State prediction of spindle and feed axis status of CNC machine tools based on LSTM

The health status of the spindle and feed axis has a direct impact on the reliability and accuracy retention level of the machine tool. Vibration acceleration data of the spindle and feed axis is collected by installing three-axis vibration acceleration sensors on the GZ002 model machine, which is processed to calculate two feature values, namely "surface quality-related features" and "dimension-related features". After normalizing and differential transformation on the feature data, the training data and label data are obtained by the time window, which are input into the LSTM model for learning and predicting the future 10 states of the CNC machine tool. Experimental results indicate that the MAE values of the predicted features in each axis and direction are all less than 0.1, and the MSE values are all less than 0.015, demonstrating the accurate prediction of the future states of the spindle and feed axis.