Using Explainable Artificial Intelligence to Interpret RemainingUseful Life Estimation with Gated Recurrent Unit

In: http://urn.kb.se/resolve?urn=urn:nbn:se:ltu:diva-80191

Abstract

In engineering, prognostics can be defined as the estimation of the remaining useful life of a system given current and past condition. This field has drawn much attention from research, industry, and government as this kind of technology can help improve the efficiency and lower the costs of maintenance in a variety of technical applications. An approach to prognostics that has gained increasing attention is the use of datadriven methods. These methods typically use pattern recognition and machine learning to estimate the residual life of equipment based on historical data. Despite their promising results, a major disadvantage is that it is difficult to interpretthis kind of models i.e. to understand why a certain prediction of remaining useful life was made at a certain point in time. Model interpretability is however of crucial importance to facilitate the use of data-driven prognostics in domains such as aeronautics and energy, where certification is critical. To help address this issue, we use the Local Interpretable Modelagnostic Explanations (LIME) from the field of eXplainable Artificial Intelligence (XAI) to analyze the prognostics of a Gated Recurrent Unit (GRU) on the C-MAPSS data. We select the GRU as this is a deep learning model which a) has an explicit temporal dimension and b) has shown promising results in the field of prognostics. Our results suggest that it is possible to obtain information about feature importance of the GRU both globally (for the whole model) and locally (for a given RUL prediction.