A Hybrid Experimental and Explainable Machine Learning Framework for Predicting Mechanical-Induced Degradation and Reliability in Lithium-Ion Batteries

Abstract

Li-ion batteries are common in electric vehicles and in energy storage systems, but its performance and reliability are greatly influenced by these mechanical loading factors, which are vibration, compression, and impact. This paper proposes a hybrid experimental, machine learning-based model to examine mechanically induced interdependence between the electrical properties and degradation characteristics of li-ion batteries. Mechanical tests on various battery form factors were controlled in order to produce multi-domain data on voltage response, internal resistance, capacity fade, and temperature variations. They used these datasets to train and test several machine learning models, such as Linear Regression, Support Vector Machine, Random Forest and Gradient Boosting, to learn the nonlinear relationship between mechanical inputs and electrical degradation indicators. The findings prove that ensemble models, and especially Gradient Boosting, have a high prediction accuracy with an R² value of up to 0.94. Explainable AI methods and feature importance analysis were used to discover that mechanical strain and compressional force are the most influential factors that lead to degradation. Moreover, the suggested framework can be used to detect faults in the early stages and achieve trustworthy Remaining Useful Life (RUL) estimates with various loading conditions. This research is useful in the creation of smart battery control, as well as the creation of safer and more durable lithium-ion batteries to be used in mechanically active conditions.