Abstract:The space pointing determination technology is a basic and key technique in aerospace development. Orbital multi-physics field has distributed-parameterized, strong time-variant, and inter-coupling features, which has become the major obstacle for significant advances in measurement accuracy level. In this article, we firstlyintroduce the complex effects of space multi-physics fields. Then, the research status of multi-physics field error is summarized from threeaspects of analysis and suppression methods, on-ground and on-orbit calibration consist of the facilities and key technologies. By accurately simulating and measuring the multi-physics field as orbital environment,and creating an innovative milliarcsecond space pointing calibration system, the infinitesimal error sources caused by space multi-physics field can be unambiguously identified,the inner coupling mechanism of space multi-physics field can be revealed and the theoretical blank in space complex environment effects on pointingdetermination can be compensated. In this way, the space pointing measurement accuracy is pushed to milliarcsecond level. Finally, some suggestions and prospects are given based on the key problems in the research and engineering implementation of space multi-physics field effects.

Abstract:To address the problem of vortex metering in wet gas,tlle overreading correction and wet gas flow measurement method are proposed, which is based on acceleration detection. The high-frequency response three-axis acceleration probe is designed, and the sensing element,probe size and its package are optimized.The calibration results in dry gas show that the metering accuracy is ±1.0%and the linearity is 1.06% in the Reynolds number range of 4.43×10＋～1.81×10³.Then，the tests are implemented on various wet gas conditions(carrier gas pressure and flowrate,liquid flowrate),and the output frequency and acceleration amplitude are acquired. The models of meter overreading,and the acceleration amplitude areestablished with the gas and liquid Weber numbers, respectively. Finally, the wet gas flow measurement model is formulated by the simultaneous equations, and solved by using the Newton iteration algorithm. The test results show that the gas prediction error is within ±1.0% with uncertainty of 0.46%,and full-scale prediction error for liquid flow within ±15% with uncertainty of 10.04%. Comparedwith the error up to 8% by the traditional method,the accuracy of wet gas metering is largely improved. Boththe gas and liquid flows can be tested online.

Abstract:To eliminate the external precision displacement sensor, reduce the volume of the micro assembly and micro operation system, and reduce the system cost, the self-sensing method is used to obtain the output displacement of the piezoelectric actuator. Firstly, the deformation of the piezoelectric actuator under the action of voltage is utilized. Meanwhile, the piezoelectric actuator is be polarized and charges on the surface of its surface is generated. Adisplacement self-sensing method of the unbalanced current integral piezoelectric actuator is proposed. Secondly, considering the influence of the leakage current and dielectric absorption current of the piezoelectric actuator, as well as the bias current of the operational amplifier on the self-sensing accuracy, the self-sensing expression and the self-sensing parameter identification method which can accurately reflect the output displacement of the piezoelectric actuator are given. Finally, on the basis of identifying the self-sensingparameters of the piezoelectric actuator,experiments evaluate the effectiveness of the proposed precision displacement self-sensing method. The results show that the self-sensing displacement has a sub-micron resolution of 0.24 μm. In terms of static displacement self-sensing，the maximum relative error of self-sensing displacement is 1.17% in the displacement range of 0～118.9μm. In terms of dynamic displacement self-sensing， the maximum relative error of self-sensing displacement is 1.45% in the frequency range of 0~100 Hz. The proposed precision displacement self-sensing method cannot only realize the static long-time, but also achieve the dynamic fast high-precision and high-resolution displacement self-sensing.

Abstract:To quantitatively measure the thickness d and refractiveindex n of the damage layer on sapphire substrate produced during chemical mechanical polishing,a spectral ellipsometric method is proposed. First, the change of spectral polarization state(i.e., amplitudes ratio and phases difference) of light(from 250 to 1650 nm) reflected on sapphire substrate is measured. Then, the thickness and refractive index of damaged layer are extracted by optical modeling and inversion of measured data. The d and n of substrates polished by Al,O% and SiO, abrasives are measured. Thed of the former fluctuates with the better result about 1.4 nm after 40 min, and the latter continues decreasing with the better result1 nm after 20 min. The n produced by two abrasives is both smaller than that of sapphire crystal. Meanwhile, the experiment and simulation analysis indicate that the change of phase difference is similar with d,which can be used to quickly evaluate the change of damage layer thickness because of no requirements for modelling and inversion. Therefore, the proposed method can be utilized tomonitor the machining process as a nondestructive optical way.

