Abstract:The traditional machine learning methods for wearable sensorbased human activity recognition tasks usually require manual feature extraction, and the deep neural network that can automatically extract human activity data features is becoming a new research focus. At present, DeepConvLSTM, which is combined of convolutional neural network (CNN) and long shortterm memory (LSTM) recurrent neural network, has better recognition accuracy than other recognition methods. To solve the difficulty of training neural networks with long shortterm memory recurrent unit, the paper proposes a fusion model based on convolutional neural network and gated recurrent unit (GRU), and the performance on three public data sets (ACT data set, UCI data set and OPPORTUNITY data set) is compared with convolutional neural network and DeepConvLSTM. The experimental results show that the recognition accuracy of the model on three public data sets is higher than that of convolutional neural network and is close to DeepConvLSTM, but the convergence speed of the model is faster than that of DeepConvLSTM.

Abstract:The uniformity of magnetic field is an important factor affecting the detection accuracy of inductive abrasive sensor. In order to improve the uniformity of the magnetic field of the sensor and study the distribution of coil parameters on the magnetic field of the sensor, the mathematical model of the threecoil of the inductive abrasive particle sensor is established according to the principle of magnetic field superposition. The calculation formula of the magnetic field of the sensor is derived and the evaluation coefficient K of the magnetic field uniformity is proposed. The magnetic field of the inductive abrasive particle sensor was simulated, and the influence of the inner diameter of the coil, the width of the coil, the spacing of the coil and the number of turns of the coil on the magnetic field distribution of the sensor was analyzed. Through the range analysis and analysis of variance, the significance of the optimal combination of coil parameters and the influence of coil parameters on the uniformity of the magnetic field are obtained. The experimental results show that when the coil is the optimal parameter, the sensor has good monitoring ability for the abrasive grains in the range of 70 to 200 μm. The deviation of the abrasive particles in different radial monitoring results is less than 5%, which provides a reference for the design of the sensor coil.

Abstract:In order to achieve the road surface flatness measurement, the magnetic levitation technology was used to measure the vibration of the car during driving to achieve road surface flatness measurement. A magnetic levitation measurement model of double magnetic suspension oscillator was designed. The force relationship of the magnetic levitation oscillator was obtained by experimental data, and the equation of motion of the magnetic levitation oscillator was established. A simulation model of the magnetic levitation measurement system was established. A flat roads, narrow speed bumps, wide speed bumps, urban roads and longdistance travel measurements were measured using the designed measurement model. The wavelet analysis method is used to perform 5 layers of wavelet decomposition on the measured vibration signals to obtain approximating signals and detail signals and wavelet reconstruction of the measured waveform. Road features were obtained in five different situations. This method provides a new idea for the road flatness detection method.

Abstract:Brillouin optical time domain relectometer (BOTDR) based on short time Fourier transform (STFT) algorithm can achieve fast temperature detection. However, there are frequency leakage and fence effects, resulting in poor temperature measurement accuracy. In view of the above problems, this paper builds on the STFTBOTDR temperature measurement system. Through the optimization of window function and operation point number, the frequency leakage caused by STFT algorithm is suppressed, and the accuracy of STFTBOTDR temperature measurement is improved. In the experiment, the time domain window length is set to 16 m, and the window sliding step is 05 m, which compares the measurement accuracy under different window functions and operation points. The results show that when the Hanning window is used and the number of calculation points is 1 024, the accurate detection and positioning of the temperature change of the fiber end of 96 km can be realized with an error of 1012℃. The measurement accuracy is ±25 MHz. When the window function is not used, the measurement accuracy is ±125 MHz, and accurate measurement of temperature change cannot be achieved. The research results provide a reference for the optimization of the accuracy of the STFTBOTDR temperature detection system.

