Robotic disassembly has gradually become an important basis for intelligent production.It is particularly important to improve the efficiency of robotic disassembly line and make its load balance.The influence of disassembly task order on disassembly line can not be ignored.To solve the balancing problem of robotic disassembly lines considering disassembly sequence, a multi-objective optimization model was established with the number of workstations, smoothness index and maximum operation time of all workstations as optimization objectives.The ANP-entropy weight method was proposed to assign weight to each objective, and an improved discrete particle swarm optimization algorithm was proposed.Experiments show that the designed coupled weighting method can quantify the target weight more reasonably.It is verified that the algorithm is effective and its performance is better than the comparison algorithms.It also shows that the proposed research method is effective and feasible, and can solve the problem efficiently.

In order to solve the mechanical optimal path planning problem, a RRT^* algorithm (GRU-RRT^*) integrating convolutional neural network GRU is proposed.In this study, the gated loop unit GRU is applied to the sampling based planner (GRU_Sampler).The neural network planner is used to provide nodes for rapid exploration of the random tree, predict the most promising reachable state of the next random sampling, significantly reduce the running time of the algorithm, and improve the planning efficiency of the improved RRT^* algorithm.In addition, the adaptive variable step size is adopted, and the environment information is used to generate an appropriate step size to avoid the problem of reducing the convergence speed caused by the step size.Finally, the cubic spline interpolation algorithm is used to smooth the trajectory of the manipulator to reduce the impact loss of the manipulator.The GRU-RRT^* algorithm is simulated in 2D, 3D and Rviz maps respectively.The experimental results show that the improved algorithm effectively reduces the search time, improves the efficiency of robot arm planning, and has good smoothness.

Aiming at the problems of the traditional rapid-exploration random tree (RRT) algorithm in path planning, such as slow convergence speed, long path and low success rate in complex environment, an improved RRT path planning algorithm was proposed.In this algorithm, an expansion angle control strategy based on the duty cycle of obstacles is introduced to control the expansion direction of new nodes through the distribution of obstacles near the parent node, which increases the planning speed of the algorithm and ensures the success rate of the algorithm.Then the path is smoothed by pruning optimization and second order Bessel curve.Multiple simulation results show that compared with the original RRT algorithm, the improved RRT algorithm generates shorter path length, faster time, higher expansion angle convergence, and greater planning success rate.

In view of the fact that the offline Cartesian space trajectory planning of the traditional manipulator cannot solve the problem of the contact with the environment, a real-time compliant trajectory planning method based on admittance control was proposed; The trajectory planning of manipulator in Cartesian space was divided into position planning and attitude planning, and the two were normalized by the time scale parameter s(t).The trajectory planning method was illustrated by taking the most common linear motion and circular motion as examples; An admittance controller was constructed, and the information obtained by the six-dimension force sensor was substituted into the controller to calculate the adjustment of the end pose; On the basis of trajectory planning, admittance control is introduced to solve the problem that trajectory planning cannot perceive the forces of the external environment in real time; Finally, the verification experiments of the manipulator were built, and the experiments of linear motion and circular motion were carried out; Experimental results show that the proposed method can not only achieve the high precision of the manipulator trajectory, but also actively comply the contact force of the external environment, which improves the stability and safety of the control system.

Aiming at the accuracy and efficiency of solving the transmission accuracy of the herringbone planetary gear train considering the error, a semi-analytical method for modeling and analyzing the transmission accuracy of the herringbone planetary gear train is proposed to accurately and efficiently analyze the influence of manufacturing errors on the transmission accuracy of herringbone planetary gear train.Based on the slicing method, a quasi-static analytical model of the herringbone gear planetary gear train considering the pitch error and double helix alignment error is established, in which the compliance matrix of the inner and outer gear pairs of the herringbone gear is obtained by the finite element substructure method; The nonlinear differential equation system of the model analyzes the motion (displacement/rotation angle) of the planetary transmission under the action of error and load, and forms a semi-analytical method for analyzing the transmission accuracy of the herringbone planetary gear train; Taking the herringbone planetary gear train of three planetary gears as an example, the influence of pitch deviation and centering error of double helix on transmission accuracy under load condition is analyzed, and the calculation efficiency is more than 300 times higher than that of finite element under the same precision condition, which illustrates the validity and accuracy of this model.

