The authors of [1] point out that various research groups suggested many physical Cell capacity fade models which sometimes assume very different degradation mechanisms, SEI growth, Loss of active material, and Lithium plating:
Ramadass et al [2] have formulated a model attributing the capacity fade to the Loss of lithium inventory on the negative electrode. Safari et al [3] model assumes that the solvent decomposition reaction at the anode leading to SEI growth is the source of capacity fade. The models of Arora et al [4] and Tang et al [5] have incorporated lithium plating as a reason for capacity fade. Dubarry et al [6] have assumed an exponential loss of active material to model the nonlinear capacity fade behavior. Yang et al [7] have attributed the nonlinear capacity fade occurring in lithium ion batteries to an exponential increase in lithium plating.
Despite appearing contradictory on the surface, all or most of these results might be largely correct: different research groups have used different protocols to cycle cells of various form and chemistry, so the leading degradation mechanisms in their experiments might genuinely be different.
From the above observation, the authors conclude that battery management systems should implement simple semi-empirical models instead of physics-based models:
The complexity of these detailed models makes them largely unattractive for developing estimation and optimization algorithms that can be implemented in microcontrollers. [...] Many assumptions, together with the many parameters, present in detailed battery degradation models complicate an understanding of the major influencing factors of battery degradation.
The development of estimation and optimization algorithms for battery life management that are suitable for microcontroller implementation requires simple semi-empirical models that can capture degradation accurately.
I agree that this is not necessary to use physics-based models in the BMS: for field operation (rather than research), we should always choose the most accurate model available, regardless of whether this model is physics-based or semi-empirical. However, the model used in the BMS should also be able to correctly attribute the changes in the cell sensor output to shifts in different Cell parameters to Estimate the risk of cell failure. Therefore, a cell model used in the BMS could be too simple: such a model might estimate some cell parameters inadequately when it "stretches itself" to explain some unusual cell sensor outputs. In other words, that would be an underfitted model. (Incidentally, this effect is deliberately used by An accurate equivalent circuit cell model with parameter functions trained by an adversarial network.)
A related question is whether to estimate only "relatively fundamental" Cell parameters or derived parameters also (such as cell capacity).
Related:
See also:
[1] Analysis of the effect of resistance increase on the capacity fade of Li-ion batteries
[2] Development of first principles capacity fade model for Li‐ion cells (2004)
[3] Multimodal physics‐based aging model for life prediction of Li‐ion batteries (2008)
[5] Two-Dimensional Modeling of Lithium Deposition during Cell Charging (2009)
[6] Synthesize battery degradation modes via a diagnostic and prognostic model (2012)