Abstract:Abstract:In order to promote the profound study of the generation mechanism and detection methods of stimulation frequency otoacoustic emissions (SFOAEs) and push forward the clinical application of SFOAEs, the research development of stimulusfrequency otoacoustic emissions (SFOAEs) is reviewed in this paper, which includes SFOAEs generation mechanism, SFOAEs detection methods, the relationship of SFOAEs detection and assessment with hearing loss, and the evaluation potential of SFOAEs to the frequency selectivity of the auditory system. The generation source, transmission pathway and the component characteristics of SFOAE signals are discussed in the introduction section of SFOAEs generation mechanism. In the SFOAEs detection methods the commonly involved twotone suppression, nonlinear compression, spectrum smoothing and swepttone methods are introduced emphasisly. Finally, the clinical application of SFOAEs is prospected at the end of this paper.

Abstract:Abstract:Due to the constraint of imaging field of view or Xray penetration ability, computed tomography is difficult to realize nondestructive testing and quality evaluation of largescale plateshell objects. However, by improving scanning geometry adjustment and reconstruction, computed laminography (CL) provides an effective way for the internal structure analysis of plateshell objects. With the increasing demand for nondestructive testing in medical, electronic, material and other fields, CL technique has been paid much attention. Based on the nondestructive testing requirements of plateshell objects, this paper reviews the research advances of CL at home and abroad. The typical CL system structures are introduced. The analysis of CL image reconstruction algorithm and product development & application are illustrated. Finally, the application prospects and development trends of CL are forecasted.

Abstract:Abstract:The static parameter identification results of the JilesAtherton (JA) hysteresis model directly affect the predictive effect of the model on the transformer core hysteresis characteristics. Aiming at the problems of poor optimization ability and long calculation time existing in current single intelligent algorithm, a hybrid algorithm combining artificial fish school algorithm and linearly decreasing particle swarm optimization algorithm with optimized inertia weight is proposed. A transformer core magnetic performance test system was set up to conduct the experiment research on the hysteresis and loss characteristics of the transformer core under sinusoidal excitation. The identification speed and accuracy of JA model parameters for the proposed hybrid algorithm and other single intelligent algorithms were compared and analyzed. The results show that the root mean square error of the hybrid algorithm identification result is only 0006 9, which is lower than those of other single intelligent algorithms. The results prove that the proposed hybrid algorithm has the advantages of being not easy to fall into local optimal solutions, faster convergence speed and higher parameter identification accuracy compared with other single intelligent algorithms. In addition, considering the influence of the dynamic loss component on the hysteresis characteristics of the transformer core under alternating magnetic field, the existing dynamic loss coefficient solution method was modified, and a JA dynamic hysteresis model was established. Comparing the forecast result of the dynamic hysteresis loop model with experiment data, the correctness and effectiveness of the proposed method are verified.

Abstract:Abstract:Under the repeated load of the wheel, the rail tread is prone to fatigue cracks. Rail tread cracks are typical rolling contact fatigue cracks. These cracks can extend from the railhead to its downside, and it may cause the breakdown of the railhead. It brings hidden huge damage to transport safety. Firstly, the finite element method is used to analyze the interaction law between ultrasonic surface waves and the cracks in the rail tread. The scattering features and ultrasonic reflection behaviors of the lowfrequency surface waves are analyzed when inclination angles and depths of the cracks on the tread change. Secondly, a surface wave electromagnetic acoustic transducer (EMAT) with a center frequency of 03 MHz is designed and established. Then, the Bscan imaging inspection is performed on the rail tread with oblique cracks to identify the cracks intuitively. To improve the worse signaltonoise ratio (SNR) of the EMAT with a larger liftoff and conducting highspeed detection, the crackreflected ultrasonic echo is denoised and reconstructed by synchrosqueezed wavelet transform (SWT). The Bscan images can be rapidly constructed by using SWT. Results show that the designed surface wave EMAT can effectively detect multiple cracks in the rail tread. SNR of the ultrasonic echos is improved at least 883 dB by using SWT denoising method. In addition, the clarity of the Bscan image and the detection efficiency are effectively enhanced.