Abstract:The difficulty in maintaining high prediction accuracy ofmachine tool thermal error prediction models under different working conditions is an important reason for thermal errors' poor actual compensation effect. This article proposes a modeling method for the thermal error of machine tools under different working conditions based on transfer learning. Firstly, the kernel mean matching algorithm is used to obtain the transfer weight between machine tool temperature data under different working conditions. And a thermal error modeling method based on transfer learning is proposed. Furthermore,the significance of differencesin thermal error data under different working conditions is tested, and a thermal error prediction modelis formulated by using the proposed method to analyze the modeling effect. Then, the actual prediction performance of the proposed modeling method and commonly used modeling methods are compared and analyzed. Finally, the compensation validation experiments areconducted to evaluate the effectiveness of the proposed method. The results show that the modeling method based on transfer learning proposed in this paper can effectively improve the modeling effect. The prediction accuracy and robustness of transfer learning combined with the LASSO algorithm under different working conditions reach 3.73 and 1.14 μm，respectively. After compensation，the thermal errors in the X/Y/Z directions of the machine tool remain within -2.3~3.1 μm，-3.4～3.9μm，and-3.3~4.6 μm，respectively.

Abstract:Crankshaft journal roundness is an important indicator for evaluating the qualification and machining accuracy of the crankshaft. To address the problem of the inability to directly use the detection data of the crankpin for roundness error evaluation due to the revolving motion of the crankpin along the main journal during the measurement process, a roundness error detection model based on a motion coordinate system is formulated to realize the conversion and processing of the detection data of the crankpin. Meanwhile, by deeply analyzing the applicable conditions of three least squares methods used for roundness error evaluation and combining the characteristics of the sampled data, high-precision detection of theroundness error of the crankpin is achieved. Take a certain model of engine crankshaft as an example,a large sample error detection test is carried out, and the results are compared with the minimum zone evaluation，with a deviation within 1 μm.The data analysis shows the theoretical correctness andpractical feasibility of the proposed method for detecting the roundness error of the crankpin.

Abstract:Shear yield stress is one of the main parameters reflecting the rheological characteristics of the magnetorheological fluid. The stable and controllable magnetic field directly affects the measurement accuracy of magnetorheologicalfluid shear yield stress. Therefore, the reasonability of magnetic field design has an important impacton the rheological performance testof magnetorheological fluid. In this article, a concave magnetic circuit is designed to change the magnetic field structure by adjusting the position of the coil. The magnetic field lines pass perpendicular through the flow direction of the magnetorheological fluid, and the detachable combined magnetic circuit design realizes the continuous measurement of magnetorheologicalfluid under the premise of ensuring the magnetic field strength. In addition, the distribution of magnetic field strength under differentcurrents is analyzed, and mechanical properties such as shear yield stress of magnetorheological fluid are tested based on the optimized magnetic circuit. Compared with the mainstream standard test instruments,the average relative deviation value of shear yield stress measured by the magnetorheological fluid test system with concave magnetic circuit structure is about 10%,and the repeat error is within 6.34%. Results indicate that the magnetic circuit structure is a feasible method for the design of magnetic field devices in magnetorheological test.

Abstract:To address the error problem of the deflection of the vertical(DOV),which restricts the precision of high-precision inertial navigation system(INS),the influence of DOV on the horizontalposition error of INS and the index requirements of DOV for error compensation of INS at all levels are studied. Firstly, the direct difference method and the fourth order Runge Kutta numerical updating algorithm of the INS error term caused by DOV are derived. The updating effects of the horizontal position errors of the two algorithms in different regions are compared and analyzed. Then, the INS is compensated with three kinds of resolution DOV grid data. Finally, the effect of DOV compensation frequency on position error compensation is analyzed and DOV compensation experiment of vehicle navigation is carried out. The simulation and experimental results show thatboth error update algorithms can effectively calculate the horizontal position error. The maximum DOV can cause a position error of nearly 3 000 m,and the horizontal attitude error and azimuth attitude error drift about 18" and 72" in 1 h. After DOV compensation,the horizontal positioning accuracy is improved by about 230 m.

