Adaptive weighted fault growth parameters: New statistic parameter health indicators for machine performance degradation assessment

Abstract

In order to monitor machine condition and assess performance degradation, the most crucial step is to construct a suitable health indicator (HI). Although significant outcomes about HI construction for machine performance degradation assessment (MPDA) during recent years, the design of HIs for simultaneously achieving clear incipient fault detection, monotonic and stable degradation trends, and similar scale ranges has not been fully studied. To overcome this problem, new statistic parameter HIs, coined as adaptive weighted fault growth parameters (AWFGPs), are proposed in this article. Firstly, generalized fault growth parameters (GFGPs) are developed from the traditional FGP by extending the detection threshold from triple standard deviations to p-quantiles. Subsequently, inspired by a revised FGP named FGP1, the AWFGPs are constructed based on the GFGP with p=0.5, where a≥1 is the only parameter. The proposed AWFGPs enhance the fault sensitivity in the late degradation stage by assigning fixed and time-varying weights to signal points below and above the detection threshold, respectively. Mathematical derivation on a complex Gaussian signal reveals that the theoretical baselines of the GFGPs and AWFGPs for incipient fault detection are 1-p and 3/(3+a), respectively. Numerical and Experimental investigations are conducted to verify the effectiveness of the proposed GFGPs and AWFGPs. Investigation results demonstrate that the proposed AWFGP with a=10 has better comprehensive performance for MPDA compared with other classical HIs.