To solve the problem of poor loading performance of the electric load simulation system of aircraft rudder due to nonlinear interference and extra torque, the linear voice coil motor was used to change the traditional loading method, and a compound control strategy based on incomplete random firefly algorithm, combined with disturbance signal feed-forward compensation control and multi-condition expert control was proposed.In the feed-forward part, the position signal output by steering gear is used as disturbance, and the feed-forward compensation controller is calculated and designed to suppress it.In the feedback part, multi-condition expert control is adopted, and expert control rules are designed by judging the simulated pneumatic load conditions and error signals.At the same time, incomplete random initialization method and dynamic variable step size search method are used to improve the firefly algorithm, so that it can complete the optimization of expert-controlled parameters.The results show that the loading accuracy and loading error of the system under the action of the composite controller are better than that of the traditional PID control system, the redundant force can be suppressed up to 78%, and the control performance is significantly improved.

To solve the problem that FastSLAM algorithm needs to increase the number of particles to improve the accuracy and the lack of particle diversity caused by resampling, a particle recombination FastSLAM algorithm optimized by improved INFO (weighted mean of vectors algorithm) is proposed.First, the latest observation information is added into the prediction of particle set and the particle fitness value is calculated through INFO.In order to enhance the ability of INFO to jump out of the local optimum, Cauchy mutation is performed on the optimal individual; Secondly, the predicted particle set is updated by improving the characteristics of strong optimization ability and fast convergence speed of INFO, so that the pose of the particle set is closer to the expected value before calculating the weight, so as to improve the estimation accuracy; Finally, in the resampling stage, the particles retained after screening and the remaining particles are recombined into new particles to increase particle diversity.The simulation results show that the IINFO FastSLAM algorithm has higher accuracy and better robustness than FastSLAM and INFO FastSLAM algorithms.

Aim at the key feature of shear performance of aviation fasteners, a three-dimensional numerical model of Ti-6Al-4V fasteners under double shear conditions of blanks and heat treatment is established.The shear resistance of Ti-6Al-4V fasteners and their shear evolution process are correspondingly revealed.The results show that the model can accurately simulate the load-displacement curve of titanium alloy fasteners under double shear conditions, and the average error between the shear strength obtained and the experimental value is 0.5%.After heat treatment, the shear strength of Ti-6Al-4V fasteners is increased by 35%.In the process of shearing, the stress concentration phenomenon of the workpiece gradually extends from the contact of the lower knife surface to the entire shear surface until the workpiece fails to be sheared.This research can provide a theoretical basis and guidance for improving the shear strength of titanium alloy fasteners.

Aiming at the phenomenon that the frequency response function of tool tip changes due to the change of machining position of machine tool spindle and tool replacement, a prediction method of tool tip machining space frequency response function based on genetic algorithm optimized inverse neural network (GA-BP) and substructure response coupling method (RCSA) was proposed.In this method, the machine tool system is divided into three sub structures:tool holder base, residual tool holder and tool.GA-BP algorithm is adopted to build the prediction model of the frequency response function of the tool holder base; The parameters of tool holder tool interface are identified by genetic algorithm; Finally, the RCSA method is used to predict the tool tip frequency response functions of different machining positions and different tools.The results show that the predicted value of the tool tip frequency response function is basically consistent with the measured value, and the errors of the first and second natural frequencies are not more than 5%, which effectively verifies the feasibility and accuracy of this method, and provides an effective reference for the prediction of the frequency response function at different positions and under different tools.

Aiming at the problem that the contour path planning in the current slicing process of FDM 3D printing tends to fall into the local optimal path length, which leads to increased energy consumption, a contour path planning method for low-carbon and energy-saving was proposed.Firstly, the mathematical description model of contour path was established according to the printing sequence of multi-contour in FDM 3D printing.Then, combining the advantages of the fast convergence speed of the genetic algorithm and the ability of the simulated annealing algorithm to accept poor solutions, a contour path planning method based on the genetic simulated annealing algorithm (GSAA) is established.It includes:Determine the geometric center of each contour and regard it as the initial point set of the algorithm; Genetic simulated annealing algorithm is used to solve the optimal connection path of geometric centers of each contour; The starting point of printing of each profile is established based on the path, etc.Finally, a case study verifies the feasibility of the proposed method.

