Development of surrogate models in reliability-based design optimization: A review.

Area of Science:

• Engineering
• Computational Science
• Optimization

Background:

• Uncertainties are inherent in engineering applications, necessitating Reliability-Based Design Optimization (RBDO).
• Traditional RBDO methods face significant computational challenges due to complex objective and performance functions.
• Surrogate models offer a solution by approximating these functions, reducing computational burden.

Purpose of the Study:

• To provide a comprehensive review and discussion of surrogate modeling methods for RBDO.
• To classify and compare surrogate-assisted RBDO methods, focusing on global and local modeling approaches.
• To analyze the performance of representative global (CBS) and local (LAS) modeling methods using a numerical example.

Main Methods:

• Review and comparison of existing reliability analysis and RBDO methods.
• Summary and comparison of commonly used surrogate models, sample selection, and accuracy evaluation techniques.
• Classification of surrogate-assisted RBDO into global and local modeling strategies.
• Demonstration of Constraint Boundary Sampling (CBS) and Local Adaptive Sampling (LAS) on a 2D RBDO problem.

Main Results:

• Surrogate models significantly alleviate the computational cost associated with reliability analysis and RBDO.
• Global and local surrogate modeling methods exhibit distinct advantages and disadvantages in RBDO.
• Comparative analysis of CBS and LAS highlights their performance characteristics in a practical RBDO scenario.

Conclusions:

• Surrogate-assisted RBDO is a crucial approach for efficient engineering design under uncertainty.
• Understanding the trade-offs between global and local modeling methods is key to selecting appropriate surrogate-assisted RBDO strategies.
• Further research prospects in surrogate-assisted RBDO are identified.