Machinery Health Prognostics With Multimodel Fusion Degradation Modeling

Degradation modeling aims to formulate the health state degradation process of machinery. Commonly used degradation models pay more attention to describing the global increasing or decreasing trend without considering the local fluctuation in the degradation process. To deal with the above-mentioned issue, this article proposes a multimodel fusion degradation modeling method. The basic idea is to fuse multiple models to describe various degradation trends of machinery involving the global trend as well as the local fluctuation. A generalized statistical degradation modeling framework is constructed, wherein the degradation process is formulated by fusing multiple models with various degradation trends. The failure event is reinterpreted under the condition of state observations fluctuating around the failure threshold. The probability density functions of the time when the state observation exceeds and drops below the failure threshold are derived, respectively. An iterative matching pursuit algorithm is developed to select the optimal models adaptively. A numerical illustration and an experimental study are conducted to verify the proposed method. The results demonstrate its superiority in health prognostics compared with two benchmark methods in cases where the degradation process has dominant local fluctuation.