Adaptive Remaining Useful Life Prediction Algorithm for Bearings

It is well known that the lifetime of bearings varies a lot even if the bearings are manufactured in the same assembly line. This is because the exposed operating conditions may be very different for each bearing. As a result, an adaptive remaining lifetime prediction algorithm is a more natural choice and is expected to be more accurate than the non-adaptive counterparts. This paper presents a new adaptive approach to predicting remaining useful life (RUL) of bearings. This approach improves an earlier method developed by us. In our earlier approach, it was observed that the prediction estimates depend on the initial conditions of the adaptive estimator. In the new approach, a bank of predictors is run in parallel and their predictions are constantly evaluated. Only the good estimators that meet the criterions are used for RUL output. Evaluations using actual bearing data demonstrated the efficacy of the proposed algorithm.