Abstract:The diameter change of the spherical grinding wheel after on-site dressing may lead to position deviation between ideal and actual grinding trajectory, which bring the machining error. Therefore, it is very necessary to obtain the change information before ultra-precision grinding. In view of this, this article proposes an on-site measurement scheme of the spherical grinding wheel, including the establishment of the on-site measurement system based on the laser tool presetter and on-site measurement steps design. The systematic error of this scheme is analyzed, the installation error model of laser tool presetter is abstracted, the error formula is deduced combined with homogeneous coordinate transformation, and the installation error is solved by the least square method. Therefore, the compensation of measurement data and high-precision on-site measurement of the spherical grinding wheel are realized. Finally, the standard ball with 12. 5 mm radius is used for calibration, and the error angle value is calculated for compensation. Then, the 12. 5 mm and 3 mm standard balls are measured for verification. Experimental results show that the measurement accuracy can be improved from 4 μm to 1. 2 μm.

Abstract:To ensure the accuracy of harmonic compensation value for the turntable positioning error, a calculating method for the harmonic error function is proposed. The harmonic compensation method for turntable positioning is described and the feasibility of the CORDIC algorithm for harmonic error function calculation is introduced in detail. Based on the principle of the CORDIC, the quantitative models of the iteration number n and the data width b are established, respectively. The overall quantization error of calculating harmonic function is deduced. The calculating method is implemented by using the laboratory-made field-programmable gate array circuit. The value of n and b are determined to satisfy the calculation accuracy. Finally, a series of evaluation tests are performed to demonstrate the effectiveness of the method by comparing theoretical and computational values. The data of experiment performed on the test turntable platform confirm that the proposed method can calculate the value of harmonic function effectively. The positioning accuracy of turntable is improved from 29. 0" to 5. 3" with this method.

Abstract:In the measurement of electronic speckle pattern interferometry, interference fringes contain much noise that affects the subsequent phase extraction. To solve the problem that the bilateral filtering cannot effectively remove high noise, a joint bilateral filtering combined with the windowed Fourier filtering is proposed to deal with electronic speckle pattern interferometry fringes with high noise. In the method, an image after initial processing by the windowed Fourier filtering is introduced as the guiding map of the joint bilateral filtering. Then, an improved similarity factor of pixel value is introduced to provide more reliable guiding information and better filtering effect. The proposed method is applied to four different simulated electronic speckle pattern interferometry fringes with high, medium, low and variable densities and the real electronic speckle pattern interferometry fringes with high noise. In addition, the proposed method is compared with the bilateral filtering and the windowed Fourier filtering. Experimental results show that the fringe obtained by joint bilateral filtering is smoothest, and the structure is the most complete. The peak signal to noise ratio of the simulated fringe filtered by the new method is improved by 1. 0 ~ 4. 2 dB, the structural similarity index and edge preservation index are the highest, and the mean square error of the phase is the smallest. In addition, the proposed method is more convenient to adjust, and the calculation time for a 330×330 simulated fringe is only about 4 s.

Abstract:The tri-eccentric butterfly valve relies on full surface contact between the butterfly plate and the valve seat sealing surface to achieve zero leakage sealing effect. The sealing surface of the processing and manufacturing accuracy has a vital influence on the sealing performance. The existing sealing surface measurement mainly relies on the off-line measurement, and there are some problems such as inconsistent measurement datum and measurement errors caused by secondary clamping. In this study, the precision in-position measurement technique of sealing surface and the original point cloud data processing method is proposed. For the measured sealing surface, the trough-clustering algorithm and the normal vector-surface fitting algorithm considering constraints are proposed. The key parameters and machining error of the sealing surface are obtained. Compared with the least square method, the relative accuracy of this algorithm is improved by more than 60% . The relative error between the cone angle of the sealing surface of the three-eccentric butterfly valve measured in situ and the result measured by the CMM is only 0. 43% , which meets the requirement of ±0. 5% . The measurement accuracy of the in-position measurement technology of the sealing surface has been verified effectively, which provides a reliable technical means for the precision measurement of high-end valves in the future.

