Machine learning pipeline for battery state of health estimation (2102.00837v1)

Abstract: Lithium-ion batteries are ubiquitous in modern day applications ranging from portable electronics to electric vehicles. Irrespective of the application, reliable real-time estimation of battery state of health (SOH) by on-board computers is crucial to the safe operation of the battery, ultimately safeguarding asset integrity. In this paper, we design and evaluate a machine learning pipeline for estimation of battery capacity fade - a metric of battery health - on 179 cells cycled under various conditions. The pipeline estimates battery SOH with an associated confidence interval by using two parametric and two non-parametric algorithms. Using segments of charge voltage and current curves, the pipeline engineers 30 features, performs automatic feature selection and calibrates the algorithms. When deployed on cells operated under the fast-charging protocol, the best model achieves a root mean squared percent error of 0.45\%. This work provides insights into the design of scalable data-driven models for battery SOH estimation, emphasising the value of confidence bounds around the prediction. The pipeline methodology combines experimental data with machine learning modelling and can be generalized to other critical components that require real-time estimation of SOH.

• The paper demonstrates a machine learning pipeline that precisely estimates capacity fade with an RMSPE of 0.45% under fast-charging conditions.
• It employs a two-stage approach, integrating rigorous feature engineering with online diagnostics using algorithms like dNNe for superior prediction calibration.
• The pipeline’s success indicates promising real-world applications in battery management systems and potential advancements through operando sensor integration.

An Academic Overview of "Machine Learning Pipeline for Battery State of Health Estimation"

The paper "Machine learning pipeline for battery state of health estimation" addresses a critical challenge in modern battery management systems: the estimation of lithium-ion (Li-ion) battery state of health (SOH) using machine learning techniques. The reliable estimation of battery SOH is pivotal for numerous applications, including electric vehicles and renewable energy storage solutions, facilitating enhanced performance and safety by safeguarding asset integrity.

Summary of the Research

The authors propose a comprehensive machine learning pipeline designed to estimate the battery capacity fade, an essential metric of battery health. The pipeline processes 179 distinct cells under varied conditions, adopting both parametric and non-parametric algorithms to achieve this objective. It engineers 30 features from the charge voltage and current curves, achieving automatic feature selection and calibrating the involved models. One of the most compelling results from this paper is the pipeline's performance on fast-charging protocol cells, where it achieves a root mean squared percent error (RMSPE) of 0.45%.

Algorithms and Methodology

The paper encompasses the development of a two-stage machine learning pipeline. The offline stage is responsible for feature engineering, data augmentation, feature selection, algorithm training, and uncertainty calibration. The online stage employs the pre-trained pipeline to diagnose the health status of an unknown battery cell.

The researchers evaluated four machine learning algorithms as the core of their pipeline:

1. Bayesian Ridge Regression (BRR)
2. Gaussian Process Regression (GPR)
3. Random Forest (RF) with Infinitesimal Jackknife (IJ) confidence intervals
4. Ensemble of Deep Neural Networks (dNNe)

Each algorithm was assessed based on its error rates alongside its capability to gauge predictive uncertainty. Notably, the ensemble of deep neural networks (dNNe) demonstrated superior calibration abilities, indicating a promising direction for future implementations.

Numerical Results and Reflections

The achievement of an rmspe of 0.45% for cells under a fast-charging protocol is particularly impressive. The research establishes that the application of sophisticated machine learning methodologies, combined with carefully engineered features, can significantly enhance the precision of SOH estimates, thereby outperforming traditional equivalent circuit models (ECMs) and electrochemical approaches. Indeed, this suggests a paradigmatic shift towards scalable, data-driven solutions in battery SOH estimation.

Implications and Speculation on Future Developments

The insights offered by this paper have profound implications:

• Practical Application: The implementation of this pipeline in real-world BMS could drastically improve battery safety and efficiency, allowing for dynamic adjustments in operational strategies based on accurate SOH readings.
• Theoretical Advancement: By proposing a rigorous process for uncertainty quantification within predictive models, this research underscores the necessity for robust and comprehensive error metrics in machine learning applications.
• Future Research Directions: For maximum efficacy, future studies might explore integration with operando sensor data to model temperature effects more accurately, potentially enhancing the robustness of SOH predictions across varying environmental conditions. Additionally, leveraging synthetic data from electrochemical simulations in model training could mitigate the limited generalizability observed with data from specific battery types.

The pipeline's adaptability to diverse charging protocols and the development of features that can effectively capture intrinsic degradation markers within batteries paint a promising future for data-driven BMS solutions. The findings stand as a testament to the potential of machine learning to address complex engineering challenges, heralding improved longevity and safety in battery technologies through insightful optimization of battery health management practices.