Abstract:Abstract:This paper proposes an image reconstruction algorithm based on Bayesian theory for electromagnetic tomography (EMT). The traditional regularization algorithms for EMT reconstruction can only achieve a single estimation. Hence, the information provided by the model is limited. A large number of reasonable parameter estimations for the model can be obtained by statistical methods. According to the sparsity of defect distribution, the conductivity distribution is divided into a series of block structures. Under the framework of sparse Bayesian learning, statistical information, including the prior information of sparse representation for conductivity distribution and the noise information in the measurement data, is taken into account. In this way, the full statistical description of the conductivity distribution can be obtained. The conductivity distribution for the surface defects of metal part is reconstructed based on the sparse Bayesian algorithm. To further prove the feasibility of this algorithm, the reconstruction results of the new method is compared with those of the conjugate gradient method and the total variation regularization method. The defect imaging experiments are implemented based on the EMT system. Compared with traditional methods, both simulation and experimental results show that the relative errors of reconstructed images based on the Bayesian algorithm with statistical information can be reduced by 20%. The quality and accuracy of defects images are effectively improved.

Abstract:Abstract:In order to solve the problem that normal electronic nose (enose) can hardly detect the low concentration gas odor of parts per billion (ppb) order, an enose preconcentration scheme and three temperature compensation (TC) methods are presented in this paper. Firstly, a preconcentrator is designed to improve the detection lower limit of the enose. Then, aiming at the problem that the detection and recognition effect of the enose decreases when the gas temperature is too high after preconcentration, three TC methods are proposed, which are the multivariate regression method, the neural network regression method and the TC ensemble learning method based on the two methods. Finally, enose detection experiment on the interior decoration material ppb level gas odor in vehicle was conducted. Two materials of polyurethane (PU) leather and polyvinyl chloride (PVC) leather were used to prepare the gases to be measured in the experiment. The average recognition accuracy rates before preconcentration, after preconcentration without TC and after preconcentration using the above three TC methods are 6114%, 8064%, 9167%, 9121% and 9506%, respectively, which verifies the effectiveness of the proposed enose preconcentrator and the TC methods.

Abstract:Abstract:Aiming at the problems that the spectral lines obtained using Fourier transform infrared spectrometer are enormous, and directly using all the spectral lines to perform multiple linear regression easily leads to overfitting, poor stability and long analysis period. In this paper, a bootstrap soft shrinkage variable selection method based on the combination of frequency and regression coefficient is proposed. This method selects the variables based on the weight of the variables; in each iterative process, the new weight of the variable is calculated according to the regression coefficient and frequency of the variable, and the soft shrinkage of the variables is realized through weighted bootstrap sampling technology. The method was verified using the infrared spectrum datasets of corn. On the corn oil dataset, the root mean square error of prediction (RMSEP) and correlation coefficients (Rp) are 0020 2 and 0976 5, respectively, the number of variables is reduced from the original 700 to 13. On the corn protein dataset, the RMSEP and Rp are 0027 9 and 0996 8, respectively, the number of variables is reduced from the original 700 to 16. The result shows that the proposed variable selection algorithm can select fewer and more precise variables, and has practical application value.

Abstract:Abstract:Learning from demonstration provides a convenient and feasible way for the firco (coexistingcooperativecognitive) robot to accomplish complex and various domestic tasks. Based on this approach, the nonexpert can easily teach the robot new motion skills to accomplish various domestic household tasks without tedious programming process. The skill learning approach is the key of learning from demonstration approach. Firstly, this study introduces the learning from demonstration approach and analyzes the learning framework. Then, the current studies are categorized into two kinds based on policy learning and representation approach. The typical approaches of each category are introduced in detail. Finally, these two categories comparatively are analyzed and the prospect of the development direction of this direction is proposed.

