Geometric and kinematic modeling of double wishbone suspensions with advanced AI–quantum optimization techniques

The design and optimization of automotive suspension systems, particularly the double wishbone suspension (DWS), remain a foundational yet highly challenging problem in vehicle dynamics due to the need for precise control of camber and caster angles under strongly non-linear and constrained geometric relationships. This dissertation presents a comprehensive investigation into the geometric and kinematic modeling of DWS systems, together with the systematic development and evaluation of advanced artificial intelligence (AI), quantum computing, and hybrid optimization techniques for suspension design. The research progresses from a parametric 2D kinematic model focused on camber behavior to a full 3D symbolic formulation enabling simultaneous camber and caster optimization under strict loop-closure and geometric feasibility constraints. In the 2D phase, classical deterministic solvers (Gradient Descent, Interior Point, Nelder–Mead), metaheuristics (Genetic Algorithms, Particle Swarm Optimization, Ant Colony Optimization), and stochastic search strategies are comparatively evaluated, with the Interior Point method demonstrating superior accuracy and computational efficiency. A hybrid quantum–classical framework combining D-Wave quantum annealing for global exploration with Sequential Least Squares Programming (SLSQP) for local refinement further improves convergence quality and constraint satisfaction. In the 3D phase, high-dimensional multi-objective optimization is addressed through a stacked ensemble framework integrating classical solvers, a reinforcement learning (RL) formulation enabling sequential geometry adaptation, a surrogate-assisted multi-objective Bayesian optimization approach based on Gaussian Processes and Expected Hypervolume Improvement (EHVI), and a QUBO–SQP hybrid algorithm combining global feasibility search with high-precision local refinement. All strategies are evaluated using consistent performance metrics including root mean square error (RMSE), convergence robustness, and computational cost. Results demonstrate that ensemble and RL methods offer strong adaptability in multi-objective settings, while the QUBO–SQP hybrid achieves near-zero objective values. Complementary review studies contextualize AI and quantum optimization in suspension and broader automotive applications, establishing a unified and scalable framework for intelligent, quantum-enhanced mechanical design.