Developing machine learning-based models to estimate time to failure for PHM

Abstract

The core of PHM (Prognostic and Health Monitoring) technology is prognostics which is able to estimate time to failure (TTF) for the monitored components or systems using the built-in predictive models. However the development of predictive models for TTF estimation remains a challenge. To address this issue, we proposed to develop machine learning-based models for TTF estimation by using the techniques from machine learning and data mining. In the past decade, we have been working on the development of machine learning-based models for estimating TTF and applied the developed technology to various real-world applications such as train wheel prognostics, and aircraft engine prognostics. In this paper, we report two kinds of machine learning-based models for estimating TTF, including multistage classification, on-demand regression. The multistage classification improves the TTF estimation over one stage classification by dividing the time window into more small narrow time windows. A case study, APU prognostics, demonstrates the usefulness of the developed methods. The results from the case study show that the machine learning-based modeling method is an effective and feasible way to develop predictive models to estimate TTF for PHM.