Due to a large amount of calculation required while optimizing, it could not ensure the efficiency using traditional finite element method.In this paper, the reduced-basis method (RBM) is used for rigid-flexible coupled motion stage to achieve fast optimization.This article introduces the fundamental theory of reduced-basis method and the simulation model of the rigid-flexible coupled linear stage, then analyzes the error of the displacement results calculated by the reduced-basis method.The maximum relative error is 1.8×10^-3, which shows that this method not only has high efficiency, but also has good accuracy.Finally, genetic algorithm was introduced to optimize the structural parameters of the stage, and ABAQUS was used to verify the optimization result.The results have shown that the tilting angle of the optimized stage is reduced by 17.2%, proving that the combination of reduced basis and genetic algorithm can not only avoid the occurrence of local optimal solution, but also achieve rapid optimization, and greatly shorten the design cycle.

Independent joint servo systems in manipulators driven by tendon-sheath transmission are affected by time-varying pose and joint flexibility, which will lead to vibration problems, and then affect the motion accuracy of manipulators.In this paper, the control strategy of fuzzy tuning is proposed to suppress the vibration of manipulators indirectly by improving the angle control accuracy of servo systems.Firstly, according to the principle of tendon-sheath transmission, the dynamic model of the flexible joint servo system of manipulators is established, and the transfer function from the load speed to the electromagnetic torque in the speed loop of the system is obtained.Then, according to the pole placement strategy, the proportional integral (PI) controller parameters of manipulators in different positions and posture are appropriately selected, and the fuzzy adaptive control strategy is applied to realize the real-time change of the controller parameters.Finally, the effectiveness of the proposed method is verified by numerical simulation and manipulator control experiments.The experimental results show that the proposed control strategy can better counteract the speed fluctuation caused by the change of moment of inertia, thus improving the speed output of the flexible joint.

Proportional-integral-derivative (PID) controllers are widely used in the field of precision control, but there are still some difficulties in the adaptive and precise adjustment of its parameters.This article is aimed at the precision motion stage driven by linear motor.By constructing internal model controller to design PID parameters, the difficulty of parameter adjustment can be reduced.In order to establish the precise mathematical model required by the internal model controller, based on the research of reinforcement learning methods, this paper proposes an adaptive internal model filter coefficient optimization method based on the probabilistic inference for learning control (PILCO) algorithm.Fitting the probabilistic dynamics model based on input filter coefficient and output error, and the filter coefficient is further optimized through strategy evaluation and strategy optimization.The proposed internal model control method is verified by following error experiments on the linear motor motion stage.The experimental results show that the method proposed in this paper can significantly reduce the motion following error.Compared with the general internal model control method, the control method in this paper can reduce the average overshoot error of the trapezoidal curve positioning process by 86.667%, and the average following error of the sinusoidal curve is reduced by 85.950%, which effectively verifies the adaptive control performance of the method in this paper on the precision motion stage.

For the complicated bolt disassembly problem of shield segment die, this paper has fitted and optimized the movement trajectory of the manipulator when the bolt trajectory is obtained.The planar four-bar mechanism and flexible parts are combined to realize the established function of the manipulator.Firstly, the plane four-bar mechanism was used to fit the bolt trajectory, and the objective function was designed by the sum of the distance between the given trajectory and the fixed-point trajectory of the plane four-bar mechanism.The four-bar parameters were optimized and solved by particle swarm optimization algorithm.Then a flexible part is added to the planar four-bar mechanism to optimize the trajectory of the manipulator, and the two forms are compared by simulation analysis.Finally, the experimental model of the bolt-removing manipulator was built to verify the stability of the manipulator's working process and the integrity of its basic actions.

For the joint trajectory tracking control problem of industrial robots, a novel predefined-time sliding mode trajectory tracking controller is designed.First, by analyzing the joint transmission mechanism of the industrial robot, the joint motion model driven by the permanent magnet synchronous motor is established.Then, a predefined time sliding mode trajectory tracking controller is designed, and the non-singularity and predefined time convergence of the controller are analyzed to ensure that the trajectory tracking error can be converged within a predefined time.Finally, experiments are carried out on industrial robots, and the results show that the designed controller has good trajectory tracking performance and can ensure that the tracking error converges within a predefined time.

