Files
Download Full Text (822 KB)
Date of Graduation
5-2025
Description
We consider the problem of learning and optimizing the performance of a system by conducting a limited number of physical and digital experiments within a design space. Physical experiments are assumed to be unbiased but costly, while digital experiments (e.g., simulations) are less expensive but may introduce bias due to the limitations of the simulation model. This problem is relevant in many fields, such as optimizing engineered systems where performance (e.g., mechanical properties and reliability) depends on various design variables and external/internal factors. Without digital experiments, optimizing the system’s performance amounts to evaluating a noisy and expensive-to-assess black-box function, a task commonly handled using Bayesian Optimization (BO). Our research extends BO by incorporating digital experiments between subsequent physical experiments, aiming to (i) improve simulation model calibration and (ii) identify solutions that are likely to generate desirable physical experiment results. We introduce “Simulation-Enhanced Bayesian Optimization” (SEBO), a methodology that integrates these steps, and evaluate it using various one- and two-dimensional benchmark functions. A bias function is used to model the simulation model’s bias across the design space and its parameters. We compare SEBO to traditional BO, with preliminary results demonstrating SEBO’s advantages in optimizing experimental efforts; SEBO outperforms traditional BO for well-behaved functions, requiring fewer physical and digital experiments to achieve a desired objective function value. By effectively combining physical and digital experiments, SEBO offers significant potential for improving the design and optimization of engineered systems, reducing costs, speeding up design processes, and overall providing more efficient solutions in engineering and manufacturing.
Publication Date
2025
Document Type
Book
Degree Name
Bachelor of Science in Data Science
Degree Level
Undergraduate
Department
Data Science
Advisor/Mentor
Sullivan, Kelly
Disciplines
Engineering
Keywords
Engineering
Citation
Kim, L., Sullivan, K., Liao, H., & Pohl, E. (2025). Simulation-Enhanced Bayesian Optimization of System Designs using Hybrid Physical and Computer Experiments. 2025 Research Poster Competition. Retrieved from https://scholarworks.uark.edu/hnrcsturpc25/14
