Machine learning assisted design of bioinspired double helicoidal composites under dynamic three-point bending

Purpose This study focuses on optimizing the dynamic fracture behavior of the double helicoidal structure using Bayesian optimization based on Support Vector Regression (SVR) to improve its mechanical properties. Design/methodology/approach To begin, the uniform Latin hypercube sampling method was employed to generate a range of double helicoidal specimens within the design space, featuring helicoidal angles that ranged from 0 to 90 degrees and aspect ratios from 0.5 to 2. Finite element numerical simulations were then conducted to evaluate the stress intensity factor of the specimens under dynamic three-point bending loads. Bayesian optimization based on Support Vector Regression (SVR) models was subsequently developed using the simulation results and design parameters of the specimens. Findings Through iterative optimization, the double helicoidal structures that exhibited optimal stress intensity factors were identified. In the final, the stress concentration and crack propagation path were investigated by analyzing the interior of the structures. The results reveal that stress concentration around a prefabricated crack opening indicates the difficulty of initiating a crack. Originality/value A Bayesian optimization method based on Support Vector Regression was employed to optimize the dynamic fracture behavior of double helicoidal structures.