Study on Screw Pairs Life Prediction Based on Hidden Markov Model and Neural Networks

Abstract

In order to estimate and predict the screw performance in the process of machining, a screw life monitoring system was built. Current signal was processed and features sensitive to cutting force were selected, a virtual force sensor was constructed to model the relation between cutting force and current by BPNN. Cutting force was indirectly calculated by the model, so the rating life of screw in different condition could be known and residual life also be reckoned by historical database. A three-way vibration sensor was installed on screw pair base; screw condition could be induced by HMM which input was 15 vibration signal features. As machining condition changed, corresponding new HMM would be built by adaptive method. Finally, the residual life of screw could be gotten by multi-HMM and BPNN. The experimental results show the model proposed in the paper is effective and high precision.