Abstract:A zeropoint fault of load cells in truck scale is a typical minor fault and it is difficult to be detected online. A method for detecting zeropoint fault online is proposed by combining a recursive principal component analysis (RPCA) with four types of fault detection indicators. In this method, firstly, the principal component model is updated online by the principal recursive algorithm based on rank 1 modification, and then the four statistics, i.e., the Hotelling's T2 statistic, the squared prediction error (SPE) statistic, the Hawkins TH2 statistic, and the principal component related variable residual (PVR) statistic, are used to construct a comprehensive evaluation method for fault detection. This proposed method for fault detection online is applied to load cells in truck scale, and the experimental results show that the accuracy of zeropoint fault detection is increased with an order of magnitude by the traditional method, which proves the effectiveness of this proposed method.

Abstract:In order to improve the positioning ability of mobile Internet of Things sensing terminals and reduce the deviation caused by channel noise, this paper proposes a coordinate sensing algorithm for mobile Internet of Things terminal based on comprehensive weight calibration and mean jump estimation. Firstly, the localization data message is divided into three parts: coordinates, hops and ID of the localization node. The localization node parses the localization data message and obtains the hops and coordinates between the localization nodes. When only the localization node receives multiple localization data messages at the same time, the localization data message is divided into three parts. The coordinates of the positioning nodes are analyzed, and a triangular positioning method is constructed according to the coordinates of the received three positioning nodes. The coordinates are obtained accurately and the coordinate jitter is reduced. Subsequently, according to the mean square error theory, a neighborhood radius prepositioning method based on weight differential correction is designed to accurately obtain the positioning coordinates. Finally, according to the weight difference of different locating nodes in the network, a terminal coordinate correction scheme based on average hop estimation is designed. The equalization mechanism is adopted to enhance the network’s ability to capture nodes, and the inductive processing method is used to optimize the locating equations and modify the node coordinates, so as to improve the terminal sensing accuracy of the Internet of Things. The simulation results show that the proposed algorithm has lower deviation degree of node coordinates and better trajectory prediction ability than the current Internet of things node positioning technology.

Abstract:As the core component of power converter, the reliability of insulated gate bipolar transistor (IGBT) is widely concerned. High junction temperature or high temperature fluctuation are the main factors leading to the failure of power devices. Therefore, accurate monitoring of IGBT module junction temperature in real time is of great significance to improve its reliability. The junction temperature is obtained by building a thermal network model and simulating its circuit. This method is widely used because of its simplicity and feasibility and the realization of online monitoring of junction temperature. In this paper, the existing thermal network models are summarized. According to the model structure, they are divided into three categories: onedimensional, twodimensional and threedimensional. Then, the research situation of parameter identification methods for thermal network model in recent years is discussed, and each method is compared and evaluated. On this basis, the future research directions of new thermal network models and model parameter identification methods are prospected.

Abstract:The traditional convolutional neural network (CNN) model has a large amount of spatial feature information redundancy in remote sensing scene classification, which greatly affects the classification accuracy and operational efficiency of the model. In view of this problem, the paper proposes a DCNN model based on octave convolutional(OctConv). Firstly, the feature map outputted by the convolutional layer is decomposed into two parts according to the frequency, and using global average pooling to compress the lowfrequency part with less feature mapping information into a quarter of the current size, then using OctConv to replace the traditional convolution operation, to achieve highlow frequency feature selfrenewal and information interaction, finally, introducing transfer learning to improve the robustness of the model and making up for the lack of data. The experimental results show that the proposed method can achieve 9925% classification accuracy under the UC_merced_Land_Use public data set, which is 2% higher than the same type method, which shows the superiority and effectiveness of the method.

