Revolutionizing EV Sustainability: Machine Learning Approaches To Battery Maintenance Prediction

Abstract

Electric vehicle (EV) batteries are one of the most important components to consider since EV batteries contribute to a significant cost share and must present a high performance to meet customers' product quality standards. However, battery behavior and performance are highly delicate to understand and monitor due to the complex interaction between multiple physical, chemical, and thermal phenomena and the different operating conditions. Informed maintenance and repair is a general strategy aiming to prevent failures or repair when most suitable. Predictive maintenance (PdM) is based on continuously analyzing equipment performance, information often extracted from massive datasets recording the equipment condition, unexpected or sudden failure of the equipment could hence be prevented. Data-driven models are a powerful technique of predictive maintenance. In the specific area of battery-electric vehicles, Data-Driven Models of Battery State-of-Health for Predictive Maintenance are the most insightful references. Baert et al. highlight the potential use of machine learning and AI techniques to predict automotive battery failures. CNNs are helpful for automatically transforming raw input data (time series in this specific case) into a statistical design of feature maps that the neural network can effectively learn. Nonetheless, other aspects of the dataset need to be considered. Maximal predicted achievable accuracy does not accurately reflect the goodness of an EV battery PdM model. Predictive maintenance algorithms are primarily studied in the IoT and asset optimization fields. Hence, the dataset is also relevant. The chosen architecture should reflect the data available. It should also be chosen to limit the data missingness, given that incomplete data is one of the worst enemies of machine/deep learning models.