We study a technique of predictive maintenance for systems with solid oxide fuel cells (SOFCs). SOFC-based systems are a digital ecosystem that utilizes energy efficiently because they generate power using oxygen in the air and hydrogen taken out from gas and also make hot water using heat generated during the power generation. Such systems are complicated as they are equipped with many sensors to control fuel flows (i.e., gas and air) and power generation according to power demands. For predictive maintenance of such complicated systems, developing individual algorithms for each individual failure sign is infeasible because of the variety of possible failure signs. Thus, it is necessary to develop a novel algorithm capable of grasping various failure signs. In this work, we develop a method to detect failure signs by performing change-point detection as well as classification analysis of change points to understand the cause of detected change points. In the change-point detection technique, failure signs are detected based on the reconstruction errors calculated with principal component analysis. In the change-point analysis technique, we solve a classification problem on the detected failure signs. Moreover, we propose a method to tell a failure sign that may correspond to unknown types of faults. We show experimental results with data obtained from real SOFC-based systems and confirm the validity of the proposed method.
|Number of pages
|IEEJ Transactions on Electronics, Information and Systems
|Published - 2021
All Science Journal Classification (ASJC) codes
- Electrical and Electronic Engineering