A Data-Driven Approach for Electric Vehicle Powertrain Modeling

Electrification in the automotive industry and increasing powertrain complexity demand accelerated, cost-effective development cycles. While data-driven models are recently investigated at component level, a gap exists in systematically integrating them into cohesive, system-level simulations for virtual validation. This paper addresses this gap by presenting a modular framework for developing powertrain simulations. By defining standardized interfaces for key components-the battery, inverter, and electric motor-our methodology enables independently developed models, whether data-driven, physics-based, or empirical, to be easily integrated. This approach facilitates scalable system-level modeling, aims to shorten development timelines and to meet the agile demands of the modern automotive industry.

💡 Research Summary

The paper addresses a critical gap in electric‑vehicle (EV) powertrain development: the lack of a systematic way to integrate data‑driven component models into a cohesive system‑level simulation. While recent studies have applied machine‑learning or statistical techniques to individual components such as batteries, inverters, or motors, they rarely consider the interactions among these subsystems. To bridge this gap, the authors propose a modular framework that defines standardized input‑output interfaces for the three core components—battery, inverter, and electric motor—allowing any combination of physics‑based, behavioral, or data‑driven models to be swapped in without redesigning the whole system.

The methodology follows a V‑model development process. At the top‑down stage, high‑level vehicle requirements (range, acceleration, efficiency, thermal limits) are decomposed into subsystem specifications. Each subsystem is then modeled independently, typically in Python using modern AI libraries (TensorFlow, PyTorch, etc.). The trained surrogate models are exported as Functional Mock‑up Units (FMUs) or directly linked into Simulink for system‑level integration, preserving backward compatibility with rule‑based controllers.

A hybrid architecture is presented where rule‑based control logic (driver torque request, speed regulation) interacts with data‑driven “plant” models. The battery model predicts terminal voltage based on current, state‑of‑charge, and temperature; the inverter model converts DC‑link voltage into three‑phase AC voltages using a simplified modulation index and frequency; the motor model consumes these voltages and an external load torque to output shaft torque, speed, and AC currents. Energy balance is enforced by calculating motor AC power, applying an inverter efficiency map to obtain required DC power, and feeding the resulting DC current back to the battery model. A separate thermal subsystem is also required to capture cooling‑circuit dynamics.

Key advantages of this approach are modularity and flexibility. Designers can pair a high‑fidelity physics‑based battery model for a novel chemistry with a lightweight data‑driven inverter model, achieving a tailored trade‑off between accuracy and computational speed. This enables rapid design‑space exploration, hardware‑in‑the‑loop (HIL) validation, and the development of advanced control strategies.

The authors acknowledge several challenges. Data‑driven models are only as good as the datasets used for training; models trained on standard drive cycles may fail under aggressive driving or extreme temperatures, highlighting the need for diverse, high‑quality data. Integration of heterogeneous models also raises issues of interface consistency, time‑step synchronization, and numerical stability, which the FMU standard helps mitigate.

Future work includes validating the framework against real‑vehicle test data, performing multi‑objective optimization of the machine‑learning algorithms (balancing prediction error, runtime, and memory footprint), and extending the methodology to full‑vehicle HIL simulations. By systematically combining deterministic physics with adaptable data‑driven surrogates, the proposed framework aims to accelerate EV powertrain development, reduce reliance on costly physical prototypes, and support the emerging “New Automotive” paradigm of software‑defined, highly modular vehicles.