Abstract:To address the problems of low accuracy and poor reliability in many existing automatic measurement methods of shield tail clearance, a new visual measurement method based on double line laser reference ruler is proposed inthis article. The observation datum is constructed by two parallel narrow linear lasers, and the key features of shield shell and segment are captured by traditional image processing technology. Combined with the ruler back calculation error compensation technology, the high-precision real-time calculation of the measured value of shield tail clearance during shield tunneling is realized. Field experiments show that the repeatability measurement accuracy of the proposed method is better than 1.2 mm, the absolute measurement accuracy is better than 2 mm, and the measurement error is less than 1.5 mm. This method can realize the automatic real-time accurate measurement of shield tail clearance and has high reliability. It has been applied inmany subway tunnel construction sites in China.

Abstract:To solve the problem of obtaining the servo motion anglequickly and accurately in the process of tracking recovery of laser tracker,this article proposes a calculation method of servo motionangle for tracking recovery of laser tracker. Firstly, the principle of tracking recovery of laser tracker based on active infrared detectionand the characteristics of visual target detection of cooperative target are expounded, and the problem of nonlinear change of servo motion angle caused by the non-coaxis and non-parallel problem of optical axis and visual axis in the tracking recovery process is analyzed.Secondly,the calculation model of the relationship among the target distance, the target pixel deviation value and the servo motion angle is formulated, and the model parameter tuning based on the tuning data is completed. Finally, the verification experiment and the accuracy analysis are carried out. The results show that the difference between the tracking recovery angle obtained by the calculation model and the actual tracking recovery angle is not more than 0.007 7°, and the maximum offset distance error is less than 2 mm,whichmeet the application requirements of laser tracker tracking recovery.

Abstract:In unstructured environments, the 6-DoF object grasp is ahighly challenging task in the field of intelligent service robotics. In such scenarios, robots need to deal with interferences from objectsof different sizes and shapes,as well as environmental noise,making it difficult to generate accurate grasp poses. To address this problem, this article proposes a grasp generation method based on multi-scale features fusion and grasp quality evaluation. Firstly, an adaptive radius query method is introduced to solve the issue of key points query anomalies caused by uneven point cloud sampling in real environments. Secondly, a grasp generation network is designed to fuse multi-scale features and grasp quality assessment, which enable the generation of rich 6-DoF grasp candidates. Finally a grasp quality assessment method is defined, which includes force closure score,contact surface flatness,edge analysis,and centroid score. These criteria are applied to generate new grasp confidence score labels on a standard dataset and incorporated into the grasp generation network. Compared with the current state-of-the-art method FGC-GraspNet, the experimental results show that the described method improves the average accuracy by 5.9%, the success rate of single-object grasp by 5.8%,and the success rate of multi-object scene grasp by 1.1%.In summary,the proposed method has effectiveness and feasibility, which has good adaptability in single-object scenes and multi-object scenes.

Abstract:In actual engineering,poor joint control parameters caneasily cause end-jitter in industrial robots. Recognizing the cause of the jitter can help locate joint anomalies and optimize control.However,there are problems with identifying the cause of jitter in industrial robots, such as high redundancy of cyclic signals, multiple jitter directions, and missing sample labels. Therefore, a method for recognizing the cause of jitter in industrial robots based on data fusion and the improved momentum contrast(MoCo)is proposed. Firstly, the data of each sensor at the end of the industrial robot are sequentially subjected to data dimensionality reduction, data expansion, horizontal splicing fusion, and dimensionality reduction to construct fusion samples that reflect sufficient and comprehensive jitter direction and state information. Data dimensionality reduction before fusion can reduce the redundancy of periodic samples and improve the efficiency of sample fusion,while dimensionality reduction after fusion can avoid the complexity of model training caused by excessively long fusion samples. Secondly, a small number of labeled fusion samples are supervised by the positive encoder classification channel output information before MoCo to guide feature clustering. Then,an improved contrastive learning strategy is implemented. The unlabeled fused data features extracted by the positive encoder arecompared with the cluster centersof the negative sample features saved by the momentum encoder, and the cluster centers with the highest feature similarity are removed to reduce the false negative sample interference of the comparison category error. And the encoder training is completed by symmetrically swapping the inputs of the two encoders for two comparison loss calculations. Finally, the cause ofjitter in industrial robots is identified by adding a Softmax classifier to the encoder classification channel. The experimental results showthat the recognition accuracy of the proposed method the causes of industrial robot jitter in different working conditions is larger than 90%, which shows the effectiveness of the method.

