Abstract:With great accuracy and high precision, the three-dimensional scanning technology has been preliminary applications in workpiece measurements, volume measurements, etc. However, for the objects of large size and low texture characteristics (like transmission line), the three-dimensional point clouds are more sparse, the matching accuracy is still problematic, and the measurement accuracy cannot meet the requirements of the special high-voltage grid construction, which means that its application may have some difficulties in this regard. In order to solve the above problem, the paper establishes the mathematical model of the binocular camera system. The accuracy of the image matching cost calculation is improved by improved Census transform and Gaussian weighting operation. Combined with the image edge preserving filtering method based on texture information, the dense point cloud of the target surface is calculated, and the 3D surface reconstruction of the measured target is realized by splicing the dense point cloud. Through the 3D measurement experiment of 220 kV grid infrastructure, the results show that the measurement error is less than 1mm. the 3D measurement method and system proposed by the papers, which meet the basic quality design and acceptance requirements of transmission line project, can be widely used in the power grid infrastructure project, to improve the accuracy of acceptance.

Abstract:The photovoltaic (PV) array wireless sensor node and fault diagnosis system based on STM32 is designed aiming at the problems of the existing methods of PV array fault detection are lack of accurate and efficient and difficult to locate faulty PV modules. STM32F103 is used as the core controller of the wireless sensor node hardware. It is mainly composed of power module, main control module, data sampling module, data communication module. Using defined communication protocol, the measurement results are reliably transmitted as data frames through ZigBee wireless module, the monitor computer can monitor the voltage data of PV modules in the process of inverter scanning the current-voltage (I-V) characteristics of PV array. In the experiments, the common shading and short-circuit of bypass diode failure of PV modules are artificially caused. The voltage waveforms of PV modules are measured by the wireless sensor node, which can realize fault diagnosis and location in PV array. The experimental results show that the system can obtain the voltage waveform of PV array when the inverter scans the I-V characteristics of PV array, facilitating maintainers to locate PV module faults timely and accurately.

Abstract:At present, manual operation is the main mode of tunnel trolley walking. In order to improve its intelligence and construction efficiency, a trolley path tracking control method based on Fuzzy pure pursuit is proposed. The combination of pure pursuit algorithm and fuzzy control changes the mode of traditional trolley walking by wheel rail or tire manual operation. The abstract mathematical model of trolley is established to determine the turning radius in the process of trolley tracking path, the simulation experiments of tracking straight line and lane changing curve are carried out, and the test trolley is built to test the pure tracking model and fuzzy pure pursuit model, and the experimental results are compared. The results show that the maximum lateral error of the control system proposed in this paper is no more than 5cm in the process of linear path tracking, and converges at the tracking path x = 7.3m. The average absolute error after stabilization is 0.004m. The simulation and experimental results verify that the proposed control system is effective in improving the convergence speed and stability.

Abstract:To address the shortcomings of traditional signal source that can only produce single signal and can not be adjusted, and to meet the testing requirements of a certain type of telemetry device, a design scheme of detection table based on the output controlled by upper computer and adjusted by hand is proposed. This paper discusses the detailed analysis and design of the structure and working pri-nciple of the detection table, as well as a brief introduction of the upper computer. The detection table is centred on FPGA, digital mea-suring board and electric relay with the characteristics of high integration, multiplexing, simple structure and convenient operation. The 13 channels of analog signals are adjusted by the upper computer to achieve 0-30V analog voltage. The 1 channel of -230V~0V analog voltage is obtained by the 220V inverter.  The test results show that the digital output of the test table is completely correct, and the out-put error of the analog output is less than 1%, which fully meets the requirements of the test, and has strong anti-interference ability, good reuse and practical value.