Abstract:Abstract:Aiming at the walking assistant demand of smart walker on slope, a method of implementing active/passive torque is proposed. The method utilizes the characteristic of controllable resistance torque of magnetorheological (MR) brake, combines the MR brake, worm gear reducer and DC motor to compose an active/passive hybrid actuator. Controllable active and passive torques are obtained through the coordination operation of the MR brake and DC motor. An MR brake with the rotor section of Tshape was designed, then finite element method was used to analyze the magnetic circuit inside the brake, and the mechanical performance parameters of the brake were obtained. On this basis, the active/passive hybrid actuator was developed, its mechanical performance test system was established and the experiment study was completed. The experiment results show that the proposed method can provide both controllable active torque and controllable passive torque, and the developed active/passive hybrid actuator can meet the functional requirements of walking assistant of the smart walker on slope.

Abstract:Abstract:Walking assistance exoskeleton robots are able to assist the people with lower limb dyskinesia to stand and walk upright. However, they have shortcomings in locomotion distance and speed compared with the electric wheelchair. A deformable walking assistance exoskeleton robot is designed, which has two independent forms of walking assistance exoskeleton and electric wheelchair that can be converted to each other. Aiming at the two motion processes of deformation and walking, the static stabilityoriented motion planning for the system is studied. The deformation principle of the deformable walking assistance exoskeleton robot is introduced. The motion planning method based on the projection point of the center of gravity (CoG) of the system is proposed. Moreover, the static stability of the exoskeleton robot during deformation and walking processes is analyzed. The deformation and walking motion experiments of a healthy subject wearing the exoskeleton were conducted. The stability margins in the deformation and walking experiments are greater than 54 mm and 20 mm, respectively. Experiment results demonstrate that the designed deformable walking assistance exoskeleton robot can achieve stable and reliable deformation switching and continuous walking utilizing the proposed static stabilityoriented motion planning method.

Abstract:Abstract:Aiming at cooperative robot, a flexible array type tactile sensor is independently developed and packaged into a tactile handle that can sense the grip posture and force of human hand. A convolutional neural network (CNN)based method is proposed, which can distinguish the three modes of loose griping, tight griping and inadvertently touching the tactile handle. The recognition accuracy reaches 982%. A variable admittance control strategy is proposed to adjust the virtual damping of the manipulator in real time according to the state with which human hand grasps the handle. Based on this tactile handle, the local posture change of human hand can be sensed in real time, the operator operation intention can be accurately estimated, and the local perception information is transmitted to the robot to control its motion. Taking the UR collaborative robot as the experiment platform, using the tactile handle as the perceptual input, the humancomputer interaction experiment was conducted, and the motion accuracy of the manipulator was evaluated. Experiment results show that the tactile handle has good intentional perception capability.

Abstract:Abstract:In the research of companion robots, gait phase detection is the key to maintaining manmachine synchronous motion. However, improving detection accuracy requires collecting and analyzing more gait phase information, which results in a long detection delay and is unable to meet the realtime requirements. In this paper, a progressive gait phase detection algorithm targeting to synchronous motion of companion robots is proposed. The algorithm mainly constructs the physical layer and decision layer of the probabilistic generative model based on the inertial measurement unit and Bayesian information criterion, and performs preliminary rapid gait phase detection; when the detection fails to reach the decision threshold, a memory network is introduced in the decision layer to predict the gait phase parameters for next period of time, thereby provide more decision information for the probabilistic generative model, and progressively complete the accurate incremental detection of the gait phase based on multiple decision results. The experiment results show that the proposed gait phase detection algorithm achieves an accuracy of 978%; the decision time is 283 ms, which is about 30% reduction compared with the adaptive Bayesian algorithm.

Abstract:Abstract:Because surface electromyography (sEMG) has nonstationary, aperiodic and chaotic characteristics, the traditional feature extraction method is difficult to be compatible in realtime characteristic and accuracy. In this paper, an improved energy kernel feature extraction method based on sEMG is proposed to process the acquired EMG signals. Firstly, based on the EMG oscillator model, the newly proposed Threshold Matrix Count (TMC) feature extraction method is described in detail. Then, the myoelectric sensors were stuck on the surfaces of 10 different muscles of the leg to detect the EMG signal during different motion processes of the lower limb. After acquiring the required EMG signals, the EMG signal characteristics of the 10 muscles were extracted and ten different feature vectors xk can be obtained. After analysis, four muscles were selected as effective muscles. Finally, the effective muscle feature vectors xk were combined to obtain a feature matrix Xk, which is inputted into the BP neural network for training, and four motion patterns were identified. The experiment results show that the calculation efficiency of the proposed energy kernel feature extraction method is improved by 13 times and 9 times compared with those of the traditional two energy kernel feature extraction methods. At the same time, compared with the commonly used time and frequency domain feature extraction methods, after training the obtained model possesses better stability and the average recognition accuracy reaches 952%.