Abstract:A nonlinear ultrasonic testing method based on the finite amplitude method is proposed for testing micro fatigue damage, which is hard to be distinguished by the traditional ultrasonic testing. First, the Detection platform which can be used to adjust probe’s acoustic coupling performance is designed to keep the stability of the nonlinear ultrasonic detection signal. Second, the relative nonlinear coefficient of the microdamage zone is extracted under various excitation voltages to investigate the influences of the excitation voltage on the testing effect. Finally, the testing abilities of the nonlinear ultrasonic testing on the microdamage are discussed combining with the metallographic measurement. The results indicate that the relative nonlinear coefficient can effectively characterize the micro fatigue damage, and the excitation voltage is the key detection parameter of the nonlinear ultrasonic finite amplitude method, which decides the detection efficiency and detection resolution.

Abstract:In order to improve the detection probability of seismic wave signals in low SNR environment, combining with adaptive line spectrum enhancement technology, this paper proposes a new realtime detection algorithm for ship seismic wave signals based on sliding power spectrum. Firstly, the signal is processed by adaptive line spectrum enhancement technology, then the power spectrum estimation is performed on the processed signal, the average power in a frequency band is selected as the detection feature quantity, and the corresponding adaptive detection threshold operator is designed. Finally, realtime detection of ship seismic wave signals is completed according to certain detection rules. The measured data show that the detection algorithm can effectively accomplish target detection with low signaltonoise ratio, and the detection probability reaches 93% when the signaltonoise ratio is -15 dB.

Abstract:An intelligent fault diagnosis method used for magnanimous monitor data is proposed to deal with the problems of power transformer fault diagnosis. Firstly, a selfpowered RFID sensor tag is designed to measure the transformer vibration signals, which has advantages of simple structure, convenience and low cost. The measured transformer vibration signals have characters of large quantity, high dimension, complex components and low signal to noise ratio. A stacked autoencoder (SAE) of deep learning is employed to extract features of the measured vibration signals, where features of the same status are highly aggregated and features of the different statuses are obviously separated. A weighted native bayes (WNB) classification model is employed to the transformer fault diagnosis based on the extracted magnanimous feature data. To further improve the performance of fault diagnosis method, chaotic quantumbehaved particle swarm optimization is proposed to optimize the parameters of SAE and WNB classification model, respectively. A 10 kV transformer fault diagnosis results show that the proposed RFID sensor tag can reliably collect the vibration signals, and the fault diagnosis method has a high correct rate of fault diagnosis.

Abstract:Signal integrity is important in highfrequency and highspeed circuits. The discontinuity of differential vias seriously affects signal integrity. According to the design requirements for lowreflection, hightransmission and impedancestabilization of differential vias for differential and commonmode signals in highspeed printed circuit boards (PCB), firstly, the equivalent physical model and the circuit model of the differential vias are established to analyze the differential signal and the common mode signal of the differential vias. Then, based on the PCB stack structure and wiring pattern design, threedimensional electromagnetic simulation software HFSS is used to set different vias center distance, antipad diameter and ground vias number. Time domain impedance, return loss, and insertion loss of differential vias are simulated and analyzed. Besides, Sparameters and impedance changes in the time domain are used to analyze the differential performance and commonmode performance of vias. Finally, through the simulation results, it is found that the center of the vias hole is 38 mils(1 mil=0025 4 mm), the antipad diameter is 32 mils, and the set of double vias and ground vias optimizes the performance of differential signals and common mode signals. This paper presents a new idea for optimizing the performance of differential vias and provides a reference for highspeed differential vias design.

Abstract:The rubidium atomic magnetometer based on nonlinear magnetooptical rotation (NMOR) effect can obtain the Larmor frequency and external magnetic field value through the free relaxation sinusoidal oscillation signal with specific time length, thus, interceptlength of relaxation signal can affect the performance of the atomic magnetometer. In this paper, the influence of the interception time of the free relaxation signal on the sensitivity and the measured magnetic field was studied near the background magnetic field of 500 nT. The measurement results show that the sensitivity calibrated by the noise power spectrum decreases gradually and reaches a stable value of 023 pT/Hz1/2with the increase of the interception time, the measurement of the alternating magnetic field shows that the field values measured according to the free relaxation signals with long interceptionlength represent the average results. Therefore, the high sensitivity index of the NMOR atomic magnetometer in this paper is obtained to some extent through mathematical processing to average the fluctuation of the magnetic field value. This method can be used for reference to calibrate the sensitivity of other atomic magnetometers or experimental devices and analyze the measured magnetic field values.