Aiming at the issue of unknown friction terms and external disturbances exist in the actual trajectory tracking of the robotic manipulator, which tends to lead to insufficient tracking accuracy, and considering multivariate constraints and coupling, a robotic manipulator trajectory tracking predictive control scheme incorporating fuzzy compensation is proposed.Firstly, the inverse dynamics control is used for feedback linearization, and the model predictive control is used to deal with the multi-variable constraint problem and achieve online receding horizon optimization.Secondly, the fuzzy compensation is introduced to eliminate the effects of unknown friction terms and external disturbances.Finally, it is demonstrated by co-simulation experiments.The results show that the proposed control scheme has better dynamic and steady-state performance compared with only model predictive control.And the mean absolute error of the two-joint trajectory tracking is reduced by 88.63% and 88.30%, while the root mean square error is reduced by 96.84% and 97.85%, respectively.

In order to improve the control performance of permanent magnet synchronous motor (PMSM), a vector control system with ADRC and improved ESO is designed.First of all, the traditional ESO is improved by using smooth nonlinear functions to improve its ability of signal smoothing.The improved ESO is paralleled with ADRC to improve the accuracy of signal processing.Then, the electromagnetic torque and load error, speed error and time integral of the system are actively compensated to the current loop to improve the response speed and anti-interference of the system.Finally, the crow algorithm is used to optimize the parameters of ADRC to reduce the complexity of manual parameter tuning.Simulation results in Simulink show that the improved system has better robustness and stability than ADRC and PI control.

In order to solve the problem that when the dual three phase permanent magnet synchronous motor (DTP-PMSM) is affected by the external and internal factors, the electrical parameters will change and the accuracy of the sensor will decline.a sensorless control scheme with parameter identification is proposed.In this scheme, MRAS is used to realize resistance and inductance online identification.In order to reduce the influence of harmonics, it is extended to the static coordinate system.A new adaptive law is designed using Popov's hyperstability theory, and BP neural network is used to optimize the gain of the adaptive law, realize online adjustment of the gain, and improve the identification accuracy of the system.In addition, in order to further improve the robustness of the system, an improved phase-locked loop based on combining linear extended state observer with sliding mode control is proposed.The simulation results show that the proposed method can achieve accurate identification of electrical parameters and improve the system control performance.

Aiming at the problem that it is difficult for traditional speed control methods to satisfy the precise control in high precision applications when load torque changes irregularly and parameters change time, a data-driven intelligent control method based on maximum entropy deep reinforcement learning (SAC) was proposed.This method uses system data to train SAC agent to replace the speed loop in traditional PID control and realizes intelligent control by directly outputting voltage.Under the experimental conditions of variable load and variable speed, compared with the traditional PID and DDPG intelligent control methods, the average maximum speed error of the proposed method is improved by about 54.8% and 24%, and the average convergence time is improved by about 37.8% and 27.8%, respectively.Experimental results show that this method has good anti-jamming ability, tracking effect and stability.

Aiming at the problem of decreasing current response accuracy caused by model parameter mismatch and one-beat delay in permanent magnet synchronous motor (PMSM) system, a novel robust deadbeat predictive current control (NR-DPCC) method is proposed.Firstly, the predictive current control model of PMSM is established, and the sensitivity of electromagnetic parameters to conventional DPCC is analyzed in detail.Secondly, the inductance and flux linkage parameter identifier is designed by using Adaline neural network.On this basis, a variable step size neural network weight adjustment algorithm is proposed for motor systems.Finally, based on the parameters and current prediction values of online identification, the NR-DPCC method is proposed and verified by the RT-Lab hardware in the loop simulation experiment platform.The research results show that compared with the conventional DPCC method, the proposed method can not only accurately track the parameter changes of the motor online, but also effectively improve the current response accuracy.

For the trajectory tracking problem under the hysteresis nonlinear characteristics of shape memory alloys, a backstepping adaptive neural network control method was proposed.Firstly, based on the established Lyapunov function, a nonlinear controller for the SMA actuator is designed by a backstepping method.Then, adaptive neural networks are used to approximate the nonlinearity of the system and the uncertainty of the parameters, and the stability of the closed-loop system is proved based on the constructed Lyapunov function.Experimental results show that compared with the traditional PID controller, the proposed control method is successfully applied to the SMA actuator and almost compensates for the hysteresis.

