A novel learning function for adaptive surrogate-model-based reliability evaluation.

Area of Science:

• Engineering
• Computational Science
• Reliability Engineering

Background:

• Classical reliability analysis faces challenges with complex engineering structures, leading to increased errors and costs.
• Adaptive surrogate-model-based methods offer a balance between computational efficiency and accuracy in reliability evaluation.
• The learning function is critical for adaptively selecting update samples in these methods.

Purpose of the Study:

• To propose a novel learning function for adaptive surrogate-model-based reliability evaluation.
• To develop a learning function independent of Kriging model prediction variance, enhancing model flexibility.
• To demonstrate the computational efficiency and accuracy of the proposed method through comparative case studies.

Main Methods:

• Development of a new learning function for adaptive sample selection in reliability evaluation.
• Integration of the learning function with surrogate models, not limited to Kriging.
• Validation through four comparative cases: a series system, a nonlinear numerical example, and two practical engineering scenarios.

Main Results:

• The proposed learning function effectively selects optimal update samples.
• The method demonstrates significant computational efficiency and accuracy compared to existing approaches.
• The learning function's independence from specific surrogate model predictions broadens its applicability.

Conclusions:

• The novel learning function provides an effective and versatile approach for adaptive reliability evaluation.
• This method offers a promising solution for analyzing complex engineering structures with improved accuracy and efficiency.
• The proposed technique can be combined with various surrogate models, extending its practical utility.