A Soft Sensor for Prediction of Temperature Rises on a Ball Screw Shaft Using Extreme Learning Machine

Abstract

A high speed ball screw driving system generates considerable heat and causes significant thermal expansion, which affects the accuracy of position. In order to properly deal with thermal errors and resultant thermal deformation, continuous precision thermal monitoring along the ball screw shaft is required. However, the working temperature of the ball screw shaft is difficult to measure online. Therefore, in this paper we propose a soft sensor based on an extreme learning machine (ELM) method for predicting the distributions of temperature rises on the ball screw shaft of a feed drive. ELM is an emerging learning algorithm for single-hidden layer feedforward neural networks (SLFNs) training that has recently attracted many researchers' interest due to its impressive generalisation performance at fast training speed. The performance of the ELM soft-sensor is investigated on the thermal expansion process of the ball screw system simulated by the finite element method (FEM). The results show that the ELM-based soft sensor provides good generalisation performance with much faster speed than the traditional backpropagation artificial neural network (BP-ANN).