Abstract:In the 3D measurement of gear, the position and attitude calibration of sensors directly affects the accuracy of measurement results. In this article, a position and attitude calibration method of the linear laser sensor is proposed, which is based on feature standard parts. And a standard part with certain geometric features is designed. This method decouples the position and attitude relationship between the sensor and the instrument. Through the geometric relationship between the sensor and the standard part during the movement, the three attitude angles of the sensor are calculated and adjusted to the calibrated zero position. Then, the offset distances of the three positions of the sensor are calculated through calculating equal region mean by multiple offsets. The verification experiment is implemented on the gear measuring center. After multiple calibration, the dynamic measurement repeatability of the tooth profile deviation of the measured gear is 3. 94 μm. The evaluation standard deviation is less than 0. 3 μm. The evaluation results are consistent with the traditional contact measurement results, which show that the position and attitude calibration method of linear laser sensor based on feature standard parts has good repeatability and accuracy for the 3D gear measurement.

Abstract:In the process of using the ball bar to identify the geometric error of the translational axis, the parameter vector matrix at any position in the formulated identification model is an ill-posed matrix, which may result in inaccurate solutions or no solutions in solving the identification model. To address these issues, a method of ridge estimation based on the virtual observation method to solve the identification model solution is proposed. The translational axis of the machine tool is taken as the research object. Based on the rod length data measured by the ball bar, they are substituted into the established mapping relationship between the error element and the change of the ball bar rod length. The geometry is solved by the virtual observation method the polynomial coefficient of the error term. This method improves the ill-condition of the identification matrix from the cause of the matrix. Then, the identification of the error elements related to each axis is identified. Finally, simulation and experimental results show the correctness of the identification method. And the ill-posed of the identification matrix is improved. The results provide the theoretical basis for accurately identifying the geometric errors of the machine tool.

Abstract:A double-sided magnetic sensing composite structure is proposed using FeGa magnetostrictive films with variable cross-section on the top and the bottom of the AT-cut quartz wafer. The variable cross-section shape is utilized to reduce the demagnetization factor of the convergent area of the magnetic films, further increasing the strain in the electrode area. Meanwhile, the change in the strain direction along the wafer thickness direction can be avoided by adopting the double-sided composite structure. Therefore, the sensitivity of the resonant magnetic field sensor can be improved effectively by combining above two methods. The derived equations show that the sensitivity is positively correlated to the strain transfer coefficient determined by the composite structure, the piezomagnetic coefficient of the magnetostrictive material, and the thickness of the magnetic film, while inversely proportional to the square of the quartz wafer thickness. The sensor samples are fabricated by sputter deposition, with 1 μm thick variable cross-section FeGa films on the top and the bottom surface of the 200 μm thick AT-cut quartz wafer. Experimental results show that the Q value and sensitivity of the sensor are 5 489 and -0. 82 Hz / Oe, respectively. Therefore, when the thickness of the quartz wafer is reduced to 7. 5 μm, the sensitivity of the sensor can reach upwards to - 583 Hz / Oe. In addition, the sensitivity of the sensor can be further increased by using a higher piezomagnetic coefficient magnetostrictive material.

