The Study Of Following Error Prediction Method For Machine Tool Feed Drive System Using Running Data

In order to predict the following error of the machine tool feed drive system,it is necessary to construct a dynamic response model of the feed drive system.The traditional mathematical model and Simulink simulation method,electromechanical simulation method;there is a problem of insufficient expression of nonlinear factors in the feed drive system and inaccurate parameters in the mathematical model.In this paper,the following error offline prediction model is constructed based on the running data of the machine tool feed drive system.It realizes the offline prediction of the following error.For machine tool feed drive system,the repetitive nature of dynamic response is the basis for modeling.In the same machining instructions and parameter configuration,the actual trajectory of the workbench should have repeatability.By analyzing the composition and control principle of the feed drive system,the proportional gain value of the feed drive system position loop is solved,the following error is decomposed into a steady-state following error proportional to the command speed,and the non-steady-state following including the nonlinear response error error.The steady-state following error value can be obtained by the command speed.The nonlinear characteristic of the non-steady-state following error is strong.It is necessary to construct a neural network model to achieve a nonlinear mapping between the command data and the non-steady-state following error.Taking the straight line,the full circle and the spiral track as examples,the running data is preprocessed,the characteristic relationship between the running data is analyzed,the motion state interval is divided and marked,and the input and output of the neural network are determined.The long-term and short-term memory network(LSTM)suitable for time series problem modeling is selected to construct the non-steady-state following error prediction model of the feed drive system,and the multi-layer perceptron network(MLP)is used as the comparison model.The input and output of the model have been analyzed and determined,and the structure and hyperparameter of the model are adjusted to have a good prediction accuracy in the test set.The experimental results show that the constructed LSTM model can have a good tracking error prediction accuracy under different processing trajectories(such as straight line,full circle,spiral,free curve)and different feedrates.The prediction error of the LSTM model is smaller than the MLP model.It is shown that the dynamic response model of the feed drive system can be effectively constructed based on the running data of the machine tool feed drive system,and the accurate prediction of the following error can be realized.