Author:

Keywords:

prognostics, logistic regression

Abstract:

Despite critical components, very little attention has been paid for wet friction clutches in the monitoring and prognostics research field. This paper presents and discusses an overall methodology for assessing the health (performance) and predicting the remaining useful life (RUL) of wet friction clutches. Three principle features extracted from relative velocity signal measured between the input and output shaft of the clutch, namely (i) the normalized engagement duration, (ii) the normalized Euclidean distance and (iii) the Spectral Angle Mapper (SAM) distance are fused with a logistic regression technique into a single value called the health index. In logistic regression analysis, the output of the logistic model (i.e. the health index) is restricted between 0 and 1. Accordingly, the logistic model can guide the users to assess the state of a wet friction clutch either in healthy state (e.g. health index value of (close to) 1) or in failed state (e.g. health index value of (close to) 0). In terms of prognostics, the logarithm of the odds-of-success g defined as g = log(h/[1-h]), where h denotes the health index, is used as the predicted variable. Once history data are sufficient for prediction, the weighted mean slope (WMS) method is implemented in this study to adaptively build a prognostics model and to predict the trajectory of g until it crosses a predetermined threshold. This way, the remaining useful life (RUL) of a clutch can be determined. Furthermore, an experimental verification of the proposed methodology has been performed on two history datasets obtained by performing accelerated life tests (ALTs) on two clutch packs with different friction materials but the same lubricant. The experimental results confirm that the proposed methodology is promising and has a potential to be implemented for real-life applications. As was expected, the estimated RUL converges to the actual RUL and the uncertainty interval decreases over time that may indicate that the prognostics model improves as more evidence becomes available.