Abstract:Due to the special operating environment of Marine cables, higher and higher spatial resolution and temperature resolution are required for existing distributed optical fiber temperature measurement systems. However, in the distributed optical fiber temperature measuring system based on Raman scattering demodulation, the spatial resolution, temperature resolution, temporal resolution and temperature measuring distance are mutually restricted. To design a temperature measuring system for meeting the on-line monitoring requirements of the ship power system, the influencing factors of the spatial resolution of the system are analyzed. On this basis, the equivalent model of distributed optical fiber temperature measurement system based on the optical time domain reflection technology is formulated, and the priority of each influencing factors is obtained. A more accurate influence model of system temperature resolution is deduced, which is based on the Raman scattering light intensity formula. Finally, the interaction formula among spatial resolution, temperature resolution, temporal resolution and temperature measurement distance is achieved, which provides a reference for the design of performance index of the distributed fiber temperature measurement system. By further increasing the accumulation times and decreasing the pulse width of pump light, the spatial resolution can be improved while ensuring the spatial resolution and temporal resolution in a short measurement distance. An experimental platform is established to evaluate the proposed method. By adjusting the accumulation times and pulse width in 500 m temperature distance, the temperature resolution is 0. 65℃ , the spatial resolution is 1 m, and the temporal resolution is 1 s.

Abstract:A fiber Bragg grating accelerometer based on the rotating beam is designed to attain high accelerated sensitivity in lowfrequency vibration. By analyzing its vibration model and MATLAB numerical calculation, the optimized structural parameters of the sensor are achieved. Its theoretical sensitivity and natural frequency are 1 725 pm/ g and 68. 4 Hz, respectively. Meanwhile, its sensitivity response characteristics are simulated by COMSOL, and the simulation results are closely consistent with the theoretical analysis. Experimental results of frequency response and amplitude characteristics indicate that when the acceleration changes from 0~ 2 g and working frequency in a range from 0. 5~ 20 Hz, their central wavelength of FBGs is linearly related to the vibration acceleration. In addition, a high sensitivity up to 1 495. 2 pm/ g and a good repeatability is achieved. The sensor has a simple and compact structure, in which the elastic energy consumption is decreased by bearing during the vibration of the cantilever beam. Therefore, its sensitivity is significantly improved. And the low-frequency vibration signals are detectable.

Abstract:This article detects the shape of objects based on the flexible fibre optic grating ( FBG) sensors encapsulated in PDMS material. The influence of the number and layout of sensors on the reconstruction accuracy of objects of different shapes is discussed. The accuracy of the algorithm is evaluated by using COMSOL simulation software. Four FBG flexible sensors are designed and calibrated for curvature. The sensors are distributed on the aluminium plate. The minimum relative errors for the equally distributed 3 and 4 FBG flexible sensor reconfiguration are -6. 12% and 0. 48% , respectively. Experimental results show that the reconstruction is better when 4 sensors are utilized. Therefore, by selecting 4 sensors in 7 different sets of dispositions, the layout has small relative error under the condition of 4 different weights. The specific is that the interval values of one end of the aluminium alloy plate are 260, 100 and 90 mm, respectively. For the other end, the interval values are 150, 150 and 150 mm, respectively. The effect of reconstruction is well and all have the smallest relative errors in the seventh group of 6. 22% , -1. 01% , -2. 52% , and -3. 22% , respectively. The designed FBG flexible sensor provides the basis for applications in the field of soft robot shape reconstruction.

Abstract:Clock synchronization provides a common time reference for different nodes in the wireless sensor networks to deal with distributed tasks. The wireless sensor networks require accurate clock synchronization to keep the consistence and coordination of data among nodes. However, the unpredictability of message delays in the synchronization process may affect the synchronization accuracy significantly. To eliminate the impact of delay fluctuation, the clock synchronization model is formulated, which considers the influence of delays with Gaussian distribution in the synchronization process. And a Kalman based delay estimation algorithm on distributed clock synchronization is proposed, which can effectively deal with the influence of delays. This algorithm utilizes one-way message exchange mechanism, and all parameters could be updated independently in one node. In addition, the calculated global time is considered as a reference to update the clock offset. In this way, the synchronization accuracy is improved. The MATLAB simulation and the nRF52832 experiment testbed with 5 nodes indicate that this algorithm restricts the synchronization error into 10 μs under microsecond delays. Compared with other algorithms, this method could achieve better synchronization accuracy and performance.