Abstract:In the switched reluctance motor speed control system, PI controllers often exhibit problems of poor dynamic performance, low control accuracy and large torque ripple. The PID controller can effectively improve the control performance, but because the PID controller has a differential link, a large system noise is likely to cause oscillation, resulting in unstable operation of the motor. To solve this problem, a parameter optimal fractionalorder PIDμ controller is designed based on conventional PI controller. The speed response of the system is selected to construct the target optimization function, and the gray wolf optimization algorithm is used to adjust the controller parameters to obtain the optimal parameters. The double loop model of SRM speed control system is established, the fractionalorder PIDμ controller is adopted in the speed loop and a compound current controller and PWM controller are adopted in current loop. The performance of conventional PI controllers and fractionalorder PIDμ controllers is compared in the case of sudden changes in speed and load transients. Simulation experiments show that the fractionalorder PIDμ controller retains the advantages of simple structure and convenient design of the PI controller. At the same time, it exhibits faster response speed and control precision, and has a good inhibitory effect on SRM torque ripple.

Abstract:This paper presents an image reconstruction method combining convolution neural networks (CNN) with anchored neighborhood regression (ANR), aiming at the shortcomings of the conventional anchored neighborhood regression (ANR) image superresolution method, which is inflexible and incapable to restore image details. Firstly, the elastic network regression model is proposed in ANR to give the algorithm with the characteristics of feature selection. Secondly, the lanczos3 interpolation method is used in the part of image preprocessing of CNN to accelerate the operation speed. In the feature extraction, the Swish function with selfgating characteristics is proposed as the activation function to improve the test accuracy. Finally, the correlation coefficient of the image is proposed in the evaluation of the reconstructed image and used for further evaluation of the reconstructed image. The experimental results show that the average PSNR, average SSIM and average correlation coefficient of the proposed method reach 0.982 8, 0.968 and 0.938 0 respectively. The algorithm effectively restores the details of the image and the image quality is further improved.

Abstract:With the development of sensor array technology, more and more array technology is used in the fault diagnosis research of equipment. In this paper, based on the fault characterization and positioning problem in multirolling bearing diagnosis in equipment, a theoretical study on bearing fault location theory based on 2DMUSIC algorithm Ltype acoustic array is carried out. The main purpose of the research is to optimize the array parameters. According to the acoustic signal characteristics of the rolling bearing, the 2DMUSIC algorithm is used to simulate the dual source signal under the Ltype microphone array. Then, by changing the parameters of SNR, array spacing and array number, the optimal acoustic array parameters are determined. The simulation results show that, under the condition that the two signal sources have been determined, by optimizing array parameters, fault location resolution can be improved and sound source location system can be optimized. The research has a good theoretical guiding significance for the rapid positioning and realtime diagnosis technology of bearing faults in multibearing drive systems of rail locomotives.

Abstract:For the tracking problem of radar working state by electronic reconnaissance system, a radar working state tracking method based on fused hidden Markov model is proposed. First, the radar working process is modeled as a hidden Markov model by this algorithm. Second, by recognizing the reconnaissance radar phrase sequence, the tracking of radar working state is realized on a single platform. Finally, the DS evidence theory is used to fuse the recognition results of multiplatform to realize the multiplatform fusion tracking. The recognition rate of the algorithm is simulated, and the simulation result shows that the proposed algorithm can improve the tracking accuracy of radar working state under error observation. When the error rate of observation is 20%, the tracking accuracy reaches 93%.

