Use of Model-Based Datasets and Neural-Networks to Support Real-Time Driving Suggestions System for Components Preservation in Battery Electric Vehicles University of Sannio

Format:

Book

Conference/Event

Author/Creator:

Bernardi, Mario Luca, author.

Contributor:

Capasso, Clemente

Iannucci, Luigi

Sequino, Luigi

Conference Name:

17th International Conference on Engines and Vehicles (2025-09-14 : Capri, Italy)

Language:

English

Physical Description:

1 online resource cm

Place of Publication:

Warrendale, PA SAE International 2025

Summary:

Nowadays, Battery Electric Vehicles (BEVs) are considered an attractive solution to support the transition towards more sustainable transportation systems. Although their well-known advantages in terms of overall propulsion efficiency and exhaust emissions, the diffusion of BEVs on the market is still reduced by some technical bottlenecks. Among those, the uncertainty about the expected durability of the vehicle's onboard battery packs plays a key role in affecting customer choice. In this context, this paper proposes the use of model-based datasets for training a driving support system based on machine learning techniques to be installed on board. The objective of this system is to acquire vehicle, environmental, and traffic information from sensor' networks and provide real-time smart suggestions to the driver to preserve the remaining useful life of vehicle components, with particular reference to the battery pack and brakes. For the generation of the training dataset, first, a set of onboard measurements is performed with the vehicle running in different operational conditions in terms of driving style, environmental temperature, road surface, and traffic intensity. Then, experimental tests are carried out to parametrise and validate battery electro-thermal simulation models, which are used, in combination with an electric vehicle model and the related brake-wearing sub-model, to perform long-term analysis through multiple runs of the acquired driving cycles. The proposed system employs federated learning to enhance prediction models while preserving data privacy. Vehicles contribute locally trained parameters to a global model, reducing data transfer overhead and adapting to evolving driving conditions. Federated averaging minimises model drift across clusters, ensuring consistency. Edge computing processes data locally, enabling low-latency decision-making. Optimised neural networks ensure efficient execution on embedded systems, enhancing real-time driver support. By integrating federated learning and edge AI, the system achieves robust, scalable, and privacy-preserving optimisation for next-generation electric mobility

Notes:

Vendor supplied data

Publisher Number:

2025-24-0115

Access Restriction:

Restricted for use by site license