Research On Modeling And Control System Driven By Data And Image For Relining Manipulator Of Mill

When the large ball mill is working,the liner is worn by collision and friction between the liner and the material.Therefore,the liner must be replaced regularly.The traditional method of relining requires manual handling,disassembly and installation by workers,which has high labor intensity and unsafety.When installing the liner,it uses slings to assist the installation due to the large mass of the liner.During installation,it is necessary to constantly adjust the position of slings,which seriously affects the installation efficiency.To improve the relining efficiency,the main contents of research on modeling and control system driven by data and image for relining manipulator of mill are as follows:The mechanical structure of the relining manipulator of mill,the end fixture and the liner handling fixture are designed.Static finite element analysis of the liner fixture is carried out by Solid Works Simulation.Based on the structure of relining manipulator of mill,the kinematics model and dynamics model are established by D-H method and Lagrange method,respectively.A visual inspection system for liner is established.The parameters of the binocular camera are obtained by Zhang’s camera calibration method,and the binocular camera is used to measure the distance from the camera to the liner.YOLOv4 is used to classify the two kinds of liner and extract the coordinates of liner bolt holes.The experimental results show that the method can obtain the actual coordinates of the center point of the liner bolt holes and the distance between the camera and the liner.A recurrent stochastic configuration network(RSCN)is proposed,which is used to train the uncertain parameter prediction model of the relining manipulator of mill.The parallel configuration method(PCM),stochastic configuration networks with chaotic maps(SCNCM-I and SCNCM-II)are proposed.Based on stochastic configuration networks with chaotic maps and Harris Hawk Optimization Algorithm(HHO),a hierarchical learning strategy(HLS)is established.PCM uses uniform and normal distributions to randomly assign input weights and biases of hidden layer nodes and designs an iterative learning algorithm to improve configuration efficiency.SCNCM-I uses multiple errors and chaotic maps to adjust the nodes block size to improve the training speed.On the basis of SCNCM-I,SCNCM-II uses the node removal mechanism to reduce the complexity of the model structure.HLS is the integration of SCNCM-I,SCNCM-II and HHO to improve the training speed and reduce the complexity of the model structure.The feedback channel is added to HLS,which gets the recurrent stochastic configuration network.Experiments on an approximation function and benchmark data sets show the effectiveness of the methods.The recursive terminal sliding mode control method(RTSMC)based on RSCN disturbance observer is designed.A nonlinear function is established for the design of terminal sliding surface.According to the dynamic model of the relining manipulator of mill,a recursive terminal sliding mode controller is designed.An adaptive law is added to the uncertain parameter prediction model,which establishes a disturbance observer to compensate the parameter uncertainty and external disturbance of the model.A dsat function is constructed to realize the adaptive adjustment of switching gain.Simulation results show that the method has great trajectory tracking ability and anti-jamming ability.