Intelligent power-split device optimization using a multi-surrogate framework integrating Bayesian learning and physics-based design for next-generation hybrid vehicles

This article introduces a knowledge-augmented design process for power-split devices in hybrid electric vehicles, one of the important devices where improvement in fuel consumption and emissions is critical. In this work, a three-phase framework is presented for mechanical and informatics-driven design that combines computational intelligence with traditional mechanical design methodologies through the coupling of data-driven and engineering-based modelling methodologies. Phase 1 used the validated ADVISOR simulation platform to produce detailed data on vehicle performance, creating a solid base for optimization. In Phase 2, multi-surrogate framework was designed with Bayesian Neural Networks (BNNs) and regression models, which produced an accurate prediction with a mean absolute percentage error (MAPE) of 0.552748% and a mean uncertainty of 0.001535. The better performance of BNN is due to its advanced capability of representing complex nonlinear relationships between gear parameters and performance metrics. A transition, the theoretical optimizations are incorporated towards potential manufacturing designs by employing a computer-aided engineering approach in Phase 3 using the SOLIDWORKS platform to transfer theoretical data into a practical design before ultimately running the designs through ANSYS to evaluate performance. With this methodology, major developments were achieved, such as 2% increased fuel efficiency and 6.74% increased load carrying capacity via gear geometry design optimization. Sensitivity analysis indicated that teeth configurations of sun gears exhibited a sensitivity index of 1.4554, whereas that of ring gears was 0.9686, providing important guidance on the location of the design priority. This combined system not only connects the gap between theoretical optimization and practical implementation but also sets up a new paradigm of sustainable automotive design. This methodology balances improvements in performance against the trade-offs in manufacturing feasibility, showing promise that it can become a part of the automotive powertrain development process beyond just the developed prototype, which is a clear contribution towards sustainable transportation.