RESIDUAL LIFE PREDICTIONS FOR BALL BEARING
BASED ON NEURAL NETWORKS

XI Lifeng¹  HUANG Runqing¹  LI Xinglin²  LIU C Richard³  LEE Jay ⁴

(1. School of Mechanical Engineering, Shanghai Jiaotong University, Shanghai 200030;
.2. Hangzhou Bearing Test & Research Center, Hangzhou 310022;
3. School of Industrial Engineering, Purdue University, West Lafayette 47907, USA;
4. School of Engineering, University of Cincinnati, Cincinnati 45221, USA)

Abstract: A new scheme for prediction of ball bearing’s remaining useful life is dealt with based on self-organizing map and back propagation neural networks. One of the key issues in bearing life prediction is to set up an appropriate degradation indicator from its incipient defect stage to final failure. Different from degradation features ever used, it uses the minimum quantization error (MQE) indicator deriving from SOM, which is trained by six vibrations features including a new designed degradation index for performance degradation assessment. Then using this indicator, back propagation neural networks focusing on the degradation periods are trained. Based on weight application to failure times (WAFT) technology, a remaining useful life prediction model of ball bearing is developed successfully. The validation results show that the proposed methods are greatly superior to the currently used L10 bearing life prediction.

Key words: Self-organizing map  Neural network Ball bearing  Prediction model  Residual life

CLC No: TH 133.1

国家自然科学基金资助项目(50128504，50405016). Received 20061020, received in revised form 20070518