Challenges of machine learning-based RUL prognosis: A review on NASA's C-MAPSS data set

The estimation of a system's or a component's remaining useful life (RUL) is considered the most complex task in predictive maintenance, at the same time the most beneficial one. In this brief review paper, we survey the state-of-the-art in machine learning-based RUL prognosis based on research on NASA's C-MAPSS data set. We identify the frequently used models, comparatively evaluate model performance and survey the used feature extraction methods. As a main contribution, we formulate challenges in the field, independently of the C-MAPSS data set. Among the challenges are interpretability, model uncertainty and domain adaptation, i.e. transfer learning. The identified challenges may serve to identify potential research directions, in order to further push the field of machine learning applied to RUL prognosis.

I. Introduction

Modern machinery is becoming increasingly complex to enable faster and flexible operation. In order to make optimal use of machines and to ensure planning security, the failure-free execution of an intended task is more important than ever. With the ever growing digitization of fields like manufacturing [1], automotive [2]–[4] and aviation [5], predictive maintenance (PdM) has gained a lot of attention. At the same time, machine learning (ML) has made its way from research to industrial applications. Advanced ML methods, e.g. deep learning (DL), have achieved unprecedented accuracies [6]. Data-driven PdM, specifically ML, is now widely used for the detection, diagnosis and prognosis of faults. The prognostics task to predict an asset's remaining useful life (RUL), i.e. the time span from the current time point to the end of the useful life [7], is considered as the most complex part of PdM [8], yet the most beneficial one. Predicting the time of failure enables not only cost-effective scheduling of maintenance, but also related activities like spare part provisioning [9], [10].