Diagnosing battery degradation is critical for understanding the state of a cell in operation to predict remaining useful life. Degradation Mode Analysis (DMA) is increasing in popularity as a tool for diagnosing the state-of-health (SOH) in more detail than capacity-based methods. By attributing degradation to loss of lithium inventory (LLI) or loss of active material (LAM) on each electrode, the gap between the diagnosed state and the underlying degradation mechanisms is reduced.
DMA relies on a fitting procedure, which makes the potential sources of error numerous. As a thermodynamic method, DMA works best close to equilibrium, such that it is usually performed on experiments at low currents. But the resistance increase of a cell with degradation also increases the overpotential during electrochemical measurement at a fixed current in Amps. This effect can be exacerbated by the selection of reference performance test (RPT) procedures; best practices would suggest taking pseudo-open circuit voltage (pOCV) measurements at currents lower than C/25 (1) but this may be impractical, particularly in real-world applications. Inhomogeneous distribution of degradation inside the cell is also known to suppress features in the OCV profile that are required to make a DMA fit (2). The literature provides no singular agreed method for conducting DMA. Options include fitting to the unaltered pOCV data or to its derivatives and there are multiple potential approaches to correcting for the overpotential caused by cell resistance.
Previous attempts to validate DMA methods have used synthetic datasets generated from known degradation mode combinations (3,4). In reality, however, there is no ground truth from which the degradation modes predicted by these methods can be validated without tearing down a cell (5). It is therefore impossible to quantify the error in a DMA fit in-situ.
This work uses a physics-based battery degradation model in PyBaMM (6) to generate a synthetic dataset under different conditions, on which the error in DMA results is quantified by comparison with the degradation modes directly predicted by the model. This approach allows the performance of DMA methods to be quantified:
• at different cell states of health (SOH),
• reached as the result of different dominant degradation mechanisms that impact cell performance beyond LLI and LAM through mechanisms such as porosity change,
• using different RPT procedures, and
• with the effect of degradation heterogeneity captured by a distributed model (7).
We use PyProBE (8), an open-source Python package for battery data processing that includes a library of post-processing methods, including methods for DMA. Multiple approaches for performing the DMA fit can therefore be run within the same environment, allowing comparisons to be made.
We demonstrate that careful selection of RPT procedure and fitting technique is critical for accurate prediction of degradation modes, with failure to do so leading to erroneous results. We show how the accuracy of DMA changes as the cell degrades, which must be considered when interpreting results. Finally, we provide users of DMA in research and industry an insight into the magnitude of error in their diagnostic tools so that they can attribute informed uncertainty to their results.
1. Dubarry M, Anseán D. Best practices for incremental capacity analysis. Frontiers in Energy Research. 2022;10. https://www.frontiersin.org/articles/10.3389/fenrg.2022.1023555
2. Lewerenz M, Fuchs G, Becker L, Sauer DU. Irreversible calendar aging and quantification of the reversible capacity loss caused by anode overhang. Journal of Energy Storage. 2018;18: 149–159. https://doi.org/10.1016/j.est.2018.04.029.
3. Dubarry M, Beck D. Big data training data for artificial intelligence-based Li-ion diagnosis and prognosis. Journal of Power Sources. 2020;479: 228806. https://doi.org/10.1016/j.jpowsour.2020.228806.
4. Birkl CR, Roberts MR, McTurk E, Bruce PG, Howey DA. Degradation diagnostics for lithium ion cells. Journal of Power Sources. 2017;341: 373–386. https://doi.org/10.1016/j.jpowsour.2016.12.011.
5. Chen J, Marlow MN, Jiang Q, Wu B. Peak-tracking method to quantify degradation modes in lithium-ion batteries via differential voltage and incremental capacity. Journal of Energy Storage. 2022;45: 103669. https://doi.org/10.1016/j.est.2021.103669.
6. Sulzer V, Marquis SG, Timms R, Robinson M, Chapman SJ. Python Battery Mathematical Modelling (PyBaMM). Journal of Open Research Software. 2021;9(1). https://doi.org/10.5334/jors.309.
7. Li S, Kirkaldy N, Zhang C, Gopalakrishnan K, Amietszajew T, Diaz LB, et al. Optimal cell tab design and cooling strategy for cylindrical lithium-ion batteries. Journal of Power Sources. 2021;492: 229594. https://doi.org/10.1016/J.JPOWSOUR.2021.229594.
8. Holland T, Cummins D, Marinescu M. PyProBE: Python Processing for Battery Experiments. Submitted. https://github.com/ImperialCollegeLondon/PyProBE