Canonical correlation analysis of dimension reduced degradation feature space for machinery condition monitoring

Health indicator (HI) is an important metric to characterize a degradation process and detect the fault initiation time of rotating machineries as early as possible for condition monitoring. In recent years, many HIs are developed and reported, however, a constructed HI which is more sensitive to incipient defects, free of massive history degradation data and is able to quantitatively measure a degradation process is still remain to be studied. To address this issue, a novel HI constructed through canonical correlation analysis of dimension reduced degradation feature space is proposed. In the proposed method, the intrinsic low-dimensional degradation feature space is mined by a dimension reduction model based on auto-encoder from a high-dimensional statistical feature matrix. After that, the degradation state is measured by calculating the ca-nonical correlation from the feature space of the baseline and the subsequent collected moni-toring data. On this basis, a new HI is formed to quantitatively characterize a degradation process for condition monitoring. The performance of the proposed HI is investigated by comparing with some typical state-of-the-art HIs, such as L2/L1 norm, kurtosis, negative entropy, Gini index, smoothness index, Hoyer measure, etc., the experimental validation results demonstrate the proposed HI is able to detect incipient fault and is more sensitive to the early stage degradation process.