Data-driven prognostics based on health indicator construction: Application to PRONOSTIA's data

Abstract

Failure prognostics can help improving the availability and reliability of industrial systems while reducing their maintenance cost. The main purpose of failure prognostics is the anticipation of the time of a failure by estimating the Remaining Useful Life (RUL). In this case, the fault is not undergone and the estimated RUL can be used to take appropriate decisions depending on the future exploitation of the industrial system. This paper presents a data-driven prognostic method based on the utilization of signal processing techniques and regression models. The method is applied on accelerated degradations of bearings performed under the experimental platform called PRONOSTIA. The purpose of the proposed method is to generate a health indicator, which will be used to calculate the RUL. Two acceleration sensors are used on PRONOSTIA platform to monitor the degradation evolution of the tested bearings. The vibration signals related to the degraded bearings are then compared to a nominal vibration signal of a non-degraded bearing (nominal bearing). The comparison between the signals is done by calculating a correlation coefficient (which is considered as the health indicator). The values of the calculated correlation coefficient are then fitted to a regression model which is used to estimate the RUL.