Experimental and data driven measurement of engine dynamometer bearing lifespan using acoustic emission

Abstract

In this study, experimental studies are carried out to capture the lifespan of engine dynamometer bearing using acoustic emission signal. As there is not a through comparison of the usage of different novel neural networks for bearing remaining useful life prediction, here, different feedback neural networks namely Elman, Jordan and LSTM neural networks are applied for prediction of bearing lifespan based on acoustic emission waveforms. Different time–frequency features are presented and prognostic feature selection is used for features size reduction. Correlation-based feature selection (CFS) is used to identify the correlated features. Comparison of different neural networks showed that Jordan feedback neural network with Bayesian Regularization training algorithm has the least error in estimating the bearing lifespan and is therefore selected as the best neural network for this purpose. Also, it is shown that acoustic emission is a good method for bearing RUL prediction. The time to start prediction for the the definition of RUL based on the different tests was calculated about 13.4 h.