Machine Learning-Driven Bayesian Optimization of Transmission Gear Ratios for Fuel Economy Enhancement in Conventional Passenger Vehicles

The reduction in greenhouse gas emissions from conventional vehicles powered by internal combustion engines remains a critical challenge for sustainable transportation. Improving fuel efficiency through optimized transmission gear ratio design directly influences engine operation under diverse driving conditions. In this study, a machine learning-based Bayesian Optimization (BO) framework, integrated within vehicle modeling, was employed to optimize gear ratio configurations for a six-speed transmission representative of midsize passenger cars. The Worldwide Harmonized Light Vehicle Test Cycle (WLTC) was used to assess fuel economy under transient operating conditions. The BO algorithm was structured to minimize fuel consumption while ensuring acceptable performance and compliance with WLTP regulations. The results showed that the optimized gear ratios reduced overall fuel consumption by 6.2% compared to the baseline, without requiring any modifications to the engine or other hardware components. Although the largest percentage reductions occurred in cruising (−22.9%) and deceleration (−14.4%), deceleration contributed the largest absolute share of the total saving, whereas acceleration contributed a significant share owing to its dominant baseline consumption (68.9%) despite a smaller relative reduction (−4.2%). Sensitivity analysis indicated that upper gears, particularly sixth gear, had the greatest influence on optimization outcomes. The findings demonstrate that Bayesian Optimization provides an effective and computationally efficient methodology for transmission gear ratios optimization, offering a machine-learning enabled pathway to enhance fuel economy in passenger vehicles.