Abstract:This article proposes a logistics robot scheduling system and its bidirectional synchronous skip point search algorithm to address the increasing demand for safety, real-time, and efficiency in path planning for logistics robots in material delivery. Firstly, the functional requirement of the logistics robot scheduling systemare analyzed and the overall implementation scheme is designed. Secondly,the ortho-hexagonal raster map is used to replace the traditional raster map, and on the basis of which an improved bidirectional simultaneous jump point search(JPS)algorithm isproposed. This algorithm adjusts the node expansion rules of the traditional JPS algorithm to adapt the new map, and introduces the grey wolf optimization algorithm to guide the relative direction of expansion,ensuring that both sides of the search can meet.By comparing the proposed algorithm with the traditional algorithm through simulation tests, the proposed algorithm is 34% faster than the unidirectional JPS algorithm for orthogonal quadrilateral raster graphs and 23% faster than the unidirectional JPS algorithm for orthogonal hexagonal raster graphs. The results show that the proposed algorithm is more efficient and secure in planning. In addition, the physical experiments conducted show that the proposed two-way synchronous JPS algorithm has a better path length compared to the comparison method, and the logistic robots are safer to walk under the supervision of the scheduling system.

Abstract:Due to the inherent uncertainties associated with extraterrestrial bodies' regolith environments, the implementation of shallow drilling sampling tasks can be easily hindered by potential obstacles. There is few researches addressing the perception of the environment and the adjustment of mechanism posture during the sampling process. To improve the success rate of sampling and ameliorate the resistance of end-effector during operations, a novel autonomous obstacle avoidance control method is proposed for extraterrestrial regolith sampling tasks. A three-degree-of-freedommanipulator is designed, which is capable of assisting drilling with rotational motion for obstacle avoidance,and kinematic analysis is performed. By utilizing a three-dimensional force sensor to acquire the sampler's force information, the admittance control and rotational motions are combined for obstacle avoidance. To evaluate the feasibility of the method, several sets of simulation experiments and physical experiments are conducted. Experimental results show that the proposed method can effectively assist the sampler in avoiding obstacles, and significantly reduce the resistance encountered by the end effector during sampling, with improvements in the maximum forces experienced in the X,Y,and Z directions reaching 46.7%, 57.0%,and 64.9%,respectively. In addition, the method addresses the deficiencies of conventional compliant methods, which are prone to interference when applied to shallow drilling.

Abstract:Acupuncture is a practical discipline,and the experience of operational training is very important for acupuncture medical students to master acupuncture techniques. To address the problems of low accuracy and realism of traditional training methods, this article establishes a virtual acupuncture training platform based onTouch X and unity. The force model is formulated by collecting data from the needle punched pork experiment and the lifting and twisting techniques. For the insertion method, the whole model is divided into two parts,including penetration force modeling and insertionforce modeling. Compared with the existing penetration force model, this article adds attenuation force modeling,which makes the layering sense of the needle punching process more realistic. The Karnopp model is used to solve the damping function per unit length of different tissue layers. Finally, the formula method is used to combine the penetration force and the insertion force to establish a complete mana model of the insertion. The data-driven modeling is used for the twisting maneuver,which is divided into two parts,including dynamic twisting and static decay. Both of them have test set accuracy values of 0.985 08 and 0.992 49,respectively.Finally,the model is evaluated by left-right hand comparison experiment and user perception experiment. The results show that the two acupuncture manipulative models have high accuracy,strong authenticity and good scalability.

Abstract:In the external damping system, the external velocity error has direct influence on the alignment result of the gyrocompassing loop. Based on the characteristics of the gyrocompassing loop,this manuscript deduces the effect of different kinds of sea current velocities to the alignment result in theory, which is further verified and analyzed through simulation. Since the sea current velocity satisfies some mathematical model in a specific sea area, it is possible to be measured and modeled before applied to the damping system for the compensation purpose. Therefore, a current velocity calculation method is introduced by directly comparing the velocity estimated in GNSS/INS integration to the velocimeter's output. The sea current velocity model is formulated, which is based on the calculation result. During the gyrocompassing process, the velocity model is further used to compensate the velocimeter's output in the specific area. The marine experimental results show the effectiveness of the proposed method on improving the attitude accuracy and convergence speed. The horizontal attitude accuracy is improved by 36", and the yaw angle is improved by 25". The horizontal convergence speed is improved by 25 minutes.

