Artificial neural network-based thermal error modelling in ball screw

In the pursuit of high precision in a modern CNC machining system, it is significant to eliminate thermal error. The paper first makes a brief outline of the characteristics and the training methods of neural networks. And it is successful to apply the neural network model to model the thermal error in the CNC machine linear feed system. The expected results is achieved that the maximum prediction error reduced to 2 um, laying a further foundation for thermal error compensation. The text describes the actual modeling process in detail. And a new method of data preprocessing is come up with according to the specific characteristics of training data. It is an innovative point of this paper to apply the method to model training better.