Entropy-based domain adaption strategy for predicting remaining useful life of rolling element bearing

A machine's remaining useful life (RUL) is a challenging issue that indicates how much useful life is left. In the past, models based on statistical characteristics taken from signals collected by attaching sensors to a machine have been created to estimate how much longer a machine will be functional. This works fine when the machine is subjected to the same operating conditions. However, in the real-time scenario, the machine is operated under different conditions. As a result, the model developed under one condition does not work satisfactorily for another condition. In this work, we have explored various features for the prognosis and estimation of the RUL of the rolling element bearing. A health indicator has been developed by applying Isometric Mapping (ISOMAP) to the tangent entropy features extracted from the vibration data's 16-bandpass filtered frequency domain signal. The pro-offered health indicator exhibits the monotonic degradation trend and can be utilized to develop the data-driven model for determining the RUL of the bearing. For the effective working of the model under different operating conditions, feature-based and instant-based domain adaptation is applied so that the model can be reliably used under cross-domain operating conditions. After applying domain adaptation, it has been found that the model can give the same level of performance as it gives when whole data of different conditions are used for developing the model.