Session: 03-03-01: Applications of AI

Paper Number: 162111

162111 - Generative and Uncertainty-Informed Inverse Design Reveals Diverse Fracture Mechanisms of Bio-Inspired Materials 

Data-driven engineered materials design offers prospects for developing structures and systems with evolutionary material properties. In contrast to forward modeling, the inverse design process is challenging due to its intractability and non-uniqueness of potential solutions. To circumvent this challenge, we introduce Generative and Uncertainty-informed Inverse Design (GUIDe), a method utilizing probabilistic forward prediction, statistical inference, and Markov Chain Monte Carlo sampling to generate designs with targeted nonlinear behaviors. Unlike commonly used $\mathrm{response} \rightarrow \mathrm{design}$ inverse model-based methods, such as tandem neural networks and conditional generative models, the proposed method is based on a $\mathrm{design} \rightarrow \mathrm{response}$ forward model, which provides a framework that estimates the likelihood of achieving desired response targets considering predictive uncertainty. Specifically, given a forward model that formulates the distribution of materials’ nonlinear constitutive relation, it can estimate the likelihood of possible material designs satisfying the target constitutive relation and then sample new designs based on likelihood. Thus, the proposed method operates as a generative model for design solutions, capturing the one-to-many mapping from the target to designs and characterizing the distribution of acceptable outcomes. We demonstrate the method’s effectiveness through a case study on designing the trilinear traction-separation law of bio-inspired composites to achieve desired stress-strain responses, realizing extrapolation from out-of-range properties to designs and discovering diverse feasible regions in the design space that reveal various fracture mechanisms corresponding to the same stress-strain relationship. Therefore, this method can provide deeper insights into the fundamental physics governing material failure.

Presenting Author: Haoxuan Mu Texas A&M University

Presenting Author Biography: Haoxuan Mu is a Ph.D. student at Texas A&M University.