Prediction And Compensation Of Contour Error Of CNC System Based On LSTM Neural Network

Contour error control is an essential method to improve machining accuracy.The current commonly used contour error control methods mainly include cross-coupling contour error control and iterative learning contour error control.These two methods are both passive contour error control approaches,which compensate on the produced contour error.Therefore,it is significant to realize the active contour error control and suppress the contour error generation.Aiming at the task-changing motion control system,this thesis proposes a method of contour error prediction and compensation based on the deep LSTM neural network.Based on the deep LSTM neural network,each axis’ s tracking error prediction models are established to estimate and predict the contour error.According to the predicted contour error,the reference trajectory is modified to realize the active contour error control.The deep LSTM neural network’s features contain not only linear characteristics that are selected based on the simplified modeling,but also nonlinear characteristics determined by the analysis of the operating data.Considering both the linear characteristics and nonlinear characteristics,the effective prediction of single-axis tracking error is achieved.Based on the linear interpolation approximation contour error estimation algorithm and the spline approximation contour error estimation algorithm,a two-directions backtracking piecewise interpolation approximation contour error estimation method is proposed,the main idea of which is to determine the contour approximation interpolation method according to the local curvature of the actual position curve.Combining the advantages of linear interpolation approximation and spline approximation,the proposed method improves not only the error tolerance rate but also the efficiency and accuracy of contour error estimation.In actual applications,for the reference position data to be executed after trajectory planning and interpolation,the deep LSTM neural network tracking error prediction models are used to predict each axis’ s tracking error.And the proposed contour error estimation method is used to estimate the contour error,the components of which are compensated to each axis’ s original reference position to generate a new reference trajectory that will be sent to the controller FIFO buffer to perform the position control late.Thereby the on-machine prediction and active contour error control are realized.Finally,on a three-axis CNC machine tool,the contour control effect of the active contour error control scheme was verified.The contour error of the experimental trajectory was effectively suppressed.