Stationary subspaces-vector autoregressive with exogenous terms methodology for degradation trend estimation of rolling and slewing bearings

Degradation trend estimation (DTE) of rotating machinery plays a vital role in prognostics and health management (PHM). It enables us to foresee future conditions and avoid unexpected risks. Recently, considerable accomplishments in the field of rotating machinery PHM has achieved through regression analysis based data-driven prognostics, which assist in directly analyzing and exploring the relationships between degradation trend and characterization indicators. Internal static structures still widely exist in most of them, inevitably restricting the natural extrapolation or generalization to future moments. Thus, the autoregression theories with complete mathematical foundations are first introduced and extended the methodologies for rotating machinery DTE. Meanwhile, the characterization ability of degradation or damage information from a single indicator rather than multi-endogenous indicators considering their causality and interactions may significantly reduce in the existing regression analysis based prognostics, and it further influences the final prognostics. Therefore, the idea of exploring internal dynamic structural regression based prognostics containing establishing multi-endogenous degradation indicators with weak-stationary traits and an interpretable and lightweight vector autoregression based DTE modeling method is motivated. The above dilemmas are well addressed through the in-depth study of autoregression based prognostics, namely stationary subspaces-vector autoregressive with exogenous terms (SSVARX). To be specific, multi-channel vibration signals are first picked up, and non-stationary signals are converted into time and frequency domain based weak-stationary degradation indicators via double stationary subspace decomposition and differential operation. Then the above two domain endogenous variables are feed into our proposed DTE models after stationarity test, order determination, and impulse response analysis. Finally, promising results from two run-to-failed life tests of rolling and slewing bearings are obtained via our multi-endogenous variables based extrapolation model. Compared with existing prediction methodologies, SSVARX of this study achieves not only high-accurate prediction results but also fast-computing speed and reasonable mathematical supports. (c) 2020 Elsevier Ltd. All rights reserved.

Automated summary

The study addresses the dilemma of predicting degradation trends in rotating machinery through vector autoregression-based DTE, achieving successful results and overcoming limitations in existing methodologies.