Implementation of an Early Stage Fuel Cell Degradation Prediction Digital Twin Based on Transfer Learning
Meiling Yue, Khaled Benaggoune, Jianwen Meng, Demba Diallo
Abstract
Digital twins are now being widely applied in fault and lifetime prediction of complex systems. In this article, a digital twin for fuel cell degradation prediction via transfer learning method is proposed. As the fuel cell degradation is very susceptible to operation conditions, a multi-input data-driven behavior model of fuel cell degradation is constructed based on a connected convolutional neural network and long short-term memory network to capture both spatial and temporal characteristics hidden in the data. Transfer learning method is applied in order to leverage the knowledge from historical datasets to reliably predict the fuel cell degradation in real-time operation, especially in the early stage. The developed degradation prediction digital twin is cross-validated using two fuel cell aging experiment datasets, and the results showcase the effectiveness and generalizability of the proposed approach. This article contributes to developing an early stage fuel cell degradation prediction digital twin, which is tolerant to different degradation patterns and can achieve real-time degradation prediction.