Abstract:When millimeter wave sensor is used to measure respiratory signal of distant human body, it is easy to be interfered by environmental clutter, which leads to more noise in the signal. Therefore this paper proposes a new GA-VMD-WT denoising method. The fitness function has been designed according to the characteristics of breathing signal and then, VMD parameters are optimized by GA algorithm using the fitness function.The optimized VMD parameters were used to decompose the noise signal, subsequently. In the end, the denoised signal can be obtain after the wavelet threshold for the decomposition results.The proposed method not only avoids the over-decomposition problem in VMD decomposition, but also improves the SNR by 8.5025dB, 7.6642dB and 3.3637dB, respectively, compared with other traditional denoising algorithms. Experimental results show that the denoising effect of the proposed method is good, and more useful signal information can be retained.

Abstract:Studies of robot wheelchair human-robot interaction have shown that long-term use of a single mode of interaction can easily lead to misjudgment of user operation intentions and decrease of control stability. The complete autonomous mode can also cause people frustration due to the lack of user control experience. Aiming at the problems of hard mode switching and lack of dynamic adjustment ability to environmental changes in the existing robot wheelchair interaction based on human-robot cooperation, this paper uses the gesture interactive control and proposes a robot wheelchair dynamic shared control method based on the combination of user behavior and autonomous navigation. Firstly, the user's palm coordinates are tracked based on the Leap Motion sensor to generate the user's gesture speed command; Secondly, the autonomous navigation control command is generated based on RPLIDAR A1 lidar sensor and autonomous navigation algorithm; Finally, the weight of human-robot control command is updated in real time based on various constraints such as distance, fatigue and error, so as to realize the dynamic shared control of robot wheelchair. The experimental results show that the dynamic shared control method can dynamically adjust the role allocation between different modes according to the wheelchair operating environment and user operating performance, avoid the direct hard switching between different modes, and has a better user experience.

Abstract:To improve the accuracy of text classification and expand different classification tasks, a text classification method combining one-dimensional convolutional neural network (1D-CNN) and bi-directional long short-term memory (Bi-LSTM) network is proposed. Firstly, in order to solve the difficulty of representing synonyms and polysemy, GloVe model is used to represent word features, making full use of the advantages of global information and co-occurrence window. Then, 1D-CNN is used for feature extraction to reduce the input feature dimension of classifier or prediction model. Finally, the classification module Bi-LSTM is optimized, which hidden layer is composed of two residual blocks, and the attention mechanism is introduced to further improve the accuracy of prediction. Binary classification and multiple topic classification experiments are carried out in multiple public data sets. The experimental results show that compared with other excellent methods, the proposed method has better performance in accuracy, recall and F1 score, with the highest accuracy of 92.5% and the highest F1 score of 91.3%.

Abstract:A motor bearing fault diagnosis model based on an improved Bayesian network is proposed for the problem that the motor bearing vibration signal is affected by noise interference in feature extraction and low accuracy of traditional Bayesian network fault diagnosis. The adaptive ensemble modal decomposition of noise method is employed for noise reduction of data, which increases the model robustness; the Grasshopper Algorithm is optimized by using differential evolution and simulate anneal algorithm to enhance the global and local search ability of the Grasshopper Algorithm; the optimized Grasshopper Algorithm is applied to the Bayesian network structure and learning to construct the bearing fault diagnosis model; through comparing with other methods, it is proved that the method has stronger learning ability and higher accuracy rate for multi-fault classification of bearings, and the experiment fault diagnosis for some samples result reaches 97.15% and the average accuracy rate can reach 98.73%.

Abstract:The structural safety and health inspection of carbon fiber reinforced polymer is very important. Electrical impedance tomography utilizes the conductive properties of carbon fiber composites to detect structural damage. This technology has the advantages of low cost, non-invasive and simple operation, and has become a research hotspot of scholars in recent years. There are serious ill-conditioned problems in the image reconstruction process of EIT. In this paper, an electrical impedance tomography algorithm with improved L1/2 regularization operator is proposed. The method uses the L1/2 norm to construct a sparse regularization penalty function, and uses the Bregman alternating direction method of multiplier iterative algorithm to solve the new objective function to improve the performance of the algorithm. To verify the effectiveness of the algorithm, four typical damage types of CFRP laminates were designed using COMSOL software and a 16-electrode EIT test system was built for verification. The simulation and actual experimental results show that, compared with other algorithms, using the BADMM iterative algorithm to solve the L1/2 regularization method improves the image correlation coefficient, reduces the image error, and effectively improves the accuracy of the reconstructed image, and the method has strong robustness to noise.