Abstract:State of charge(SOC)is the key parameter of the lithium-ion battery management system, which needs to be estimated accurately to ensure the battery's safe operation. The traditionally used data-driven SOC estimation methods(e.g.,neuro-network)have limitations on interpretability and parameter tuning. This article proposes a novel nnethod by combining the local model network(LMN)and the beetle antenna search(BAS)algorithm. Firstly,LMN,known as a grey-boxmodel that can model complex non-linear systems with some extent of interpretability,is employed to partition the working condition space into some sub-regions that can be represented by simple models. Then, they are combined by validation function. Secondly, during the training of LMN,BAS optimization is utilized to find the optimal splitting location and orientation globally, which reaches a good trade-off between the model identification accuracy and the computation complexity. Finally,the proposed SOC estimation method is compared with two existing methods on a lithium-ion battery dynamic characteristic dataset. The RMSE is less than 0.4% on the training set under simple test driving cycle,and less than 0.9% on the testing set under complex test driving cycles. The performance on different temperatures is relativelystable too. Therefore, it shows anexcellent identification accuracy and generalization capability of the method. The advantage of the proposed method is verified on real measured dataset too.

Abstract:In bad weather such as rainy and snowy, the performance of LiDAR can be seriously affected due to the block of rain and snowflakes,which brings great difficulties to 3D target detection.Aiming at this problem, a dynamic outlier filtering algorithm based on the Mahalanobis distance is proposed. First, by establishing the KD tree, the Mahalanobis distance of outlier points is calculated to remove snowflakes noise with different Euclidean distances. After the verification of the Canadian Adverse Driving Conditions open Dataset and practical experiments, the accuracy of the filtering algorithm proposed in this paper is improved by 7.88% and 7.72% relatively, compared with the DROR filtering algorithm in medium and heavy snowy weather. In practical rainy experiments, the algorithm proposed in this study exhibited a relative improvement of 10% precision compared to the DROR filtering algorithm. For target detection applications, the detection accuracy of vehicle and pedestrian detection via the filtering algorithm is also improved by 19.26% and 20.39% relatively, compared to the algorithm using only Pointpillars,which verifies theeffectiveness of this method in dataset and real experimental scenarios.

Abstract:In the visual measurement of rod's angular velocity, the background noise caused by short exposure time leads to serious registration error by mixing and blocking optical flow trajectory. Toaddress this issue,the article proposes a trajectory compensation and noise clipping algorithm. The algorithm takes the difference motionstatistical characteristics between background noise and rod's optical flow trajectory as prior model,and cuts out the background noisetrajectory segments to remove the trajectory aliasing by detecting the speed mutation points through the kernel double sample hypothesis test. By introducing a new optical flow acquisition method which takes the rod hinge fulcrum as the reference, the rod's hinge points of each frame are registered and warpedto the position of the first frame to separate the rod's trajectory from the total composite high-order cycloid to form a low order ideal arc as compensation prior. Secondly, the trajectory is clustered into arcshaped trajectory groups with different radii which fitted by Pratt. Thirdly, the full length trajectory is semi-supervised learned from the v-SVR regression of arc trajectorygroups which acts as a geometric constraint and from x-t dynamic regression to realize compensation. The comparative measurementexperiment of the brush assembly angular velocity and displacement of the needle bed conjugate cam shows that this algorithm can improve the accuracy by 3.26% compared to traditional algorithms such as VBM3D and MeshFlow. The computational complexity is reduced by2 orders. It has broad application prospects in visual fault diagnosis of mechanical rotation motion and digital acquisition of mechanical instruments.

