Health State Prediction of Rolling Element Bearing Using Phase Space Reconstruction and Improved GMDH

摘要

Rolling element bearings are not only one of the most common parts in rotating machinery, but also one of the most frequent reasons for unplanned shutdowns of the mechanical system. The health state of rotating bearings is therefore significant to the safety in production, which involves the techniques for feature extraction and health state prediction of rolling element bearings. Various statistical indicators of vibration signals are used as time-domain features to describe the health conditions of machines, whereas traditional features are probably not sensitive to the early failure of rolling element bearings. In this paper, a novel method based on the phase space reconstruction and improved GMDH (Group Method of Data Handling) network is proposed to predict the health degradation of bearings. Its effectiveness is finally investigated on real vibration signals, and the result shows that the prediction accuracy of the proposed method is superior to that of the grey prediction model and the LSTM (Long Short-Term Memory) model.