A framework to interpret deep learning-based health management system with human interactions

Abstract

Deep learning has shown good performance in detecting a product's faults and estimating the remaining useful life of a product. However, it is hard to interpret deep learning-based health management systems because deep learning is often regarded as a black box. In order to make a maintenance decision based on the result of the management system, humans need to know how it gave the outcome. This study aims to develop a framework that utilizes human interactions during system development to understand the internal process of deep learning. The study will demonstrate the framework on bearing datasets.