Abstract:The free-head 3D gaze tracking is of great significance. The traditional eye tracking methods have problems of low accuracy, complex equipment and limited wearing. To address these issues, a monocular head free 3D eye tracking technology based on machine learning is proposed. Two lightweight, high-precision and real-time eye tracking models are formulated, which can estimate the gaze point and gaze direction, respectively. For the gaze point estimation model, Dlib is used to locate the facial feature points to get the eye image. Then, PNP is used to get the head pose. Two kinds of information and part of the feature point coordinates are taken as the input into the multi-channel convolutional neural network. Finally, the gaze point is estimated. The gaze direction estimation network is a simplified version of gaze point estimation network. The proposed eye tracking technology is combined with the electric sickbed to establish a set of electric sickbed system based on eyeball drive, which allows patients to use their eyes to control the operation of the electric sickbed. Experimental results show that the error of the proposed gaze point estimate model on MPIIGaze dataset is 4. 1 cm. The error of gaze direction estimation network in ColumbiaGazedata set is 7. 2°, and the accuracy of the two models is 6. 8% and 2. 7% higher than those of iTracker and UlinFT, respectively. The system of eye driven electric sickbed can improve the living standard of patients and meet the needs of patients.

Abstract:Squat is known as the king of strength trainings. However, the incorrect positions may produce irreversible damage to the human body. This study proposes a method utilizing plantar pressure to detect common incorrect squats position. Insoles with eight pressure sensors are utilized to collected 5 sets of database, which are correct squat and 4 common incorrect squats. An algorithm is proposed to segment those continuous pressure data. Then, the pressure nephogram is also analyzed. Three sets of deep neural network are designed as classifiers, which are Att-LSTM, LSTM, and CNN, respectively. Experimental results show that accuracies of these models are 90. 2% , 83. 0% and 79. 8% , respectively. The results suggest that the utilization of sensor insoles with LSTM and attention mechanism as the classification algorithm is a valid method to detect squats position.

Abstract:The artificial anal sphincter (AAS) is permanently exposed to biological tissue fluid after implantation. The design of stable and reliable pressure sensing module plays an important role in the reconstruction of defecation perception and ensuring the safety of blood supply. In this article, based on the characteristics of the acquisition of physiological parameters of rectal pressure, firstly, the material size and other parameters of the sensor are selected through finite element analysis and design, and the slot structure of the sensor installation and the cable trough of the pressure module circuit of the prosthetic arm are designed. The packaging technology and installation method of the new sensor are studied through the experiments on the anti-moisture performance of 9 series of protective adhesives, and a new packaging form is designed for the circuit module of the system, which improves the waterproof performance of the whole system. Experimental results show that the strain gauge resistance of the new sensor changes within 1‰ in the underwater environment for 20 days, the sensor has a high linearity, and the output operating voltage range is appropriate. After the underwater test of the AAS system, the average fluctuation rate of initial sensor voltage is only 4. 1% , which preliminarily verifies the good applicability and application feasibility of the new pressure sensor designed in the vivo environment.

Abstract:The TAS2Rs in extra-oral tissues may become a potential therapeutic target in the treatment of related diseases. In this study, cells expressing of TAS2Rs are used as sensitive elements. According to their physiological characteristics and coupling with different sensors, the heterotopic expression of taste receptors and their application in personalized drug screening are explored. The personalized drug screening platform for different disease models is established. Firstly, based on the electric cell-substrate impedance sensing (ECIS) and the endogenous T2R38 receptor expressing Caco2 cells, a drug screening platform for isothiocyanates is developed. The EC50 of phenylthiourea is calculated as 157. 6 μM. Secondly, combined with the microelectrode array (MEA) detection system, the inhibitory effects of diphenidol ( 5 ~ 160 μM) and salicin ( 0. 001 ~ 100 μM) on the contraction of cardiomyocytes are investigated. Furthermore, based on three-dimensional airway smooth muscle cells (ASMCs) array and the gel imaging system, the diastolic effect of tangeretin (20 μM) on airway smooth muscle cells is explored.