Abstract:To address the problems of untimely indoor air quality regulation and high energy consumption of ventilation system, this paper designs and implements a fuzzy PID-based indoor air quality measurement and control system. Firstly, the functions of data collection, data communication and fresh air control are achieved with STM32 microprocessor as the core, combined with various air quality sensors, communication module and control module, respectively. And the air quality software display platform is designed based on Java Web technology. Secondly, a fuzzy PID control algorithm is designed by selecting the representative indoor CO2 and TVOC as pollutant control indexes. Then, simulation verifies the effectiveness of the fuzzy PID controller by simplifying the ventilation system to a first-order pure hysteresis inertia link. Finally, the experimental results show that compared with the traditional control method, the system not only has fast regulation speed and high stability, but also saves 22.2% of system energy consumption, which has some application value.

Abstract:In order to reduce the impact of individual characteristic differences on the accuracy of the non-invasive blood pressure prediction model and improve the prediction accuracy, a Bayesian optimization (BO) XGBoost non-invasive blood pressure prediction method is proposed. Firstly, the multivariate linear model is established to obtain preliminary blood pressure prediction values based on pulse transit time (PTT) and body mass index (BMI). Then combine human characteristic parameters as the input of XGBoost blood pressure prediction model. Then use Bayesian optimization to automatically optimize XGBoost hyperparameters. Finally， the BO-XGBoost model is used to predict the blood pressure and compare with other methods. The experimental results show that the average absolute error of diastolic and systolic blood pressure based on the BO-XGBoost blood pressure prediction model meets the standard of less than 5mmHg formulated by AAMI (American Medical Instrument Promotion Association), which is better consistent with the method of mercury sphygmomanometer.

Abstract:The P-V curve of photovoltaic system is multimodal due to the effect of partial shading. This reduces the tracking efficiency of the traditional maximum power point tracking algorithm. To handle this effect, this paper proposes a two-layered maximum power point tracking algorithm. The differential evolution algorithm is placed in the slave layer, whereas the gray wolf optimization algorithm is placed in the master layer. In order to search the optimal duty cycle that maximizes the power output of photovoltaic system, the methods of replacement and feedback are employed to strengthen the cooperation between two algorithms. Firstly, the duty cycle is considered as the individual and the gray wolf of each algorithm, respectively. Then, differential evolution algorithm is used to search multiple groups of individuals rapidly, and the positions of the wolves in master layer are replaced by the best duty cycle in each group. Finally, the grey wolf optimization algorithm is employed to optimize the positions of wolves, and the α wolf is feed back to the slave layer. This can guide the update of individuals in the slave layer. With the platform of Matlab2017a/Simulink, the proposed algorithm is applied to simulate four cases under different magnitudes of shading. The results indicate that the efficiencies of proposed algorithm are 99.63%, 99.91%, 99.41%, and 99.95% in four cases, respectively. All these efficiencies are above those of other three existing algorithms. The energy production of photovoltaic system can be well improved by the proposed algorithm.

Abstract:Aiming at the problem of increased power consumption and reduced reliability of DDS chips due to large storage space overhead, a ROM storage space compression optimization for direct digital frequency synthesis (DDS) waveform generators on field programmable gate arrays (FPGA) was designed. algorithm. Under the premise of not changing the waveform precision, the ROM is compressed by storing the relative increment of the amplitude sequence to reduce the waveform data bit width, and then the amplitude accumulator can be used to restore the real amplitude sequence. Build the project in the Quartus II 13.0 development environment and pass the test on the FPGA development board. After testing, the DDS signal generator can generate five different waveforms, occupying a total of 9240bit storage space. The results show that this DDS optimization algorithm saves more than 96% of resources compared with the traditional DDS waveform generator, which can reduce the system power consumption and improve the system running speed.

