Bayesian Optimization Framework and Subjective Bayesian Update Model for the Open-Hole Tensile Strength Design of Laminated Composites

Abstract

This study proposes a Bayesian optimization framework and a related subjective Bayesian update model for the open-hole tensile (OHT) strength design of laminated composites, considering empirical stacking sequence constraints. First, a surrogate model for OHT strength is constructed in terms of lamination parameters using a Gaussian process model, serving as the Bayesian optimization framework, based on experimental strength test results. Subsequently, multiple laminate candidates satisfying the empirical constraints are identified through batch Bayesian optimization using a genetic algorithm with efficient genetic operators, including gene repair strategies and adaptive mutation. To efficiently update the surrogate model, a subjective Bayesian updating method utilizing low-fidelity strength analysis results instead of data from additional strength tests is proposed. Here, “subjective” refers to the engineer’s judgment in defining the uncertainty range of the analysis results to ensure that the approximation accuracy is not compromised. This subjectivity is modeled using a subjective probabilistic framework using the three-point estimation method from the Program Evaluation and Review Technique, which defines pessimistic, most likely, and optimistic values. The Bayesian update process, guided by this subjective stochastic model, is applied to enhance the accuracy of the surrogate model for strength design. The effectiveness of the proposed method is demonstrated through OHT strength design examples of laminates.