Abstract:At present, many image duplicationpaste tampering detection algorithms mainly rely on the gray level information of the image to detect image features, but do not consider the color factor of the image, resulting in the deficiency of error detection and missing detection in the detection results. Based on cosine modulated Gauss filtering, an image copypaste tampering detection algorithm based on color metric factor coupled with local feature clustering is designed in this paper. The CMG is used to obtain the scale response value of the image, and the candidate feature points are extracted by the extremum calculation. The spectral reflection model of the pixels is used to establish the color measurement factor, which is used to determine the image feature points from the candidate feature points. The neighborhood circle of the feature points is constructed and the quaternion exponential moments in the circle are obtained to form the feature vectors. The Euclidean distance between feature points is calculated by using eigenvectors to complete image feature matching. By using the R, G and B values of matching point pairs, the local features of feature points are formed, the clustering of image features is completed, the forgery content is located and copied and pasted, and the tampering detection results are obtained. The simulation results show that compared with the current copypaste forgery detection method, the proposed method has higher detection accuracy and robustness for simple copypaste forgery and complex combination forgery.

Abstract:A direct instantaneous force control scheme based on FOA-RBF was proposed for the tubular linear switched reluctance actuator with radial flux. The FOA-RBF was trained offline by the samples obtained from bench measurements, and nonlinear mapping from current and translator position to output force was completed. After training, FOA-RBF was applied for the realtime estimation of output force, and dynamic adjustment of required force between adjacent phases was realized with the combination of force distribution function. The output force ripple and peak phase current were effectively restrained. The proposed control algorithm makes use of the advantages of generalization and approximation ability of FOA-RBF. Fast force estimation and high precision are achieved, which can meet the requirement of realtime control. Based on the actual measurements of force and electromagnetic characteristics, the system simulation model was constructed and applied to verify the control method. Finally, the experimental platform was built, and the effectiveness of FOA-RBF and proposed control algorithm is further verified.

Abstract:Since the news content in the field of shipbuilding industry is long and professional, and contains a large number of professional vocabulary, there is currently little research on the classification of news texts in this field and the lack of corresponding shipping industry news corpus. This paper builds a shipping industry news corpus, and proposes a new text classification algorithm for ship industry news. Firstly, based on document frequency, chisquare statistic and topic model LSA, it conducts feature selection and feature dimension reduction, after mapping the documentword matrix into the documenttopics matrix, the processed features are finally classified by using support vector machine. Experiments on the classification of news texts show that the proposed algorithm can effectively solve the problem of high dimensional and high sparsity of text vectors and has better classification effect than traditional methods under the premise of small sample sets and limited categories.

Abstract:Aiming at the crack problem, a translationtorsion coupling dynamics model of singlestage planetary gears was established by the lumped centralized method, considering timevarying meshing stiffness, support stiffness, integrated meshing error and meshing damping. The response characteristics of a single planetary gear under different crack degrees were analyzed, and the response characteristics of the system under different planetary gear failures were analyzed. The results show that, when a single planetary gear has a crack failure, the double frequency is the main energy. As the degree of crack increases, the vibration of the system becomes severe, and the proportion of the energy of the double spectrum increases gradually, the phase diagram gradually changes from the inner character to the singlecircle curve, and the system motion tends to be stable. When the number of planetary gear failures increases, the vibration severity of the system becomes smaller than that of a single faulty planetary gear, and the spectral energy distribution changes more obviously. The proportion of the double frequency energy is getting smaller and smaller and the internal curve is gradually separated from the external curve. The research in this paper has certain theoretical significance for the crack fault diagnosis of singlestage planetary gearbox.

Abstract:Aiming at the problem of speechlike signal synchronization in encrypted communication across cellular mobile networks, a synchronous scheme combining time domain analysis with fast correlation method is proposed. On the basis of establishing the twostage synchronization model, the synchronous signal capture is implemented via the analysis of time domain characteristics of sinusoidal signal and the postcalibration algorithm. Furthermore, the accurate positioning of synchronous signals is achieved through the correlation calculation of Linear FM signals and the synchronous deviation compensation algorithm. Experimental results show that both of the error rate and leakage rate in the detection of synchronous signals are less than 001%, and the ratio of deviation that more than 5 samples is less than 003% in GSM and LTE network. The proposed scheme has good detection performance and low computational complexity.