Abstract:An overvoltage and undervoltage protection circuit for low-voltage systems is designed. The output voltage threshold accuracy is improved through a circuit structure similar to a bandgap reference, low-voltage conditions are achieved by reducing the use of diodes, and a certain hysteresis is designed in the undervoltage protection threshold voltage of the circuit to prevent the system repeatedly turning off near the threshold, and the functional correctness of the circuit is verified in the simulation of the low-voltage power management system application. The circuit is designed and simulated using SMIC 0.18 μm standard CMOS process and Cadence software. The results show that when the power supply voltage rises within the temperature range of -40~85 ℃, the circuit starts to work normally and the voltage threshold is 2.431~2.50 V, the threshold accuracy is 3.2%, the overvoltage protection threshold is 4.931 ~5.22 V, the accuracy is 5.7%, when the power supply voltage drops, the undervoltage detection threshold is 1.58~1.663 V, the accuracy is 4.9%. The system application simulation results show that when the power supply voltage exceeds the normal operating voltage threshold of the system, the over-voltage and under-voltage protection circuit will control the system to enter the shutdown mode and realize the protection function.

Abstract:In order to realize the high-resolution imaging of the optical synthetic aperture imaging system, the phase misalignment error between the sub-apertures must be corrected. Using monochromatic light illumination, establish a dual-aperture imaging system model, analyze the influence of translation error and tilt error in the phase misalignment error on the optical system, use the integral value of the system modulation transfer function as the evaluation function, and use the simulated annealing algorithm to perform online phase misalignment error Calibration simulation, system performance evaluation of the optical system after calibration. The simulation results show that given a specific phase misalignment error, the simulated annealing algorithm can correct the phase misalignment error of the optical system well. Due to the randomness of the algorithm, 50 sets of random errors are used for calibration simulation. The results show that the integrated value indicators of the modulation transfer function of the corrected system can all converge, the corrected Strehl ratio is greater than 0.8, and the peak-to-valley value of the wave front is less than 0.57λ, The mean square error of the wavefront is less than 0.10λ.

Abstract:Aiming at the binocular vision system of the space manipulator of the insulated bucket arm vehicle, the optimal design and calibration of the parameters of two panoramic cameras in the vision system are studied in this paper, and a model is constructed after the optimal configuration of the camera position, direction and mirror shape in the system, so as to obtain higher accuracy of 3d data measurement. In order to construct the optimal configuration of the visual system, the analytical formula of the 3D measurement error model is derived according to the error propagation analysis, the quantization accuracy of image pixels and the variation of angular resolution in the process of data calculation. Then, this formula can be used in framework optimization to find the best system configuration for different shapes of system Settings. For the general case with cuboid 3D measurement area and camera placement area, the suboptimal solution of system configuration is derived by using the proposed analytical formula, and the accuracy of the suboptimal solution is proved to be close to its optimal solution. Finally, the experimental results of simulation and practical application cases show that the proposed method can effectively improve the accuracy and robustness of binocular ranging, and the running speed is fast.

Abstract:In order to solve the problem of efficient and accurate measurement of aerosol mass concentration, an aerosol photometer is developed based on the principle of measuring aerosol mass concentration by light scattering method. The constant flow control module is designed. Based on the principle that the forward scattered light of aerosol by laser irradiation is proportional to the mass concentration, the optical signal is converted to electrical signal by photomultiplier tube, and the linear coefficient is calculated according to the high-precision comparison instrument results under the same measurement conditions. The experimental results show that the indication error of the aerosol photometer system developed in this paper can be within ± 5%, and the repeatability deviation is within 3%, which meets the requirements of aerosol photometer calibration specification, realizes the measurement of aerosol mass concentration, and has good application value.

