Selection of bearing health indicator by GRA for ANFIS-based forecasting of remaining useful life

Abstract

Within the framework of Prognostic and Health Management, an attempt has been made to enhance prediction accuracy of the bearing remaining useful life (RUL) using the adaptive-network-based fuzzy inference system. First, the gray relational analysis has been employed to select the optimal bearing health indicators from 30 vibration features of time, frequency, and time–frequency domains. The selection was based on four criteria, namely correlation, monotonicity, average increase rate, and robustness. This allowed ranking the candidate features in terms of sensitivity to the bearing degradation evolution. The effectiveness of the proposed approach was evaluated on bearing run-to-failure test carried out on a wind turbine. The results of RUL prediction, based on the average of the present and the two previous values of the optimal feature, were very satisfactory, which can effectively optimize the maintenance costs of such equipment.