Physics-Informed Neural Networks for State of Health Estimation in Lithium-Ion Batteries

One of the most challenging tasks modern battery management systems is accurate state health estimation. While physico-chemical models are accurate, they have high computational cost. Neural networks lack physical interpretability but efficient. Physics-informed neural tackle aforementioned shortcomings by combining efficiency with accuracy models. A physics-informed network developed and evaluated against three different datasets: pseudo-two-dimensional Newman model generates data at various points. This dataset fused experimental from laboratory measurements vehicle field to train a in which it exploits correlation internal modeled states measurable health. The resulting performs best synthetic achieves root mean squared error below 2% estimating stays within 3% for test data, lowest observed constant current discharge samples. outperforms several other purely data-driven methods proves its advantage. inclusion information simulation increases further enables broader application ranges.