Transfer Adaptive Digital Twin for Cross-Domain State-of-Health Estimation of Li-Ion Batteries

Abstract

The conventional digital twin (DT) for state-of-health (SoH) estimation of lithium-ion batteries (LiBs) relies on end-of-cycle estimation to observe battery capacity or is testing-intensive in the case of internal resistance (IR). These health indicators (HIs) are often nongeneralized even under similar operating conditions. This results in data distribution discrepancy that renders the model unsuitable for real-onboard SoH estimation. Therefore, this article demonstrates the implementation of Industrial Internet-of-Things (IIoT)-based DT through Microsoft Azure services based on the convolutional neural network (CNN) with an improved domain adaption method of transfer learning (TL). To facilitate a more comprehensive understanding of real-driving scenarios, this article incorporates a meticulous selection of three driving routes and demonstrates real-time data acquisition by integrating three application programming interfaces (APIs), namely, Google Directions, Google Elevation, and OpenWeatherMap. Finally, the SoH is obtained using the real data in the reference electric vehicle (EV) model that closely emulates the real-time driving behavior using the HI extracted from readily available measurements from the LiB pack. Through hardware implementation for diverse validation scenarios, the proposed model updates extemporaneously and obtains results with errors less than 1.983% for all cases, thereby offering valuable insights toward its significance in battery health prognostics for industrial applications.