Regularization-based Continual Learning for Fault Prediction in Lithium-Ion Batteries

In recent years, the use of lithium-ion batteries has greatly expanded into products from many industrial sectors, e.g. cars, power tools or medical devices. An early prediction and robust understanding battery faults could therefore increase product quality in those fields. While current approaches for data-driven fault provide good results on exact processes they were trained on, often lack ability to flexibly adapt changes, operational environmental parameters. Continual learning promises such flexibility, allowing an automatic adaption previously learnt knowledge new tasks. Therefore, this article discusses different continual group regularization strategies, which are implemented, evaluated compared based a real wear dataset. Online elastic weight consolidation delivers best results, but, as with all examined approaches, its performance appears be strongly dependent task characteristics sequence.