PINNs Predict Battery Health from Partial Discharge Data

Researchers have introduced a novel physics-informed deep learning methodology for accurately estimating the State-of-Health (SOH) and predicting degradation in lithium-ion batteries. This approach is particularly innovative because it can derive these critical metrics from partial discharge curves, which are often the only data available under realistic and varied operating conditions. The core of the method lies in its ability to merge data-driven learning with established physical models of battery degradation. This hybrid strategy ensures consistent and robust SOH estimations, even when dealing with incomplete discharge information, achieving a mean absolute percentage error (MAPE) below 4%. Furthermore, the framework includes a real-time degradation trend estimation strategy. This allows for the detection of key aging transitions without needing extensive historical data or predefined protocols, making it highly adaptable across a wide spectrum of battery types and usage scenarios. This overcomes limitations of previous methods that required fixed discharge protocols or relied on non-adaptive, early predictions.