Early Detection of Bearing Faults by the Hilbert-Huang Transform

The operation of bearings usually results in a dynamic behavior generating stationary and non-stationary vibration signals mixed with an amount of background noise. Therefore, the condition monitoring of bearings becomes difficult since the purpose is to extract health indicators able to detect the appearance of faults, track their evolution and predict the bearings' remaining useful life. The aim of this paper is the introduction of a new approach for the health monitoring of bearings. This approach is based on health indicators extracted from raw vibration signals filtered by the Hilbert-Huang transform. The proposed approach is composed of three steps. The first step uses the empirical mode decomposition to separate each vibration signal into different intrinsic mode functions (IMFs), where each IMF is located within a specific frequency band. The second step extracts instantaneous amplitudes and frequencies for each mode in order to identify its frequency band. Finally, the third step selects the interesting IMFs according to the characteristic frequencies of the bearing failures. The Hilbert marginal spectrum of the selected intrinsic mode functions are then considered as health indicators. The proposed approach is validated by real data taken from the PRONOSTIA experimental platform.