For commercial vehicles, reliability is key since the vehicle is typically linked to the daily earnings of the owner. To ensure continuous vehicle operation, early diagnostics of critical issues and proactive maintenance are important. However, an electric vehicle is a complex and dynamic system consisting of numerous components interacting with each other and with external environments such as road conditions, traffic, weather, and driving behavior. Thus, vehicle operation and performance are highly contextual and for identifying an abnormal operation (diagnostics) the solution must consider the conditions under which it is driven.
To address this, the paper proposes an AI-based digital twin of an electric three-wheeler vehicle. TabNet a deep-learning based model is used to learn and generate near-ideal vehicle behavior. The focus of the paper is motor subsystem. The model is trained using appx 200 vehicles first 1500 km driven data. To ensure, the digital twin model learns near-ideal vehicle behavior, the vehicles used for training are the ones that did not report any issues during the initial and subsequent three months.
Results show that the digital twin model can faithfully reproduce good vehicle behavior with low error while vehicles with reported progressive issues in motor, show higher error earlier thus enabling the service team to do proactive maintenance for these key components.