Efficient computation of robust, safe, fast charging protocols for lithium-ion batteries

The design of fast charging protocols is fundamental to improving the performance and lifetime of lithium-ion batteries. It is well-known that charging operations consistently performed at very high current will negatively impact operational safety and battery lifetime, although a quantitative understanding of these relationships remains lacking. The protocol design problem is typically formulated as a model-based dynamic optimization, where safety of operations can be encoded by constraining relevant battery states. However, all models are affected by uncertainty, which in turn propagates to state predictions. In this case, charging protocols based on nominal predictions may not satisfy the operating constraints. To overcome this issue, this work proposes a stochastic optimal control approach for the efficient computation of safe, fast charging protocols, able to explicitly account for parametric uncertainties affecting the battery model and guarantee probabilistically robust constraint satisfaction. Given a description of uncertainty affecting model parameters, linearized sensitivity analysis is exploited to propagate uncertainty to the battery states, and suitable backoff values for safety constraints are computed for each time instant. The effectiveness of the methodology is demonstrated in silico, by computing five different protocols, with a detailed Multiphase Porous Electrode Theory-based model of commercially available lithium-iron-phosphate batteries.