Abstract:To meet the requirement of humanoid motion planning of the robotic arm in human-robot interaction environment, a humanoid motion planning method of the robotic arm based on reinforcement learning is proposed in this article. Firstly, based on the structural characteristics of the human arm, the shoulder angle, the elbow angle and the wrist joint motion angle are designed to reflect motion characteristics of the robotic arm. The motion data of human arm captured by the VICON system are analyzed by using normality and correlation analysis methods to achieve the motion characteristics of the human arm. Then, according to different motion characteristics rules, the corresponding reward functions are designed, and the humanoid motion model is trained by the reinforcement learning method. Finally, the humanoid motion platform of the robot arm is established, and the success rate of the humanoid motion is 91. 25% . It evaluates the feasibility and effectiveness of the proposed method, which could be used to improve the humanization of robot motion.

Abstract:This article investigates the influence of different nanofluid distributions on the treatment temperature by considering a homogeneous distribution of magnetic nanoparticles (MNPs). Two distributions are observed in experimental images and modeled by the proposed method in this article. Meanwhile, the effect of nanofluid diffusion on the concentration distribution is further studied. In addition, the difference of temperature-based on the blood perfusion rate and the traditional quantitative perfusion rate is also considered. The treatment effect in this study is evaluated by the cumulative equivalent minutes at 43℃ ( CEM43) from the perspective of the thermal iso-effect dose, which is directly related to the treatment temperature distribution. Simulation results demonstrate that the nanofluid diffusion has positive effect on the uniformity degree of nanofluid distribution inside tumor region, while it has less effect on the ultimate treatment temperature if individual critical power dissipations of MNPs.

Abstract:The limited applications of the traditional machine learning algorithms and the transfer learning algorithm are considered in this study. The improved manifold embedded distribution alignment (MEDA) algorithm is utilized to improve the detection accuracy in the cross-subject emotion recognition. The MEDA algorithm in the manifold space could reduce the data drift between domains by popular feature transformation, which can adaptively and quantitatively estimate the weights of edge distribution and conditional distribution. This article proposes an improved manifold space distribution alignment algorithm to address the problems of large feature dimension and possible bad features. An improved minimum redundancy maximum correlation algorithm is introduced for feature selection. The computational complexity is reduced, the associated features are selected, and the decision-level fusion on multiple groups of recognition results in multi-source domain is performed to further improve the transfer learning effect. The analysis results of SEED data set and the measured data set show that the distribution alignment algorithm in the manifold space is better than those of the support vector machine, transfer component analysis and joint distribution adaptation. The overall recognition accuracy is improved by 8. 97% , 4. 00% , and 2. 89% , respectively. The improved distribution alignment algorithm in manifold space has improved the recognition accuracy of each subject, and the overall recognition accuracy is improved by 3. 36% . Therefore, the effectiveness of the proposed method is verified.

Abstract:To address the problem of PCB defect detection caused by noise interference in industrial environment, a PCB defect detection method based on the multi-attention Faster RCNN is proposed. The attention mechanism is introduced into the feature extraction and feature fusion parts to obtain feature representations with anti-interference ability. First, the defective features are extracted by using a split-attention network that automatically focuses on the defective features to reduce the effect of noise interference. Secondly, a balanced feature pyramid is used to fuse different resolution features, and a non-local attention mechanism is utilized to weight the fused features to different regions within the global perceptual field to enhance their defect characterization and further suppress noise interference. Finally, based on the obtained feature representation, the regional proposal network is used to generate defect candidate box. The fully connected layer is utilized to determine defects′ position and category to obtain the detection results. Experiments are implemented on the printed circuit board defect data sets under different degrees of noise interference. The average detection accuracy reaches 92. 4% , which proves the effectiveness and feasibility of the proposed method.