Aiming at the problems of poor dynamic response speed and low synchronization control accuracy in the multi-motor speed coordination system, a new control strategy is proposed based on the deviation coupling control structure, which combines the super twisted nonsingular sliding mode controller with the error factor speed compensator.Combining the new nonsingular fast terminal sliding mode function with the super twist algorithm, the super twist nonsingular sliding mode controller is designed to shorten the response time of the system with its excellent dynamic regulation performance.The concept of error factor is introduced, and the speed compensation signal of each motor is redefined to enhance the coupling between the motors in the system.After the load disturbance, the speed fluctuation is small and the system synchronization error is reduced.The results turned out that the new method has strong tracking ability is improved by about 33%, and the synchronization is improved by about 40%.The new control strategy has better tracking and synchronization, and can realize the precise coordinated control of multiple motors.

Many problems exist for the current target detection of aluminum sheet surface defects, including the unsuitability of large-scale algorithms and computing devices in the field, and the balance between detection speed and accuracy.In this paper, a novel lightweight detection method based on attention mechanism is proposed, focusing on industrial applications of aluminum sheet defect detection.The GBANet backbone network is proposed based on the YOLOv4 framework, which is constructed based on a new convolutional Ghost module and embeds an improved attention module in the stacked Ghost blocks.The neck network is redesigned and lightened by feature fusion, the perceptual field is increased, the network is simplified by the SPPF-PANet module and the accuracy of the model for defective objects is enhanced by measures such as improved anchor box and loss function.Experiments show that the proposed method improves the mAP by 1.06% compared with the original YOLOv4, achieves a detection speed of 36.6 fps, reduces the model volume by 82.72%, and can effectively identify different kinds of defects on the surface of aluminum profiles.The proposed method can meet the requirements of defect detection in the production site of aluminum profile factories.

Aiming at the problems of complex system structure, difficult control accuracy and synchronization performance to achieve the desired control effect in multi-channel servo control system, a nine-channel actuator synchronization control system is designed.The system takes DSP and FPGA as the core control architecture, DSP is used for control algorithm, and FPGA is used for communication and external interface expansion; The system realizes distributed drive control, adopts bus structure design scheme, uses time triggered communication bus (TTP) to connect each independent actuator drive controller to realize data interconnection, and designs hardware and software in a modular manner.The experiment shows that the system control position sensitivity is 0.2 mm, the bandwidth is not less than 3 Hz, the motion speed of the actuator is not less than 100 mm/s, and the synchronization accuracy is 0.15 mm.The system has high control accuracy, good synchronization performance, and can meet the actual needs.

A fractional order dynamic surface control method based on radial basis function (RBF) neural network is proposed to solve the problem that the position tracking accuracy of permanent magnet linear synchronous motor (PMLSM) is affected by uncertain factors such as parameter variation and external disturbance.Based on the design of backstepping controller, the dynamic surface control technology is introduced to avoid the “differential explosion” problem of backstepping control.The fractional calculus is added to the virtual control variable to increase the degree of freedom of the system parameters.RBF neural network is used to approximate the uncertain factors of the compensation system.Based on Lyapunov stability theory, the stability of the controlled system is proved.The simulation results show that the proposed control method can reduce the tracking error, improve the robustness and accelerate the response speed of the system.

Aiming at the problem that microporous mechanical seals can effectively improve the performance of mechanical seals, this paper explores the change law of bearing capacity and leakage of composite micropores on mechanical seals, so as to reveal that composite micropores have better sealing performance.Firstly, the geometric model of the regular pentagonal-oval composite microporous fluid membrane was established.Secondly, the pressure distribution of the composite microporous texture on the flow field of the friction pair end face of the mechanical seal was studied by numerical simulation, and compared with the single micropores.Finally, the effects of different micropore depths h_p, rotation speeds n, area ratios S_p and pressure inlets P_i on the sealing performance were compared.The results show that when the micropore depth is about 3 μm, the leakage amount and bearing capacity of the composite micropores have reached the maximum value, and the open-drain ratio has achieved the minimum value.The bearing capacity and leakage capacity of composite micropores increased with the increase of speed and inlet pressure.The composite microporous mechanical seal has a large bearing capacity and a small leakage amount, and its open-drain ratio is better than that of a single microporous, and the composite microporous end face has good sealing performance.