Abstract:Abstract:To solve the problems of single target generation and incomplete characterization of personalized features in the study of human personalized gait generation, this paper proposes a method of human personalized gait generation based on the conditional generative adversarial networks. Firstly, a total of 51 joint angles of the whole body are set as preprocessing targets. Secondly, according to the walking parameters such as individual parameters, walking speed, joint composition and synergy relationship, data are labelled and condition information is constructed. Then, the human gait formation process is simulated by the conditional generation confrontation networks. Finally, the personalized gait with different walking characteristics is generated by adjusting the condition information. Through experimental analysis, the correlation coefficient between the personalized gait generated by the method and the real personalized walking data is larger than 098, the average absolute deviation is less than 008 rad, the absolute deviation of the threshold is below 5%, and the gait stability criterion results are within the stability interval. Experimental results show that the method can effectively generate personalized gait corresponding to different walking characteristics. Compared with similar researches, the walking features are more comprehensive and have better integrity.

Abstract:Abstract:In order to accurately extract the optimal time period and frequency band features of individual motor imagery EEG signals and effectively improve its classification accuracy, combining convolutional neural network and integrated classification method, a new multifeature convolutional neural network (MFCNN) algorithm is proposed to classify and identify motor imagery EEG signals. Firstly, the EEG signal is preprocessed, then the original signal, energy feature, power spectrum feature and fusion feature are inputted into the convolutional neural network to obtain their respective training models. Finally, the final classification result is obtained with the weighted voting based integrated classification method. The experiment analysis of the proposed method was carried out using the 2008 BCI competition Datasets 2b dataset and the actually measured data. The results show that the proposed MFCNN method can effectively improve the recognition rate of motor imagery. The average classification accuracy and average Kappa value of all the subjects in the experiment are 786% and 057, respectively. The proposed method provides a new idea and solution for the application of motor imagery braincomputer interface.

Abstract:Abstract:In minimally invasive surgical robot system, the implementation of the force feedback function can increase the flexibility of the surgeon during surgery and reduce the risk of damage to the tissues and organs of the patient. In order to achieve the force detection during surgical process, a multidimensional force sensor based on fiber Bragg grating (FBG) is designed for minimally invasive surgical robot and its decoupling methods are studied. The multidimensional force sensor is composed of three fiber gratings spaced 120° apart that are attached to the end of the surgical tool rod along the axial direction. Firstly, based on the stress analysis of the sensor, the least squares method is used for decoupling. However, the sensor has a nonlinear relationship between inputs and outputs due to the factors such as assembly and etc., so the feedforward neural network is used to carry out the nonlinear decoupling of the multidimensional force sensor. Then, the effects of the trocar translation on the decoupled force are analyzed theoretically and experimentally. The experiment results show that the feedforward neural network has good decoupling effects for the multidimensional force sensor, and the average errors in three mutually perpendicular directions are 005, 007 and 018 N， respectively. The maximum average error of the force detection after the trocar translation is 0036 N, which is negligible. The designed sensor, the decoupling method and the analysis of trocar translation effects are proved to have strong practicality.