Abstract:Machine vision is one of the important measure manners for environmental perception. It is a research hotspot in the fields of automatic driving, robot, industrial detection and so on. The fine analysis of point cloud data is one of the key technologies. To solve the problem of low segmentation accuracy of large-scale point cloud data of real scene, a bilateral feature aggregation network architecture for semantic segmentation of the point cloud is proposed. Firstly, a bilateral feature aggregation module is formulated to aggregate local features by processing the geometric information and semantic information of the point cloud. The aim is to make full use of the feature information of the point cloud. Secondly, the high-dimensional spatial correlation of nearest neighbor features is used to calculate the impact between points. The context information of local neighborhood is enhanced. A hybrid-pooling architecture is proposed to replace the max-pooling to reduce the information loss of max-pooling, and the horizontal skip connection pooling is used to enhance feature diversity. Finally, an attention module is introduced to extract global features, which can filter scale noise and enhance the spatial expressiveness of features. Experimental results show that the mean intersection over union of the proposed method is 68. 2% , and the mean accuracy is 80. 7% . These two values are 20. 6% and 14. 5% higher than those of the PointNet. The objective indicator is better than the existing representative methods.

Abstract:To meet the digital drilling and assembly need of large components in the cutting-edge manufacturing field, an automatic detection and positioning system for base hole are proposed by using the line laser scanner and the industrial robotic arm. Firstly, the hole bottom point cloud interpolation compensation is implemented for the three-dimensional point cloud of the base hole, and the rough edge feature point extraction of the base hole is achieved by using the bidirectional gradient constraint algorithm. Secondly, for extracting edge feature points, a refinement algorithm based on angle judgment and a compression algorithm based on plane fitting are used to achieve the refined extraction of edge feature points. Finally, the edge feature points are parameterized in the space circle, and the aperture and hole distance are obtained. The detection and positioning of the base hole are achieved. Experimental results show that the detection accuracy and positioning accuracy ( relative error) of the proposed method for the base hole are 1. 77% and 0. 24% , respectively. These values are better than the 2. 77% and 0. 46% of the traditional method. It can meet the digital drilling and assembly requirements of large parts in manufacturing.

Abstract:The current binocular 3D detection algorithm has the problem of slow online calculation speed due to a large number of anchor points to be selected. To address this issue, an anchor-guided 3D target detection algorithm is proposed, which is based on the stereo RCNN. This method is named as the FGAS RCNN. In the first stage, a probability map is generated for the left and right input images to generate sparse anchor points and corresponding sparse anchor boxes. The left and right anchors are used as the whole entirety to generate a 2D preselection box. The second stage is based on the key-point generation network of the pyramid feature network. The key-point heatmaps are generated by the information of these sparse anchor points. A 3D bounding box can be generated by combining the stereo regressor with these key-point heatmaps. The original image will lose pixel-level information after convolution. The instance segmentation mask generated by Mask Branch can be used to solve this problem. The 3D bounding box center depth precision can be improved by the instance segmentation mask and the instance-level disparity estimation. Experimental results show that the proposed method can reduce the amount of calculation while maintaining a high recall rate without any depth and position prior information input. Specifically, the mean average precision is 44. 07% on 3D target detection with a threshold of 0. 7. Compared with the stereo RCNN, the proposed method improves the average precision by 4. 5% . Meanwhile, the overall running time of our method is 0. 09 s shorter than Stereo RCNN.

Abstract:The trinocular stereo vision system could overcome the occlusion problem of the binocular stereo vision system. The measurement accuracy of the stereo vision system can be further improved. However, the large number of sensors may lead to the high computation load of the matching algorithm and affect the real-time performance of the system. Hence, the practical application of trinocular stereo vision systems is limited in various fields. In this article, a trinocular semi-global stereo matching algorithm and its hardware computational framework are proposed. Firstly, based on the in-depth analysis of the basic model of the trinocular stereo vision system, a hardware-friendly semi-global matching algorithm is proposed. Then, the overall framework of the on-chip system and the structure of each computational module are designed according to the parallelization and pipeline processing characteristics of field programmable gate array (FPGA) hardware computation. Finally, a complete hardware experiment system is established, which is based on Zynq-7000 SoC FPGA for algorithm implementation. The algorithm is evaluated by using dataset images and real scene images, respectively. Compared with the traditional binocular semi-global matching algorithm, experimental results show that the effective pixel filling rate of this algorithm is increased by 17. 31% , the error rate is reduced by 13. 06% , and the real-time stereo matching can be achieved at 60 fps in real scenes, which could meet the practical needs of various application scenarios.

