Computationally Efficient Machine-Learning-Based Online Battery State of Health Estimation

TL;DR

This paper introduces a lightweight, computationally efficient machine learning method for online battery health estimation, suitable for low-cost BMS, achieving high accuracy with less complex calculations.

Contribution

It presents a novel linear regression-based approach using impedance features, enabling real-time battery SoH estimation on microcontrollers.

Findings

Achieves less than 2% MAE in battery SoH estimation

Validated on two datasets with diverse conditions

Suitable for implementation in low-cost BMS hardware

Abstract

A key function of battery management systems (BMS) in e-mobility applications is estimating the battery state of health (SoH) with high accuracy. This is typically achieved in commercial BMS using model-based methods. There has been considerable research in developing data-driven methods for improving the accuracy of SoH estimation. The data-driven methods are diverse and use different machine-learning (ML) or artificial intelligence (AI) based techniques. Complex AI/ML techniques are difficult to implement in low-cost microcontrollers used in BMS due to the extensive use of non-linear functions and large matrix operations. This paper proposes a computationally efficient and data-lightweight SoH estimation technique. Online impedance at four discrete frequencies is evaluated to derive the features of a linear regression problem. The proposed solution avoids complex mathematical operations and it is well-suited for online implementation in a commercial BMS. The accuracy of this method is validated on two experimental datasets and is shown to have a mean absolute error (MAE) of less than 2% across diverse training and testing data.