The modeling of machine tool thermal error caused by environmental temperature variation is the basis of revealing the evolution law of thermal error and compensating for it.Taking a precision five-axis horizontal machining center in a common workshop as the research object, the characteristics of environmental temperature fluctuation in the workshop were analyzed, and the forming process of thermal error caused by environmental temperature variation was deduced.From the perspective of engineering practice, the experiments of machine tool thermal error with three environmental temperature measuring points were designed.A ridge regression thermal error model considering collinearity between variables was established, and the model parameters were solved by coordinate descent method.The results show that the prediction performance of the proposed model approximately improved by 33% and 22%, respectively, compared with the traditional single point and multi-point linear regression thermal error models.This study provides an effective reference model for later thermal error compensation.

In order to realize the lightweight of aluminum profile, the performance evaluation indexes are obtained by analyzing the bending and torsional resistance of aluminum profile, and the cross-sectional structure of aluminum profile is analyzed by finite element software to realize the cross-sectional structure improvement and design parameter selection; the parametric model of cross-sectional area and moment of inertia of aluminum profile is established and combined with AUTOCAD surface region command for auxiliary verification; The sequential quadratic programming algorithm is used to solve the optimal solution of the design variables in the parametric model, and the optimal design of the aluminum profile section is realized; under the condition that the moment of inertia of the aluminum section is not reduced, the area of the optimized section is reduced by 10% compared with the area before optimization.

In order to improve the wear resistance of the surface of GCr15 steel ball, GCr15 steel ball, hard ball, grinding powder and grinding liquid were mixed to enhance the grinding process of steel ball.The cross section microstructure, grain size and Vickers microhardness of GCr15 steel balls were detected by optical metallography microscope, X-ray diffractometer and Vickers microhardness tester, and then the friction and wear tests were carried out on the GCr15 steel balls with different ball intensification lapping time by friction and wear testing machine, and the average friction coefficient, wear amount and wear width were analyzed.The effect of enhanced lapping time on the surface wear resistance of GCr15 steel ball was studied.The results show that when the processing time increases from 30 min to 90 min, the thickness of the surface layer increases, the degree of grain refinement gradually increases, the Vickers microhardness increases, and the wear resistance of the ball is enhanced.

Based on the equivalent plane method, some scholars use the results of two-dimensional orthogonal cutting simulation to replace the stress field and temperature field of three-dimensional oblique cutting simulation, in order to save the simulation time and cost.However, However, at present, the application of the equivalent plane method is not perfect, and there are certain errors.In this paper, Based on the analysis of the causes of error, this paper optimizes the equivalent plane method in the temperature field and applies it to the milling simulation.Firstly, the conversion model between the tool parameters in the equivalent plane and those in the main section is established.At the same time, the important parameters of the workpiece are also treated equivalently, such as cutting thickness, matrix thickness, specific heat and friction coefficient.Finally, by establishing simulation models for comparative analysis, it is found that the two-dimensional cutting model established based on the improved equivalent plane method effectively reduces the error between the simulation results of the two-dimensional model and the three-dimensional model, in which the maximum temperature error of the contact area of tool and workpiece is reduced from 9.63% to 0.29%, and the temperature distribution of the workpiece substrate is also significantly improved.So the improved equivalent plane method can effectively reduce the simulation error.

A new hexapod wheeled-legged robot and its control system are designed to meet the requirements of mobile welding of structured box girder.First, the body structure model of the robot is designed.Then, the kinematics analysis and modeling are carried out for the robot model, and the robot's obstacle crossing gait is planned.In order to ensure the stability of the robot in the process of linear motion, the fuzzy PID control algorithm is used, and the Simulink software are used for simulation.Finally, an experimental platform is built for experimental testing.The experimental results show that the linear motion deviation of the wheel-legged robot based on fuzzy PID control method is smaller than that of the conventional PID control method, and it can achieve stable obstacle crossing motion.