Abstract:It is important for the surface quality of hot rolled steel strips to make final product. Therefore, it is necessary to strictly control the defects on the surface of hot rolled steel strips. The current you only look once (YOLO) v4 algorithm has low detection accuracy and poor performance on small-scale information. To address these issues, an improved YOLOv4 automatic detection method is proposed. First, to improve detection speed, enhance detection target feature extraction and reduce gradient vanishing, the feature extraction network CSPDarknet53 in YOLOv4 is replaced with the lightweight deep neural network MobileNetv3. Secondly, to improve the learning efficiency and accelerate the convergence speed, the K-Means clustering is utilized to generate a prior box to suit for this experiment. Finally, the confidence loss is redefined and a loss function is proposed that can adapt to the multi-scale to solve the problem of poor detection effect due to the imbalance of positive and negative samples. Compared with the original YOLOv4 model for the surface defect detection of the hot rolled steel strip, experimental results show that the proposed method enhance the mean average precision and the speed about 7. 94% and 4. 52 f / s, respectively. The accuracy of this model is improved effectively while ensuring the detection speed.

Abstract:The sample number of ship target categories collected in actual scenes is not balanced, and the model training easily leads to be overfitting. The data set of the traditional transfer learning is divided into categories, which results in low recognition accuracy of unlabeled new categories. To solve the above problems, a small sample ship identification algorithm based on the fusion of the crosstarget universal global attention mechanism and the relationship measurement network is proposed. This method introduces the universal attention mechanism into the relation network, uses the original features extracted by relation network, and smooths the target features between imbalanced categories through the universal attention mechanism, and compares them with the original features extracted by the relation network. After feature fusion, feature distance measurement is performed. This method enhances the consistency among universal features, which is conducive to learning invariant target features and improve the performance of ship recognition with few samples and few labels. In this way, the overfitting problem caused by imbalance of categories in the training process could be solved. Using the ship data set collected and produced by ourselves to test the proposed method, the recognition accuracy is improved 5. 6% (5- shot) and 3. 2% (1-shot). The impact of imbalanced category on the model ship recognition is reduced, and the robust of the model is enhanced.

Abstract:To solve the nonlinear and ill-conditioned problem of image reconstruction in the electrical capacitance tomography (ECT), an adaptive simulated annealing-Levenberg Marquardt ( ASA-LM) joint inversion algorithm is proposed. The new solution generation strategy, energy function definition and annealing strategy of the standard simulated annealing (SA) algorithm are improved. The direct local search method of LM is combined to jointly invert the ECT image reconstruction problem. Meanwhile, the Savitzky-Golay ( SG) filter is used to smooth the capacitance data required for ECT image reconstruction to enhance its signal-to-noise ratio. Finally, simulation and static experiments are carried out and compared with linear back projection (LBP), Landweber iteration and standard SA algorithms. Comparison experiment results show that the ASA-LM algorithm has advantages of high reconstruction image accuracy and fast convergence speed. The image reconstruction quality is significantly improved, and the edge information fidelity is high. The average relative error of the reconstructed image is 0. 331 1, and the average correlation coefficient is 0. 933 1.