Abstract:The flat panel detector is a key component of cone-beam CT, and the signal crosstalk between pixels is the main factor that causes the spatial resolution of the projection image to be lowerthan the possible highest resolution of the flat panel detector. It is essential to correct the signal crosstalk of the flat panel detector forimproving the detection accuracy of cone-beam CT. In this article, the crosstalk correction effect of the point spread function matrices with different accuracy, the relationship between the point spread function and line spread function, and the similarity between the relationship and X-ray imaging are studied, based on the signal crosstalk correction idea of the point spread function matrix deconvoluting projection image. In addition, a new method for calculating the point spread function matrix of a flat panel detector is proposed,which combines the knife edge method for measuring line spread function with parallel-beam CT scanning reconstruction. In DR/CTscanning experiments, the spatial resolution of DR projection image is improved from about 10 lp/mm to better than 25 lp/mm and thespatial resolution of high-energy CT image is improved from less than 4 lp/mm to better than 5 lp/mm after correcting signal crosstalk by the proposed method. Experimental results show that the proposed method can effectively correct the signal crosstalk of flat panel detector and improve the spatial resolution and contrast of cone-beam CT images.

Abstract:To achieve a cooperative navigation system for multiple autonomous underwater vehicles(AUVs)with three leaders,the localization performance evaluation function is derived without simplifying range measurement equation. In this way, the relationship among localization performance,formation configuration,cluster speed and distance between AUVs is analyzed. Theoretical analysis proves that the localization performance evaluation function can be degeneratedinto the simplest form under certain conditions,and the localization performance evaluation function in the simplest form is consistent with the corresponding results of theerror ellipse method and the Fisher information matrix method. Furthermore, Cramer-Rao lower bound based on localization estimation is derived to improve localization performance evaluation function, and the influence of rang measurement noise on the localization performance is analyzed. Finally, the optimal formation configuration, the suboptimal formation configuration and the worst formation configuration are constructed based on localization performance evaluation function analysis for the three-leaders cooperative navigation system, and the effectiveness of analysis and formation configuration is evaluated by simulation experiments. The results show that reducing cluster speed can improve the localization performance of system in the optimaland suboptimal formation configuration to a certain extent. Under the simulation conditions, the accuracy of localization can be increased by 9.11%. However, changing distance between AUVs has almost no effect on the localization performance. For a cooperative navigation systemin the worst formation configuration, reducing cluster speed and shortening distance between AUVs can both improve the localization performance. If they are both satisfied, the localization performance will be greatly improved. Under the simulation conditions,the accuracy of localization is increased by 63.17%,63.99%,and 81.50%,respectively.

Abstract:With the continuous development of machine vision, visual sensors has advantages of lightweight and low cost. Thus, visual simultaneous localization and mapping(VSLAM)is attracting moreand more attention and becoming a research hotspot. Deep learning has provided new methods and ideas to deal with VSLAM problenns. This article reviews the deep learning-based VSLAM methods in recent years. Firstly, the development history of VSLAM is reviewed, and the basic principle and composition structure of VSLAM are systematically explained. Then, various methods based on deep learning are summarized and analyzed from three aspects, including visual odometry(VO),loop closure detection and mapping. The application of deep learning in visual odometry is described in three parts,which are feature extraction and feature matching,depth estimation and pose estimation and keyframes selection. Based on the different manner of scene representation, deep learning-based methods in geometric mapping, semantic mapping and general mapping are summarized. Thirdly, it introduces various datasets and performance evaluation metrics commonly used in VSLAM at present. Finally, the challenges of VSLAM are pointed out, and the future research trends and development directions of combining deep learning with VSLAM are forecasted.

Abstract:To address the limitations of the static assumption in visual SLAM for dynamic real-world applications,a visual SLAM algorithm is proposed, which is based on Fourier-Mellin transformfor high dynamic environments. It involves Fourier-Mellin transform for motion compensation, employs frame differencing for motion mask generation, utilizes the short-term dense connection network for semantic segmentation to identify potential moving objects,combines motion and object masks to obtain the final object motion region, and eliminates the corresponding feature points in that region. Finally, the pose accuracy is optimized based on stable static feature points. Experimental results demonstrate a reduction of over 95% in absolute trajectory error and relative pose error compared to ORB-SLAM2, and over 30% compared to DS-SLAM. These evaluate its excellent localization accuracy and robustness in complex dynamic scenes. The impact of motion blur and lighting changes on motion detection is effectively mitigated,and the limitations of traditional dynamic SLAM in detecting non-prior motion objects are overcome.

