A reliability-constrained inverse design method for the macroscopic mechanical property of composites considering the uncertainty of microscopic parameters

The inverse design of composites is challenged by numerous design parameters and uncertainty of microscopic parameters arising from manufacturing processes, leading to redundant designs and increased costs. This paper proposes a reliability-constrained inverse design method that accounts for the uncertainty of microscopic parameters. The primary objective is to achieve lightweight design of composites that satisfies target macroscopic performance with specified reliability, while minimizing computational expense. The proposed method integrates self-consistent clustering analysis, Fourier amplitude sensitivity test, and artificial neural networks to construct a surrogate model to reduce the cost of computation. Furthermore, a trust-region based adaptive surrogate model method is introduced, which strategically balances the need for global accuracy during the exploration with the requirement for local accuracy near the optimal point. The effectiveness of this method was validated through inverse design experiments, which shows the method can meet the expected goals in terms of macroscopic performance and reliability. Compared to traditional optimization method, the proposed method reduces the computational cost by over 72%. This methodology provides an efficient and reliable pathway for the lightweight inverse design of composites, with significant potential for application in the material development of high-performance, reliability-critical components such as automotive and aerospace structures.