Abstract:The slug flow is a common flow pattern in the multiphase flow. The accurate measurement of flow parameter phase fraction is the prerequisite for analyzing the flow state and obtaining the average density of the mixture. It is of great significance to the fields of petroleum and chemical industry. First, according to Fresnel′s law and Maxwell′s equation theory, the distribution characteristics of light on the pipe arc is derived. A single-shot and multiple-receiving near-infrared sensor is designed to study the propagation characteristics of near-infrared spectrum in slug flow. Experiments prove that the voltage value received by each probe on the circumference of the pipeline is cosine with the installation angle of the probe. According to Lambert-Beer law, the relationship among the voltage value received by the near-infrared probe, the installation angle of the near-infrared receiving probe and the phase fraction deduced is achieved. The phase fraction measurement model of the slug flow is formulated. The experiments are implemented in a horizontal transparent pipe with an inner diameter of 50 mm, the gas flow range is between 0. 5 m 3 / h and 2 m 3 / h, and the liquid flow rate is between 5 m 3 / h and 9 m 3 / h. Research results show that the relative error of the prediction model of phase fraction is within ±7% .

Abstract:In the pipeline magnetic flux leakage detection, defect inversion is the core part of pipeline fault diagnosis. Considering the complexity of the magnetic flux leakage signal and the variability of the pipeline environment, the commonly defect inversion methods mostly use sensor uniaxial information, which may cause the defect inversion to bring the problems of low defect estimation size accuracy and poor model versatility. It is difficult to meet the requirement of practical application. This article proposes a three-axis fusion-based defect inversion algorithm for magnetic flux leakage internal inspection data, which significantly improves the inversion accuracy of magnetic flux leakage defect. The method mainly consists of two parts. First, the proposed weighted random forest algorithm is used to realize the defect inversion of single-axis signals. Secondly, the three-axis inversion result decision fusion is achieved through the designed fuzzy inference system. Then, the precise defect size is achieved. Finally, the evaluation of the method is realized through simulation data and practical pipeline data. Experimental results show that the length accuracy of the defect inversion method is increased by 23% , the width accuracy is increased by 13% , and the depth accuracy is increased by 14. 7% , which have good experimental results.

Abstract:Conventional rotor unmanned aerial vehicles (UAV) mostly adopt the collinear design. It can only generate thrust in the vertical direction, which greatly limits its application in the physical interaction tasks. To solve this problem, this article studies a fully-actuated hexarotor with double-tilted rotors. It can realize independent control of position and attitude by adopting the design method of non-collinear rotor shaft. An improved integral sliding mode ( ISM) controller with chattering suppression is proposed. It is compared with the PID controller, integral backstepping (IB) controller and traditional integral sliding mode controller. Simulation results show that the proposed improved ISM controller can achieve the independent control of the position and attitude of the rotor UAV, which can effectively overcome the uncertainty of its own model parameters and the external wind field disturbance to complete the fixed-point hovering and complex trajectory tracking. Experimental results of the implemented prototype show that the designed fully-actuated rotor UAV can maintain the horizontal attitude during the long-distance lateral motion. The pitch and roll angle errors are controlled within two degrees.

Abstract:To realize the high-precision control of the inertially stabilized platform ( ISP ) in the complex multi-source disturbance environment, a high-precision control method of the ISP based on the fuzzy sliding mode is proposed. By suppressing the total disturbance, the influence of the multi-source disturbance is reduced, which improves the control accuracy of the ISP. Firstly, all disturbances suffered by the ISP are regarded as an entirety. Based on the sliding mode equivalent control, a global fast terminal sliding mode controller is designed to realize the system state quickly and accurately to the equilibrium state without leaving the sliding mode surface. Secondly, to enhance the robustness and chattering of the control system at the same time, this article fuses fuzzy switching gain regulation on the basis of the sliding mode control platform. By designing the fuzzy rules, the switching gain in sliding mode control of the inertial stabilization platform is adjusted in real time to eliminate the influence of interference by using the switching gain. Finally, simulations and experiments are carried out. Results show that the sliding mode control method based on the fuzzy switching gain adjustment method can improve stability accuracy and disturbance rejection ability. Compared with PID control and global fast terminal sliding mode controller, the stability accuracy values are improved by 32. 7% and 15. 3% , respectively.