Component based data-driven prognostics for complex systems: Methodology and applications

Abstract—In recent years, considerable research efforts have been applied in the field of fault prognostics. However, to the authors knowledge, there are few published works that address complete and systematic methods describing the steps required to develop data-driven prognostics approaches for complex systems. This paper presents a generic component-based prognostics methodology that can be customized for different applications and which can be useful for new researchers and engineers. The paper is divided into two parts. The first part provides a description of the procedures required before constructing data-driven prognostics, such as identifying critical components, selecting physical parameters to monitor, choosing monitoring sensors and defining the data acquisition system. The second part presents a novel data-driven prognostic method for direct remaining useful life (RUL) prediction. This method relies on two phases: offline and online. In the offline phase, a method for constructing health indicators (HI) from sensor data is presented. Such HIs can be used as offline models to display the deterioration evolution of components over time. In the online phase, similar HIs are constructed from the sensor data for a new component. Then, a discrete Bayesian filter is applied to estimate the current health status. Finally, the offline database is searched to find the closest group to the online HIs. The selected offline HIs can be used for estimating the RUL of the new component under operation. The performance of the method is demonstrated using two real data sets taken from the NASA Ames prognostics data repository. Index Terms—Data-driven prognostics, remaining useful life, health indicators construction, discrete Bayes filter, Gaussian process regression. I