Abstract:Abstract:Due to the good direction sensing characteristic of piezoelectric fiber to Lamb wave, the piezoelectric fiber sensor, which is similar to the strain rosette, is adopted to achieve the detection of the Lamb wave propagation direction. Aiming at commonly used 4 structure configurations of 45°rectangular, 135°rectangular, 60°delta and 120°delta, the theoretical, simulation and experiment studies are conducted to find out the influences of various piezoelectric fiber sensor configurations on the identification accuracy of Lamb wave propagation direction in this paper. According to the piezoelectric equation and attenuation characteristic of Lamb wave propagation, the sensing response equation of piezoelectric fiber to A0 mode Lamb wave under narrowband excitation is theoretically derived. The error function of the response amplitudes of three piezoelectric fibers is defined and used to estimate the Lamb wave propagation direction. The ANSYS software was adopted to conduct the piezoelectric coupling simulation analysis and obtain the response signals of different piezoelectric fiber sensors, and corresponding experiment tests were carried out. The matching pursuit algorithm was applied to perform the signal decomposition and extract the response amplitudes. The Lamb wave propagation directions were identified according to the error function and the response signals. The simulation and experiment results are compared, the causes of simulation and experiment errors are discussed. The analysis results show that the identification errors of the Lamb wave direction detection for four different configurations of the piezoelectric fiber depend on the length of the piezoelectric fiber, configuration angle, excitation frequency, the adhesion condition of the piezoelectric fiber, and etc. In terms of overall error, the 135°rectangular configuration is the best, the error is less than 4%; the 60°delta configuration is slightly better than the 45°rectangular configuration, its error is less than 8%; due to the large size, the 120°delta configuration presents the largest error. This study could provide a theoretical basis for piezoelectric fiber sensor structure design.

Abstract:Abstract:Based on the idea of redundant fault tolerance and measurement reconstruction, this paper proposes a redundant orthogonal parallel sixdimensional force sensing mechanism with reconfigurable measurement model, and carries out related theoretical analysis and experiment research. The flexibility matrix theory is used to establish the stiffness model of the mechanism, and the correctness of the modeling method is verified through simulation. The prototype of the designed force sensing mechanism was developed, the calibration test bench was built, and the reconfiguration experiments on the measurement model were conducted, in which different measurement branch combination was selected according to different dimensional forces. The experiment results show that after the reconfiguration of the measurement model of fivedimensional force and threedimensional force, the measurement accuracy increases by 129% and 319%, respectively. The multidimensional force sensing mechanism with reconfigurable measurement model realizes the multidimensional force measurement facing to different measurement tasks, which provides theoretical and experiment bases for the study of redundant sixdimensional force sensing mechanism.

Abstract:Abstract:The contact vibration measurement method has application limitation in the flow induced vibration test of the tube bundle. To solve this problem, a new measurement device, named as the differential pressure vibration measurement device, is designed. Based on the principle of measuring the pressure differential fluctuation frequency on both sides of the heat exchange tube, the differential pressure type vibration measuring device can measure the excitation frequency of the external fluid force of the heat exchange tube. Differential pressure vibration measurement device solves the problem of interference of the traditional contact sensor to the external flow field of the heat exchange tube and the problem of limited installation space. To verify its effectiveness, the single tube flow induced vibration test and the heat exchange tube bundle flow induced vibration test are carried out respectively. Experimental results show that the designed device can accurately capture the frequency of all kinds of vibration forms of the heat exchange tube in the complex flow environment, including pump vibration, natural vibration, periodic vortex vibration and twophase flow vibration characteristics. Experimental results of single tube flow induced vibration test show that the periodic vortexinduced vibration frequency measured by the designed device is within 5% of the theoretical value. Therefore, the development and application of the differential pressure vibration measuring device is of great significance for the measurement of the flowinduced vibration test and the longterm monitoring of the vibration of the heat exchange tube.

Abstract:Abstract:The propagation of the impact signal in the reinforced bulkhead structure of spacecraft is very complex, which hinders the rapid location of space debris impact. To solve these problems, a soft threshold filtering hyperbolic algorithm based on acoustic emission is proposed. The algorithm solves the difficult problem of judging the arrival time of the attenuation of the wave in the reinforced structure by the soft threshold scheme. The S0 mode with the fastest wave speed is preserved by filtering, which reduces the influence of the modal transition of the acoustic wave on the location. The influence law of ribs on Lamb waves is analyzed by finite element simulation. The energy passing ratio of S0 mode Lamb waves in different frequency bands is obtained to determine the filtering frequency band. The feasibility of the algorithm is verified by the location experiments of 20 impact points. For the experimental plate in this study, the best location result is obtained when the frequency band is 100~200 kHz, and the average absolute error is 559 mm.

