Digital-Twin Enabled Online Remaining Useful Life Prediction of PEM Fuel Cell

Developing affordable and efficient methods to utilize hydrogen from renewable sources is a crucial hurdle that must be overcome to achieve a low-carbon economy and industrial decarbonization. One promising solution is the proton exchange membrane fuel cell (PEMFC), which is a commercially attractive fuel cell that can support the implementation of decarbonization techniques for the transport sector. However, the unstable operation and undesired degradation will significantly shorten PEMFC's lifetime, which is a serious barrier to its commercialisation. To tackle this challenge, we propose a digital twin-enabled online remaining useful life prediction method for PEMFC, based on the Long Short-term Memory (LSTM) neural network and the quantile Huber loss (QH-loss). Our approach involves using a digital model that can be learned from a short period of online monitoring data and then used to estimate the remaining useful life (RUL) for upcoming data. By utilizing the proposed digital twin method, we can simulate the performance of the PEMFC in real-time, providing accurate and timely predictions of its RUL. We conducted experiments on PEM fuel cell test rigs, varying the length of the data period used to train our digital twin model. The results of our experiments demonstrate the effectiveness of the proposed method, showing that our approach can accurately predict the remaining useful life of PEMFCs, even when trained with a short period of online monitoring data.