Fault prognosis feature extraction and selection for bearings based on statistical indicator optimization

Fault evolution feature extraction is the key to tracking and monitoring fault development, and triggering prognosis and health management. However, traditional time-frequency domain methods didn't consider the effectiveness of features for the whole lifecycle process, and it supports a low degree for fault prognosis and health state assessment. A fault prognosis feature extraction method based on statistical indicator optimization selection is proposed in this paper. Firstly, original signals of bearings in the whole lifecycle are collected, then their statistical indicators at each echo are calculated, thus those fault evolution trend curves described by all statistical indicators will be generated. Secondly, fault trend analysis is implemented based on the above fault evolution curves, then the monotonicity, trendability, identifiability and robustness of all features for fault evolution curves can be calculated. Thirdly, the monotonicity, trendability, identifiability and robustness of each feature are translated into a hybrid metric, and those hybrid metrics of all features are sorted by size, thus the fault prognosis features will be selected appropriately. Finally, the life-time data of bearings provided by University of Cincinnati are used to verify the effectiveness of the proposed method in this paper.