Abstract:Abstract:The received signal strength indication (RSSI) based indoor fingerprinting positioning algorithm has problems of referencepoints errormatching and location discovery. To solve these problems, a fuzzy clustering and regional cat swarm based positioning method is proposed. Firstly, the fuzzy clustering is used to accomplish clustering and estimate RSSI feature of the cluster center instead of the traditional hard clustering algorithm. In this way, the fuzzy clustering based twolevel matching can increase the difference between reference points, and reduce the complexity of feature matching. Then, the cat swarm optimization is utilized due to the fast convergence near the optimal solution, which is suitable for the location discovery based on the regions obtained by the twolevel matching method. Simultaneously, a feed mechanism is designed to improve the local search capability and the convergence speed of the cat swarm optimization. Compared with traditional algorithms, experimental results show that the proposed algorithm can improve the positioning accuracy by 12.5%.

Abstract:Abstract:Using deep learning technique to automatically and accurately identify the incipient fault of rolling bearing, especially the fault position, classification and severity degree, is a research hotspot in current fault diagnosis field. The traditional fault diagnosis method excessively relies on the manual feature extraction by the engineers with prior knowledge, which is difficult to effectively extract incipient fault features. In this paper, a novel improved CNNsSVM method is proposed and used for the rapid intelligent fault diagnosis of motor rolling bearing. This method adopts the combination of 1×1 transitional convolution layer and global average pooling layer to replace the fully connected network layer structure of traditional CNN, which effectively reduces the number of training parameters of CNN. In test stage, the method uses SVM to replace the Softmax classifier, which further improves the diagnosis accuracy. The proposed method was applied to the fault experiment data of the motor support rolling bearing, and the method was compared and verified with traditional intelligent diagnosis methods. The results show that the accuracy of fault identification of the improved CNNsSVM algorithm reaches up to 9986%, and the proposed method has good migration generalization ability under different load conditions and possesses the feasibility for practical engineering application. The fault diagnosis accuracy and test time of the method is obviously better than other intelligent algorithms.

Abstract:Abstract:To solve the problem of small underwater objects classification based on multiview sonar images, a deep neural network classification method with multiview is proposed. Firstly, the shadow area of underwater objects in sonar images is extracted. The main axis slope of shadow area is calculated, which is used to match sonar images to the corresponding simulated dataset. The convolutional neural network trained by this simulated dataset is applied to extract deep neural network features from multiview sonar images. The achieved feature vectors from sonar images of different views are combined as a feature vector of underwater object and predicted from support vector machine. The classifier is utilized to classify multiview sonar images collected from lake and sea trials. The average classification accuracy can reach 9333%. The performance is improved compared with the single-view classification method using convolutional neural network and support vector machine.

Abstract:Abstract:TbDyFe alloy can be used in high frequency and even ultrasonic frequency devices. The hysteresis and loss of TbDyFe alloy (TerfenolD) are serious under high frequency magnetic field (f>1 kHz). It is very important to study the dynamic magnetic properties and loss properties of the TbDyFe alloy laminates with different thicknesses for designing the devices in high frequency range. The AMH1MS dynamic magnetic property testing system was used to measure the dynamic hysteresis loops of TbDyFe alloy laminates with different thicknesses under different driving magnetic field frequencies and magnetic density amplitudes. Based on the dynamic magnetic theory and loss calculation model, the dynamic magnetic property parameters of TbDyFe alloy laminates with different thicknesses were compared and analyzed, and the loss mechanism in high frequency range was deeply studied. The experiment results show that when the magnetic density amplitude Bm is 005 T and the frequency is 5 kHz, compared with that of a single 2 mm thickness sample, the dynamic hysteresis loop of the 1 mm thickness laminate sample becomes narrower laterally, the required magnetic field strength is reduced, the coercive force and magnetic energy loss values are reduced by 264% and 281%, respectively, and the amplitude permeability is increased by 117%. Under certain magnetic density magnitude, the value of the real part and imaginary part of the complex permeability, as well as the amplitude permeability of TbDyFe alloy present a tendency of data decreasing and speed slow down as the frequency increases. At the condition of high frequency and high magnetic density (f>5 kHz, Bm>006 T), the thickness decreasing makes the loss decreasing of TbDyFe alloy more obviously.