Abstract:The current machine vision-based roughness measurement methods use image metrics with few considerations, no reference making the prediction model accuracy limited and highly influencedby the brightness of the light source. To address this problem, the article proposes a roughness measurement method based on a full-reference color image quality algorithm. Based on the analysis of rough surface imaging mechanism, this method introduces structural information based on visual saliency-induced index(VSI)and proposes an image quality assessment algorithm based on visual saliency structural index(VSIS). Meanwhile, a surface roughness measurement device for grinding samples based on the image quality assessment algorithm is designed. The experimental results show that the proposed VSIS image quality assessment algorithm has a significant correlation with surface roughness(R。). The curve relationship, obtained through the least squares method, enables low-discrepancy and high-precision predictions for grinding samples with a roughness of 0.965 um or greater. The average error and standard deviation for these predictions are measured at 0.111 μm and 0.079 μm，respectively. Compared with the roughness-correlated image characteristic index considering a single factor,VSIS has a better comprehensive performance and can overcome the influence of light source brightness to a certain extent. The method provides an alternative way for non-contact roughness measurement.

Abstract:With the development of the field of marine engineering, how to obtain high-precision three-dimensional point clouds of underwater objects has significant academic and practical value. However,the propagation of light in different media causes change in the optical path,which may lead to refraction distortion in the point clouds obtained through underwater structured light vision measurement and reduce accuracy. To address these issues, this article proposes an tunderwater rotational scanning measurement system based on structured light. An algorithm for axial calibration is proposed to register multi-view underwater point clouds into a unified coordinate system. In addition, an underwater camera imaging model is introduced, whichincorporates refraction compensation. This model accurately describes the optical path of laser propagation between different media. By applying the constraints of the underwater laser plane equation,refraction correction is performed on the underwater point clouds. Thereby, the reconstruction accuracy is improved. Experimental results of underwater measurements demonstrate that the proposed high-precision measurement method can obtain three-dimensional point cloud information of underwater objects. The measurement accuracy reaches 0.2mm when the distance to the target is between 30~80 cm,which could meet the requirements of high-precision three-dimensional measurement of underwater targets.

Abstract:This article proposes an adaptive filtering-collaborative graph optimization navigation method to address the problem of inaccurate sensor measurement covariance in the traditional graph optimization navigation method, which leads to a decrease in estimation accuracy. Firstly, a factor graph model for the INS/GNSS/e-Compass integrated navigation system is established. Then, the adaptive filter is used to pre-estimate the sensor measurement information based on the measurement variance,the measurement covariance matrix of relevant sensors is updated during the filtering process, and thepre-estimated result is added to the factor graph as the variable node. Finally, the sliding window is used to control the optimization range, and the nonlinear optimization of the variable nodes within the sliding window is performed. Thus, the final navigation states are achieved. Simulation and experimental results show that the proposed method has adaptability to the mismatch of sensor measurement covariance and can achieve efficient and reliable navigation positioning in different environments. Compared with the traditional graph optimization method,this method improves the positioning accuracy by 30%and the calculation efficiency by 12%.

Abstract:Traditional 3D point semantic segmentation networks based on PointNet++ tend to sacrifice the accuracy of minority classes to maintain the overall accuracy dominated by majority classes. A new CNN is proposed to improve the segmentation accuracy of PointNet+ when processing airborne LiDAR point clouds with long-tailed distribution,which mainly consists of two aspects. The first is cluster-based farthest point sampling(FPS).Through intra-class FPS under proportional constraints, meanshift clustering based on confidence and zoning FPS combined with neighborhood compensation, the samplesof minority classes in airborne LiDAR point clouds can be retained to the maximum extent, and can be well learned by the network through re-weighting. The second is local feature learning under the spatial self-attention mechanism. By using different spatial encoding methods, a new spatial self-attention mechanism is constructed to facilitate learning the complete structure of the target from sparse sample data. Therefore, the learning ability of the network model for minority classes is improved while ensuring the good learning ability of the majority classes. Experiments on public data set show that the overall accuracy(OA)and F,score in this article have a significant improvement,which is 6.3% and 6.6% higher than those of PointNet+Compared with other 6 networks based on PointNet+ and the top10 network model in recent publications, the proposed algorithm has the best performance,good generalization ability and application value.

