This thesis presents an optimization methodology for improving the curving and tangent performance of a forced steering passenger train. We perform multi-objective optimization using a Non-dominated Sorting Genetic Algorithm-III (NSGA-III) to obtain a set of Pareto optimal solutions representing different combinations of primary suspension longitudinal and lateral stiffness, and steering linkage parameters of the train’s bogie. Wheelset unloading, derailment risk, rail rollover risk, and train carbody lateral acceleration are considered as objective functions during optimization, representing the curving and tangent performance criteria.

Prior to optimization, a high-fidelity simulation model of the forced steering train is developed based on vehicle data from a railway transit operator. The model is validated by comparing the wheelset lateral force and angle of attack outputs to those of a conventional rail vehicle. Wheelset yaw angles generated at a test curve are also compared to vehicle design data. To simplify the optimization process and study the effects of the design parameters on the output performance, global sensitivity analysis is performed using the Sobol method. This analysis ranks the design parameters based on their impact on the curving and tangent objective functions, allowing us to ignore the least impactful parameters during optimization. Due to the computationally expensive nature of the high-fidelity simulation model, conducting numerous simulation evaluations during optimization is impractical. Therefore, a Kriging surrogate model is developed to model the input-output relationship between the design variables and objective functions. The surrogate model is trained using data from the sensitivity analysis. After optimization, to demonstrate the robustness of the Pareto optimal designs to varying operating conditions, a few test cases of the optimal designs are selected and simulated under varying curve radii, track friction, and equivalent conicity of the wheel and rail.

Key findings include the identification of steering ratio and longitudinal primary suspension stiffness as critical design variables, with minimal impact from yoke-to-yoke parameters. The optimization resulted in a Pareto optimal set, revealing a trade-off between curving performance and car body lateral acceleration, with optimal solutions varying based on lateral stiffness. Under varying curve radii and equivalent conicity, two out of four selected optimal sets demonstrated improved robustness compared to the original design. In conditions of varying curve radius and track friction, all optimal sets showed equal robustness and a significant improvement over the original design. The study underscores the importance of considering robustness across diverse operational conditions and emphasizes the need for a comprehensive optimization approach to achieve balanced vehicle performance.