top of page


Mr. John Smith

Job title



The adoption of Electric Vehicles (EVs) is primarily limited by their dependence on batteries, which have lesser. power density as compared to conventional fossil fuels as well as its ageing deterioration issues over time. Therefore, there is an urgent need to understand the modifications in battery performance characteristics with respect to changes in temperature, charging behaviour and usage pattern, low and high charge states, current variations etc. To resolve such issues, this work proposes the development of a battery digital twin model to accurately reflect battery dynamics during run time. A digital twin is a virtual model replicating a physical system's characteristics. The digital twin is developed using a physics and machine learning model trained with bench-level and vehicle level actual test data. It uses an equivalent circuit model (ECM) to predict the battery's internal resistance and polarization effect due to ionic diffusion process in the cell. ECM model generates voltage response of the cell with given current input. Model voltage is monitored continuously and checked against actual voltage from the cell. Cell internal resistance is determined at regular interval to calculate SoH of Battery. ML algorithm is used in conjunction with physical battery model. It uses historical data of vehicle to predict the RUL of battery based on driving behaviour. It uses vehicle parameters like vehicle speed, acceleration, braking data to classify type of driving behavior. Internal resistance vs SOC & temperature and OCV data generated is used to train the model. Vehicle level data is analyzed and internal resistance change is determined for vehicle level. Subsequently, SOH is calculated based on resistance change. Based on driving profile, RUL is predicted using ML algorithm. The model is validated using actual vehicle data and the discharge pattern is found to be accurate within 2% RMSE error for voltage response. As ECM do not have ageing phenomena inherent in the system, it has to be calculated based on resistance change explicitly. Driving behaviour plays important role in determining Remaining Useful Life of battery. This has been achieved with integration of ML algorithm based on vehicle data. Thus, it differentiates with other work significantly for RUL prediction due to consideration of driving behavior. Thus, hybrid battery model is presented which gives comprehensive understanding of battery behaviour and ageing phenomena. SOH and RUL of battery is determined based on trained and validated model. High accuracy and correlation observed in model response against measured data. This model can also be extended for battery prognostics.

Mr. Prasanta Sarkar, General Manager, Tata Motors Ltd

Electric Vehicle Battery SoH and RUL Prediction Using Digital Twin Hybrid ML Model Considering Effect Of Driving Behaviour

FWC2023-DGT-013 • Digitalisation


Sign up or login to the ICC to download this item and access the entire FISITA library.

Upgrade your ICC subscription to access all Library items.

Congratulations! Your ICC subscription gives you complete access to the FISITA Library.


Retrieving info...

Available for purchase on the FISITA Store


bottom of page