Abstract:To address the problems of traditional model predictivedirect torque control strategy, such as low computational efficiency, high voltage jump and poor steady-state characteristics, an improved model predictive direct torque control strategy is proposed. To improve the steady-state characteristics of the motor,36 virtual voltage vectors are constructed basedon the principle of the nearest three vectors. The action time of redundant vectors in the virtual voltagevector is adjusted by combining the principle of no beat. To prevent excessive voltage jump in the output line of the inverter from causing adverse effects on the motor, only voltage vectors with voltage jump not exceeding Ua/2 are selected as alternative voltage vectors. Meanwhile, the reference voltage vector is predicted by combining the model with the deadbeat principle,and the voltage vector located in the same sector as the referencevoltage vector is selected as the final candidate vector set. Compared with traditional control strategies under rated operating conditions, the experimental results show that the improved control strategy reduces the electromagnetic torque, stator flux amplitude error, and current harmonic distortion rate by 37.42%,32.00%,and 44.52%,respectively. The program execution time is reduced by approximately 11.52%.

Abstract:A new non-invasive measurement method of air gap magnetic density of bilateral permanent magnet synchronous linear motor (BPMSLM)based on the tunneling magnetoresistance(TMR)sensor and the dynamic integrated neural network prediction model based on noise-boosted convolutional neural network(NBCNN),long short-term memory network(LSTM),attention mechanism is proposed. Firstly, the analytical model and the finite element model of air gap magnetic field of linear motor are tormulated as data basis. The nonlinear mapping relationship between external space stray magnetic field and internal central air gap magnetic field of linear motor is explored. Secondly, the TMR sensor is introduced to measure the external stray magnetic field signalof the linear motor,the installation position of the sensor is optimized,and the similarity characteristicsof the internal and external one-dimensional magnetic density signals are matched to obtain the optimal measurement position of the sensor. Then, taking the external straymagnetic field data of the motor as the input and the internal air gap magnetic field data as the output,a high-precision mapping model of the internal and external magnetic fields of NBCNN-LSTM-Attention network is established to realize the non-invasive high-precision measurement of air gap magnetic density. Finally, the experimental platform for measuring the air gap magnetic density of linear motor and the experimental platform for comparative measurement of Gauss meter are built,which verifies the advancement and superiority of the proposed method.

Abstract:The relationship between the optical axes of the transmitter and the receiver of the non-coaxial lidar is affected by the external environment,such as vibration,temperature,pretightening force,etc. It leads to the geometric overlap variation of the lidar system.In further,the lidar detection accuracy and stability in far range is affected. To solve this problem, a hardware and software design of the self-alignment of geometric overlap function in the non-coaxial lidarsystem is proposed. The self-aligning system consists of a two-phase stepper motor driver, and a two-dimensional inclination adjustment platform. The laser echo signal strength is used as the correction criterion to find the best matching position of the receiving opticalaxis of the non-coaxial laser radar system. The combination of coarse scanning and fine scanning of self-alignment method is selected based on the comprehensive consideration of alignment accuracy and alignment efficiency. The coarse scan adopts the circular scanning,which can lock the transmitter optical axes within the inclination range of 0.4 mrad×0.4 mrad in 961 s. The fine scan adopts S-shaped scanning, which can achieve the inclination adjustment accuracy of 0.02 mrad in 441 s. The preliminary experiment results show that the RSCS of lidar echo signal after the self-alignment is distributed with the lidar equation, which presents an improved performance oflidar detection and can be applied for the all-day lidar measurement without man-made operation.

Abstract:The domain adaptation problem has been widely studiedin the field of mechanical equipment fault diagnosis. At present, the most closed set domain adaptation methods generally assume the source domain and target domain share the same label space,which is not practical in real application. This can be called open set domain, because novel fault classes may actually emerge, these conventional methods which only rely on marginal distribution alignment are difficult to separate the new emerging classes and known classes. One of the typical open set domain adaptation problems isthat the label spaces of source domain and target domain are partially overlapped. In this article, a novel open set domain adaptation method based on adversarial dual classifiers(OSDA-ADC)is proposed to address this issue. Two neural networks with the same structure are introduced for adversarial training to enhance the domain adaptive performance of the model for known classes identification. The maximization and minimization entropy strategies of source domain and target domain, as well as the binary cross scheme of target domairl sample output are used to establish a boundary to isolate unknown classes. In addition, the bearing data set and the self-priming centrifugal pump are selected to evaluate the effectiveness of the proposed method. The experimental results show that the proposed method can more accurately identify the existing known fault classes and new emerging unknown fault classes of mechanical equipment than the typical closed set domain and open set domain models. In each diagnosis task,the proposed method can achieve more than 90% accuracy.