PhD Defence Reza Azizighalehsari | Li-ion Batteries Diagnostics and Prognostics: From First to Second Life

Li-ion Batteries Diagnostics and Prognostics: From First to Second Life

Reza Azizighalehsari is a PhD student in the Department of Power Electronics. (Co)Promotors are prof.dr. T. Batista Soeiro, prof.dr.ir. G. Rietveld and dr.ir. P. Venugopal from the Faculty of Electrical Engineering, Mathematics and Computer Science.

The global shift toward electric mobility has led to a growing volume of end-of-life Lithium-ion Batteries (LiBs), presenting both critical challenges and opportunities, particularly regarding their second-life deployment. This dissertation presents novel, robust methodologies for diagnostics and prognostics of LiBs, specifically addressing the complexities and uncertainties associated with their second-life applicability, aligning closely with the objectives of the European Union’s Battery Passport initiative.

The work begins with a comprehensive review of second-life battery opportunities and barriers, outlining the growing availability of retired EV batteries and the potential to repurpose them for stationary energy storage and off-grid applications. However, these opportunities are hindered by key challenges, highlighting the complexity of assessing battery state of health without access to standardized data from first-life EV batteries.

This is followed by the development of a hybrid prognostics framework that combines statistical forecasting with deep learning using Electrochemical Impedance Spectroscopy (EIS)-derived features. EIS features, extracted from first-life cycling data, are used as inputs to a two-stage model: an ARIMA-based statistical forecaster and a BiLSTM deep learning model. Validated on an extensive dataset of 18650 NMC cells aged under varied profiles, the model achieved exceptionally low error rates (RMSE < 0.17%), providing a reliable solution for predicting State of Health and Remaining Useful Life in predefined second-life scenarios.

Recognizing the limitations of electrochemical measurements in field conditions, the next phase of the thesis explores thermal diagnostics as a non-invasive alternative. Using surface temperature as accessible data from battery management systems, a dual-model architecture is developed for the batteries. This approach accurately classifies degradation history, achieving 98% classification accuracy and diagnostic errors within ±0.3% using XGBoost and Random Forest models.

To further enhance the accuracy and interpretation of EIS data, this thesis makes another major contribution by introducing an automated parameterization strategy for equivalent circuit models (ECMs) based on the Distribution of Relaxation Times (DRT). By combining DRT with a least squares fitting (LSQF), the method improves modeling accuracy. This approach significantly enhances accuracy, robustness, and computational efficiency in interpreting large-scale EIS datasets.

It reduces ambiguity in parameter estimation, a common issue that often arises when working with battery aging datasets. This advancement positions EIS-DRT modeling as a viable candidate for real-time diagnostics and integration into smart BMS platforms.

These methodologies collectively facilitate the implementation of intelligent, data-driven battery management systems that ensure battery safety, performance consistency, and lifecycle traceability through cloud-integrated platforms. The outcomes contribute substantially to advancing battery reuse strategies, addressing key technical barriers. Finally, recommendations for future research highlight opportunities in real-time diagnostic integration, end-of-life characterization, and intelligent battery scoring systems, paving the way for the sustainable integration of second-life LiBs into energy storage infrastructures.