In order to solve the problems of high prior knowledge requirement and insufficient information utilization in time domain and frequency domain in feature extraction of single vibration acceleration signal in most mechanical fault diagnosis, a sparse self-coding fault diagnosis method based on information fusion was proposed.Firstly, the velocity and displacement signals are obtained by frequency domain integration of vibration acceleration signal, and the frequency spectrum of acceleration signal is calculated.Secondly, the spectrum, velocity and displacement signals are fused into a composite signal.Finally, the composite signal is used as the input of sparse auto-coding network for depth feature extraction, and SoftMax classifier is used for state recognition.The model was adjusted by adjusting the input information of different proportions, and compared with the traditional sparse auto-coding fault diagnosis model, the results show that the proposed method can effectively identify the rolling bearing fault and RV planetary wheel fault, and can improve the recognition accuracy while reducing the number of network layers.

By analyzing the fault characteristics and generation mechanism of spindle system, a fault feature extraction method based on variational modal decomposition (VMD) and multi-scale weighted permutation entropy (MWPE) and a failure analysis and diagnosis model of CNC machine tool spindle system based on particle swarm (PSO) optimization support vector machine (SVM) are proposed.Firstly, the variational mode decomposition method is used to decompose the collected vibration signal of the spindle system, and several effective eigenmode components (IMFs) are obtained.Secondly, the fault feature information is extracted by multi-scale weighted permutation entropy, and the SVM model is used to classify and identify the fault feature information.In order to improve the recognition accuracy of the model, the particle swarm optimization algorithm (PSO) is introduced to optimize the SVM model parameters.Experimental verification shows that the proposed signal feature extraction method and state recognition model have achieved good results in the fault diagnosis of CNC machine tool spindle system, and the fault recognition accuracy is as high as 99.56%.

Aiming at the problems of low accuracy and slow speed of high similarity casting detection in current mixed-flow casting sorting line, an efficient detection method for lightweight multi-casting based on SSD was proposed.Firstly, the backbone network of SSD model is replaced by MobileNetv2 which integrates lightweight ECA attention mechanism to solve the problem of precision loss due to lightweight design.Secondly, the reverse residual layer is used to replace the convolutional prediction layer of the original SSD, so as to further reduce the weight of the algorithm, and solve the problems of the SSD algorithm due to the complex network structure, large number of parameters and floating point calculation.The experimental results show that mAP reaches 97.10% and the detection speed is 46.84 frames/s for multi-casting detection by the proposed algorithm.Compared with SSD algorithm, GFLOPs is reduced by 41 times and the parameter number is reduced to 4.34 MB, which reduces the demand on hardware devices while ensuring accuracy and speed.

In order to reduce the cogging torque of the magnetic gear, weaken the torque fluctuation during work, so that the magnetic gear has higher transmission efficiency and is used in more occasions.In this paper, the formation process and waveform variation trend of cogging torque of magnetic gear are analyzed, and a side sinusoidal pole plate structure is proposed.Firstly, the structural parameters in the side sinusoidal pole piece that can affect the cogging torque are set.Secondly, the specific trend of the influence of the structural parameters of the set magnetic pole plate on the cogging torque is simulated and analyzed, and the curve of the influence of the grooving rate, radial thickness and sinusoidal curve amplitude on the cogging torque of the magnetic gear is obtained, and the traditional sector and the side sinusoidal pole are compared.According to the best parameters obtained by simulation, the prototype was made, and it was verified by experiments that under a certain input speed and load, the input torque fluctuation of the sinusoidal pole piece on the side of the magnetic gear application was reduced by 14.67% compared with the sector pole piece, and the output torque fluctuation was reduced by 4.80%.The experimental results show that the side sinusoidal magnetic pole piece can effectively reduce the cogging torque, weaken the torque fluctuation, and improve the transmission efficiency and service life of the magnetic gear.

Aiming at the eccentricity of permanent magnet rotor in surface magnetic detection, based on the molecular circulation model, the spatial magnetic field distribution of permanent magnet is analyzed by Biot-Savart theorem, and a system for surface magnetic eccentricity correction is developed.Firstly, the principle and workflow of the eccentricity correction method are introduced, and the bi-directional interpolation based on cubic spline is introduced for the data processing module.Then, the correction system platform is designed and built.Finally, the performance of the system is evaluated by collecting comparative experimental data.The experiment shows that the molecular circulation model has strong applicability to analyze the spatial magnetic field distribution of permanent magnet rotor.At the same time, cubic spline bidirectional interpolation makes the measured magnetic field data more consistent with the real value; Finally, the relative error between the corrected magnetic field data and the real value can be controlled within 0.9%.

