range, which can meet the accuracy requirements and solve data redundancy, making it more practical. On this basis, the LightGBM algorithm and LSTM algorithm are used to simulate the global temperature spatial sequence change law. To be specific, the LightGBM algorithm is employed to predict the temperature-time sequence changes to realize the global prediction of indoor temperature. The experiment utilizes the annual building operation data and indoor and outdoor temperature monitoring data of a tobacco storage warehouse to construct an indoor global temperature prediction model. Experimental results of the practical measured temperature data show that the temperature distribution accuracy coefficient of 5 h global forecast is 0. 955 4, and the temperature range accuracy coefficient of 60 h global predict is 0. 994 0. Compared with the ANN, BP, and LSTM algorithms, the average accuracy coefficient of the proposed model is improved by 0. 022 4~0. 014 7.

Abstract:One H-bridge is used to control one coil in a traditional magnetic bearing power amplifier. In order to reduce the number of bridge arms, some of the bridge arms can be shared with vector control technology. In this paper, a power amplifier using three-leg circuit to control two coils simultaneously is studied, and the space vector pulse width modulation (SVPWM) algorithm is implemented with FPGA, which reduces the number of IGBT by 25% . Aiming at the possibly occured situation that the command signal of power amplifiers may exceed the tracking range, this paper analyzes the tracking range of the power amplifier with a three-leg circuit, and points out that the performance of the power amplifier can not be fully utilized using traditional duty cycle restriction strategies and the distortion will be increased. This paper improves the traditional duty cycle restriction strategy, which makes the utilization ratio of the power amplifier reach 100% , and can completely eliminate the distortion of one coil at most. On this basis, two duty cycle restriction strategies with different performance are proposed. Experiment results show that the power amplifier works well when the command signals are within the tracing range, and both duty cycle restriction strategies are effective and achieve the design performance when the command signals are out of the tracking range.

Abstract:In order to solve the problem of low safety of unmanned aerial vehicles ( UAVs ) operating in complex low-altitude environment, a route planning method for UAVs with regional risk assessment was proposed, and it could quickly generate a route with low operational risk. Firstly, the complex low-altitude environment was simplified by model and risk assessment, and the low-altitude three-dimensional risk map was obtained. Taking the path risk value as the comprehensive cost, the improved ant colony algorithm was used to plan the three-dimensional space path, it effectively reduced the redundancy of the generated path. Finally, cubic B-spline is used to smooth the planned discrete path and generate a continuous flightable path with curvature and pitch angle. The simulation and experimental results show that the generated path can be exchanged for a path with a lower risk value at a lower path cost, and at the same time, the path redundancy is low, and the curvature and pitch Angle continuously change, so it is a flight-able path that meets the performance constraints of UAV.

Abstract:The process parameters of injection molding are key factors to ensure product quality. The traditional trial-and-error method relies heavily on the personal experience. The injection molding process is widely used in many important fields, such as electronics, aerospace, etc. The high-end products put forward higher requirements for the intelligent setting of process parameters. Since there are various quality requirements for molded products, and different quality indicators may restrict each other, an intelligent multi-objective optimization method of process parameters is urgently needed to obtain the Pareto optimum among different optimization objectives. Scholars have proposed some intelligent optimization methods. For example, non-dominated sorting genetic algorithms are used to solve multi-objective optimization problems. However, a big amount of sample data are required in such methods to model the qualityparameter relationship. There are problems of a large number of experiments and the poor adaptability of the different materials and molds. To address these issues, proposes a multi-objective self-learning optimization method for injection molding process parameters for the first time. During the optimization process, the gradient of each process parameter is calculated and updated in real time. The multigradient descent algorithm is conducted to optimize different quality indicators. In the optimization process, the self-learning of the influence of each process parameter is realized, which removes the need to perform large numbers of experiments for optimization model establishment. In this way, the efficient intelligent optimization of injection molding process parameters is realized. The relative error between the optimization result of this method and the analytical solution in the benchmark test function is smaller than 2% . Numerical simulation and experimental results show that this method can obtain the Pareto optimum of multiple optimization objectives efficiently.