Kinematics Design of a MacPherson Suspension Architecture Based on Bayesian Optimization

Area of Science:

• Engineering
• Computer Science
• Optimization

Background:

• Traditional engineering design relies on manual expert input and lengthy simulation/prototyping cycles.
• Current methods are time-consuming and costly, involving computer simulations and physical prototypes.
• Discipline models often lack gradient information, limiting automated optimization.

Purpose of the Study:

• To develop a Bayesian optimization (BO) system for partially automating the engineering design process.
• To create a general framework for optimizing compliance with target specifications with respect to design parameters.
• To introduce a three-tier convergence criterion for robust optimization.

Main Methods:

• Implemented a Bayesian optimization (BO) system to directly optimize design compliance.
• Developed a general framework for computing the generalized inverse of high-dimensional nonlinear functions without gradient information.
• Introduced a three-tier convergence criterion: optimal solution, infeasibility detection, and probably approximately correct (PAC) solution.

Main Results:

• Demonstrated the proposed BO approach on benchmark functions and a vehicle chassis design problem.
• Showcased the generality, scalability, and efficiency of the developed system.
• Validated the straightforward implementation of novel convergence criteria within popular BO software.

Conclusions:

• The developed Bayesian optimization (BO) system offers a general, scalable, and efficient method for automating engineering design.
• The novel three-tier convergence criterion enhances the robustness of the optimization process.
• This approach significantly reduces the time and cost associated with traditional engineering design workflows.