An adaptive multi-scale kriging framework with committee-based infill for engineering design optimization.

Area of Science:

• Computational Science
• Engineering Mathematics
• Statistical Modeling

Background:

• Kriging (KG) models face efficiency challenges with single-point or fixed multi-point infill criteria.
• Existing methods often require excessive samples or add samples inefficiently per iteration.

Purpose of the Study:

• Propose a novel AMSS-QBC-KG framework to enhance Kriging modeling efficiency.
• Develop a method that adaptively controls sample size based on model accuracy improvement.
• Reduce computational cost while maintaining or improving predictive accuracy.

Main Methods:

• Implemented the Adaptive Multi-Scale Sampling-Query-by-Committee-Kriging (AMSS-QBC-KG) framework.
• Validated the framework using standard test functions and compared it with Expected Improvement (EI), q-EI, Query-by-Committee (QBC), Adaptive Multi-Scale Sampling (AMSS), and Latin Hypercube Sampling (LHS).
• Applied the framework to an engineering case study on submarine pipeline fatigue life analysis.

Main Results:

• The AMSS-QBC-KG framework achieved predictive accuracy comparable to existing methods for low-dimensional problems.
• Demonstrated superior predictive accuracy and reduced computational cost across diverse problem types.
• Successfully verified effectiveness in a fatigue life analysis case study.

Conclusions:

• The AMSS-QBC-KG framework offers an efficient and accurate alternative for Kriging modeling.
• The adaptive sample size adjustment enhances efficiency, particularly in later modeling stages.
• The framework shows broad applicability in engineering and scientific computations.