Abstract:Abstract:From the perspective of flow degradation trend, a life prediction method based on Adaptive Networkbased Fuzzy Inference System (ANFIS) is proposed. Firstly, the modified ensemble empirical mode decomposition (MEEMD) method is used to perform multiscale reconstruction and noise reduction on the vibration data of accelerated degradation test. The kurtosis, meansquare frequency, wavelet packet energy of the reconstructed signal are extracted, which together with the signals of torque, rotation speed and pressure are used as the characteristics of performance degradation of outertooth gear pump. Then, kernel principal component analysis (KPCA) method is used to perform the multiple feature fusion. Furthermore, the establishment and analysis of the degradation evaluation indices of the outertooth gear pump are realized. The degradation evaluation indices and flow signals are used to train the ANFIS model, and the remaining life prediction model of the gear pump is obtained. In order to further verify the effectiveness of the algorithm, the gear pump remaining life prediction model is compared with liner regression model and cubic exponential prediction model. Finally, based on the Monte Carlo sample expansion method, the reliability evaluation of the outertooth gear pump is achieved. The results show that the prediction error between the result of the proposed method and the actual threshold is about 8%, the proposed method can accurately evaluate the life of the outertooth gear pumps.

Abstract:Abstract:Under complex sea conditions, the ship detection using remote sensing image is easily affected by sea clutter, thin clouds and islands, which results in low reliability of detection. In this study, an endtoend deep semantic segmentation method is proposed, which combines the deep convolution neural network with the fully connected conditional random field. Based on ResNet architecture, the remote sensing image is roughly segmented by deep convolution neural network. Using the method of Gaussian pairwise and mean field approximation, the conditional random field is established as the output of the recurrent neural network. In this way, the endtoend connection is achieved. On the dataset provided by Google Earth and NWPURESISC45, the comparison between the proposed method and other models is implemented. Experimental results show that the proposed method can improve the accuracy of target detection and the ability of capturing fine details of images. The mean intersection over union is 832%, which has obvious advantage than other models. And it can also run fast, which meets the requirements of ship detection in remote sensing images.

Abstract:Abstract:Noncontact, high precision and rapid flatness measurement of miniature objects is required in the modern industry. In this study, the 3D laser flatness measurement system based on machine vision is developed by the 3D laser measurement method. Firstly, the scanning measurement principle and the flatness measurement principle of the 3D laser profilometer are studied. The laser line image is preprocessed to enhance the accuracy of the later measurement. Secondly, the geometric feature is positioned and coordinate transformed. The data are processed again to acquire the threedimensional measurement of the object. The system provides a measuring device and method for threedimensional noncontact, highprecision, and multidimensional measurement of microobject geometry. Finally, the physical measurement experiments verify that the realized system has the advantages of accuracy, rapidity and effectiveness. The measurement accuracy can reach 01 μm.

Abstract:Abstract:Aiming at the poor precision problem of video foregroundbackground separation based on traditional robust principal component analysis, this paper proposes a new model called generalized nonconvex robust principal component analysis (GNRPCA) model. This model adopts the generalized nuclear norm and generalized norm to replace the rank function and l0 norm in the robust principal component analysis model, respectively, which can solve the problems of the excessive penalty for the surrogate functions of the rank function and sparsity function existing in traditional robust principal component analysis model, and leading to poor approaching degree. Then, the alternating direction method of multiplier (ADMM) is adopted to solve the proposed GNRPCA model. Finally, the proposed algorithm was used for video foregroundbackground separation. Simulation experiments were conducted, the experiment results were analyzed. The experiment results prove that the average Fmeasure value of the proposed algorithm is 0589 2, which is 13% higher than the truncated nuclear norm algorithm. And the proposed algorithm is more superior and effective than other video foregroundbackground separation algorithms based on robust principal component analysis.