Abstract:This paper introduces the origins and derivation of cross capacitance principle. Its development for capacitor benchmarks and the gradual reduction in biasness while measuring 1pF capacitance to 10-8 was discussed. Furthermore, the methods, principle and application of cross capacitor in sensor manufacturing and in micro capacitance detection methods was reviewed. Compared with the traditional capacitance sensor, the cross-capacitance sensor is classified as follows: variable medium type, variable area type and variable plate spacing type, and its advantages and disadvantages are analyzed. Based on the analysis of the existing cross-capacitance sensor research, how to further optimize the cross-capacitance sensor and its development direction are discussed。By summarizing and summarizing the existing cross-capacitive sensors, the capacitive sensor system is more perfect and a reference is provided for the follow-up cross-capacitor research.

Abstract:Defect detection algorithms based on deep learning technology often need a large number of image samples to train the model because of many network parameters. However, in the process of industrial production, the number of defective products is very small, and collecting a large number of defect data images is time-consuming and laborious. To solve this problem, this paper proposes a lightweight defect detection algorithm based on multi model cascade. It adopts the training method of supervised learning, and can obtain better detection results through a small number of defect samples. Firstly, CBAM attention residual module is used to extract features instead of conventional convolution layer to focus on defect features and strengthen the characterization ability of network to defects; Secondly, the SE-FPN module is designed to promote the effective integration of features at all levels and improve the segmentation effect of network on defects, especially for small defects; Finally, in the training stage, the supervised learning method is used to train the multi model algorithm network proposed in this paper. The experimental results show that the detection accuracy of the proposed algorithm on KolektorSDD data set is as high as 99.28%, and the average detection time of each image is only 10.5ms. It not only fully meets the requirements of high precision and real-time in the industrial detection industry, but also realizes the accurate positioning of defect areas. Therefore, the research content of this paper is very suitable for application in the field of on-line detection of surface quality of industrial products.

Abstract:Conveyor belt tear detection is a very important part of coal mine safety production. In this paper, a new method of detecting conveyor belt damage named audio-visual fusion (AVF)detection method is proposed. The method uses both a visible light CCD and a microphone array to collect images and sounds of the conveyor belt in different running states. By processing and analyzing the collected images and sounds, the image, and sound features of normal, tear and scratch can be extracted respectively. Then the extracted features of images and sounds are fused and classified by machine learning algorithm. The results show that the accuracy of AVF method for conveyor belt scratch is 93.66%, and the accuracy of longitudinal tear is higher than 96.23%. Compared with existing methods AVF method overcomes the limitation of visual detection condition, and is more accurate and reliable for conveyor belt tear detection.

Abstract:Aiming at the problems of low accuracy and high missed rate of the existing traditional algorithms in traffic light recognition under the condition of uneven illumination and complex background, this paper proposes a traffic light recognition method based on image enhancement.First, an improved iterative method is used to process the original image.Then, the v-channel brightness information of the original image is enhanced in the HSV color space, and the color candidate regions of traffic lights are screened by self-adjusting color threshold range.Finally, the contour of the original image obtained after dual processing was extracted respectively, and the candidate area with the same information was screened by combining the contour information of the two images, and the area of the traffic signal light was judged by calculating the area and aspect ratio of the candidate area, so as to complete the identification of the traffic lights.Experimental results show that the proposed algorithm can improve the recognition accuracy by 1.05% compared with other traditional algorithms under the condition of uneven illumination and complex background, and has good real-time performance.

Abstract:At present, micro flat motor manufacturers still use manual observation of motor FPC surface welding quality for classification, its detection accuracy is low, slow speed. To solve this problem, a defect classification detection method based on improved Faster R-CNN was proposed. Firstly, the last two layers of VGG16 are fused by multi-scale feature fusion network to replace the input feature graph of the regional proposal network in the original Faster R-CNN. Then, the accuracy, recall rate and score of the network are compared from three multi-scale feature fusion algorithms with different depths. The experimental results show that the average accuracy of defect classification detection of the improved two-layer multi-scale fusion feature map is 91.89%, 7.72% higher than that of the traditional model. Compared with the other two models, the improved model has the highest classification detection accuracy and precision.