Aiming at the problem of vertical stiffness calculation of support base, and focusing on how to improve the accuracy of finite element calculation, this paper proposes an analysis method in which the equivalent stiffness model replaces the macro structure model.Based on the finite element method, the micro rough surface simulation and analysis are carried out, and the equivalent stiffness is calculated using the integration idea.The topography data of the real workpiece surface is obtained through the contour scanner, and the micro rough surface conforming to the Gaussian distribution is reconstructed.The contact stiffness is simulated by using ANSYS software.The results are compared with different theoretical contact models, and the rationality of the finite element micro contact surface stiffness is verified.Comparing the calculated stiffness of the macro structure model with that of the equivalent stiffness model, the results show that the plastic deformation of the contact surface and the matrix deformation of the curved surface have the greatest impact on the stiffness, which provides a basis for the subsequent structural optimization of the support system.

Alignment is the problem of estimating the rigid body transformation between two point clouds, and numerous deep learning methods have been proposed to solve this problem in recent years.However, the training sets of these methods often come from time-consuming and insufficient amount of data for real acquisition and manual calibration.This problem is expected to be solved by rendering techniques.A point cloud alignment algorithm process based on Point Transformer deep learning point cloud feature extraction network is designed for 6D pose estimation application scenario.The rendering technique is applied to generate a large number of local projections of the ShapeNet dataset model in different viewpoints to construct the training set.Using InfoNCE as the loss function, the Point Transformer network is made to learn the point-by-point feature description of point clouds with a contrast learning approach.The Rendering-ICP algorithm is developed to optimize the fine collocation link as the final bit-pose estimation result.The present algorithmic procedure is validated on the Linemod dataset, using VSD regression as an evaluation metric, and a 7% performance improvement is obtained compared to the extant method.

On the basis of traditional workshop management, the real 3D visualization effect of physical workshop in virtual space is expanded, and a new digital twin-driven workshop management system architecture is proposed.The MesWork Data Factory digital twin virtual simulation platform is used to model the twin workshop information, and OPC UA data acquisition network architecture is used to perceive the production data of the equipment.The front-end program receives the real-time signal of the equipment movement, and drives the twin model movement according to the technological process, so as to realize the digital twin driven workshop system which is synchronized running and interactive characteristics of reality-virtual.The system is applied to a production workshop to realize a new operating mode of a workshop with digital twin, and the effectiveness and practicability of the system are verified.

For the vehicle routing scheduling problem of automotive painting production, in order to reduce the number of color switching of painting operations and subsequent production sequence deviation, and to achieve the lowest manufacturing cost within limited resources.Therefore, based on the characteristics of having multiple linear buffer storage areas in the paint shop, the random set of vehicles to be painted as input, the MILP (mixed integer linear programming) model was developed with the goal of minimizing the number of color switching and the deviation of the total assembly production demand queue, with inbound based on heuristic rules and outbound solved by an improved genetic algorithm, vehicle routing and scheduling schemes were exported.Finally, a routing scheduling system is developed for a new energy vehicle plant paint shop as an example, and the joint scheduling method with multiple linear buffers proposed in this paper is verified, resulting in about 80% reduction in painting switching cost and 10% reduction in total assembly production demand deviation cost.

In order to solve the problem that the enforceability of the initial scheduling scheme is reduced due to uncertain machine failure in the production process, a flexible job shop rescheduling method considering machine failure probability is proposed.The mixed driving strategy of event and cycle based on failure probability is adopted, a variety of rescheduling modes are used comprehensively, and the maximum completion time deviation, process start time cumulative deviation and process machine change are introduced as evaluation indexes.the scheduling scheme is solved by using the improved genetic algorithm combined with variable neighborhood search.Through the simulation analysis of the processed flexible job shop, the results show that compared with the single rescheduling method, the optimal rescheduling scheme has a great improvement in the performance evaluation index.the superiority of the proposed rescheduling method to deal with machine fault disturbance under different conditions is verified.