Abstract:Aiming at the problems of missing and false detection encountered in image recognition on the forest fire, a sample entropy discrimination algorithm based on K-Means clustering was proposed. First, the collected forest fire images were transformed into color gamut space, which reduced the influence of visual deviation in the process of image recognition. Then, the K-Means clustering algorithm was adapted to cluster the image subset that was expected by fire through the Euclidean distance criterion of HSV components. On this basis, the weight of the clustered image subset was identified by using the sample entropy, the entropy values of the correlative fire regions and the real fire regions were statistically distinguished. Then it was confirmed whether there was a fire in the subset of images screened by the cluster. The experimental results showed that by using the sample entropy fusion K-Means clustering algorithm, the recognition accuracy can be effectively improved in forest fire image recognition. After the detection of 60 images, the correct identification rate of fire area in all images was improved to 96.67%, and the average identification time was 16.03 s. Due to the strong robustness and convenience of the algorithm, it is able to adapt to the identification of the fire area under complex background and has better detection effects than the traditional K-Means algorithm.

Abstract:Detection of the loose particles is urgently required in the Aerospace seal relay production processes. Particle impact noise detection (PIND) is a national aerospace electronic component loose particles detection method. Aiming at the misjudgment of the loose particles signal and component signal in the traditional detection method, this paper uses the parameteroptimized decision tree algorithm to classify the detection signal. After comparing the waveforms of the component signal and the loose particle signal in the time domain and the frequency domain, select the most representative feature as the split attribute of the decision tree. The grid search method is used to find the optimal splitting criterion and splitting depth of the decision tree, then use the parameter optimization decision tree to establish the classification model. The experimental results show that using the parameteroptimized decision tree algorithm to classify the loose particles detection signals can effectively improve the accuracy of the classification results, Gmeans value and Fmeasure value.

Abstract:To meet the demand of visualization in port machinery flaw detection, the electromagnetic tomography (EMT) technology is projected to detect and reconstruct metal structure flaw. Landweber iterative algorithm is a kind of traditional image reconstruction algorithm in research of EMT, but it does not work well when it is used directly in image reconstruction of metal flaw. Firstly, the region of interest is defined according to the characteristics of the sensor. On the supposition that the flaw has been detected and moved into the region of interest, the Landweber image reconstruction is carried out which gives better result, but the shortcoming is that the flaw detection and image reconstruction cannot be fulfilled at the same time, which means that the whole process is timeconsuming. Then, based on the characteristics of the sparse distribution of the metal structure flaw, the subspace pursuit (SP) algorithm which is a kind of sparse reconstruction method is used for flaw detection and image reconstruction. The simulation results show that the SP algorithm performs is better than the classical Landweber algorithm in the aspects of image reconstruction quality, algorithm robustness and so on, the flaw detection and the reconstruction are accomplished simultaneously.

Abstract:The systemonchip (SOC) contained a large amount of embedded memories, which were tested by a method of sharing builtin selftest circuits. The insertion process of the builtin selftest circuit was limited by the area overhead, test power and test time of the SOC. Aiming at this problem, the multimemory builtin selftest was modeled as a multiobjective optimization problem, and a multiobjective clustering genetic anneal algorithm was proposed. Based on the genetic algorithm, the algorithm obtained the memory compatible group through memory clustering, adopted the heuristic method to obtain the high quality initial solution, proposed an objective function with different weights under multiple constraints, and used the simulated annealing algorithm to evade better individuals to avoid local optimal solution risk. The results show that the proposed algorithm performs better than the genetic algorithm, and obtain memory solutions for testing, which reduces the power consumption by 113% or the test time by 487%, saving onchip test resources and test time.