Abstract:Crowd counting is widely used in video surveillance, public security, intelligent business and many other fields. In recent years, with the continuous development of deep learning, crowd counting has become one of the hot topics in the field of computer vision. In this paper, according to the different feature extraction methods, crowd counting is divided into two categories: one is traditional method, the other is based on deep learning method, and the method based on convolutional neural network is analyzed and introduced. Further introduces the population count in the field of benchmark data sets and other representative data sets, the experimental results show that the larger changes in the crowded and scale, based on the convolution of the neural network method is superior to the traditional method, the scale change is bigger, more complex scenarios crowd more columns than a single network count more accurate, more effective; Finally, the future development direction of the algorithm is discussed.

Abstract:UAV plays an important role in wireless networks due to the advantages of high mobility and high line-of-sight communication probability. However, in a multi-UAV system, the traditional user association method cannot meet the on-demand deployment of UAV, and there is a problem that the load is imbalanced. To this end, a utility function is first constructed, which comprehensively considers three factors: the user's received signal-to-interference-to-noise ratio, the UAV’s load, and the spatial dispersion of the set of users served by the same UAV. Second, network load balancing is achieved by solving the utility maximization problem. In order to solve this mixed integer nonlinear non-convex problem, a load balancing algorithm based on utility function is proposed, which decomposes the original problem into two sub-problems of user association and UAV location optimization for iterative solution. Given the location of the UAV, a user association algorithm based on the maximum utility is proposed to obtain the best user association scheme. Based on the current best user association scheme, an improved distributed ordinal location optimization algorithm is proposed to obtain the best UAV location. Finally, by solving the user association and UAV location optimization sub-problems iteratively, the optimal user association scheme and UAV location can be obtained. The simulation results show that the proposed algorithm improves the load balancing level by 52.56% and 7.63% respectively compared with the maximum SINR association method and the "maximum SINR+UAV location optimization" method, which has the advantage of significantly improving the load balancing effect.

Abstract:In recent years, wireless signals using WiFi Channel State Information have played important roles in scenarios such as indoor positioning, fall detection, and identification. However, the impact of multipath effects in complex environments makes the accuracy of fingerprint positioning to be improved. To solve this problem, this paper proposes an improved Density-Based Spatial Clustering of Applications with Noise during the process of noise reduction combined with Enhanced weighted K-nearest neighbor algorithm in the online stage. First, the Hampel algorithm is used to remove outliers of the amplitude information; then, the improved DBSCAN algorithm automatically adjusts the parameter to cluster data; finally, the Enhanced weighted K-nearest neighbor algorithm is used to match the real-time positioning points from the fingerprint database. The experimental results show that the average positioning accuracy of the DBSCAN algorithm reaches 1.579m in a positioning area of about 5×10m2, and the percentage of error within 2m is increased by 42.9% compared to the traditional fingerprint method.

Abstract:Three-line HF wide band antenna is a new type of short-wave communication antenna. In order to give full play to the performance of the antenna in omni-directional communication and directional communication, this paper first analyzes and calculates the radiation impedance and antenna pattern of three-line antenna at different work frequencies by using the method of moments, and obtains the curve of antenna directivity coefficient varying with working frequency, so as to effectively guide the frequency selection of antenna in omni-directional communication or directional communication; The simulation model of three-line antenna and typical ground objects is established by using electromagnetic simulation software FEKO, and the distortion law of antenna pattern under the influence of typical working frequency and different ground objects is obtained. It is concluded that the influence of height parameters on three-line antenna pattern is the most obvious, which provides a reference for engineering practice such as location selection and erection of three wire antenna.