Prediction of battery remaining useful life: an approach using Kolmogorov-Arnold Network (KAN)

Lithium-ion batteries (LIBs) are extensively used across various applications, including electronic gadgets, electric vehi- cles (EVs), and residential energy storage systems. Despite their higher upfront costs, LIBs have become the go-to energy storage option, surpassing traditional technologies like lead-acid and flow batteries. This dominance is attributed to their notable characteristics, such as high energy density, extended lifespan, fast-charging capabilities, and minimal self-discharge, making them crucial for energy storage systems (ESS) in both commercial and industrial sectors. However, degradation of LIBs over time are due to two factors : 1) calendar ageing (results in a slow loss of capacity), and 2) cycling ageing, (which occurs from repeated charge and discharge cycles). Factors that affect these process includes extreme temperatures, high charge/discharge currents, and operation at extreme states of charge can accelerate this degradation. Due to this reason, it is imperative to find the remaining useful life of the battery as a preventive maintenance approach. The ammount of research in this area largely focuses on the deployment of various machine learning models. Recently, the machine learning community has revisited Kolmogorov-Arnold networks (KAN), positioning them as a competitive alternative to traditional feed-forward neural networks that use activation functions like ReLU and sigmoid. KAN is now being applied across various fields. In this study, it is used to predict the Remaining Useful Life (RUL) of batteries. The findings reveal that KAN outperforms several models, including support vector machines (SVM), linear regression, and feed-forward neural networks, achieving an R2 value of 0.983.