Session: 12-06-01: Scientific Machine Learning (SciML) for Characterization, Modeling, and Design of Structures and Materials

Paper Number: 148730

148730 - Computational Discovery of Microstructured Composites With Optimal Stiffness-Toughness Trade-Offs 

Stiffness—the ability to resist deformation in response to an applied force—and toughness—the ability to resist cracks—are two quintessential properties in most engineering materials because these materials must resist nonrecoverable deformation and prevent catastrophic failure under external loading in structural applications. Unfortunately, stiffness and toughness are often mutually exclusive because, to be tough, a material must be ductile enough to tolerate long cracks and absorb more energy before fracturing. It is widely admitted that the conflict between stiffness and toughness is a fundamental problem in engineering materials design. However, the systematic discovery of microstructured composites with optimal stiffness-toughness trade-offs has never been demonstrated, hindered by the discrepancies between simulation and reality and the lack of data-efficient exploration of the entire Pareto front. We introduce a generalizable pipeline that integrates physical experiments, numerical simulations, and artificial neural networks to address both challenges. Our approach uses three evaluators: (i) a mechanical tester that conducts physical measurements, (ii) a finite-element method (FEM)–based simulator that performs virtual mechanical testing in moderate complexity, and (iii) a convolutional neural network (CNN)–based predictor that executes machine learning inference. All evaluators take an arbitrary microstructure design as input and then measure or predict its Young’s modulus and toughness as output, with varying evaluation speed and accuracy. The mechanical tester runs very slowly due to labor-intensive specimen fabrication and testing (~10⁴ s per sample), but this provides ground truth performance values for a microstructure design. At the opposite end of the spectrum, we have the predictor, which runs extremely fast (~10⁻⁵ s per sample) but yields relatively inaccurate results. In between, we have the simulator, which runs reasonably fast (~1 s per sample) given its moderate complexity and delivers intermediate accuracy. Faster evaluators, acting as surrogate models, conduct multi-objective structural optimization and propose microstructure designs on the Pareto front to slower evaluators. Slower evaluators, being more accurate, validate the performance of these designs and use them as additional training data to improve the accuracy of the faster evaluators. Without any prescribed expert knowledge of material design or mechanics, our approach implements a nested-loop proposal-validation workflow to bridge the simulation-to-reality gap and find microstructured composites that are stiff and tough with high sample efficiency. Further analysis of Pareto-optimal designs allows us to automatically identify existing toughness enhancement mechanisms, which were previously found through trial and error or biomimicry. On a broader scale, our methodology provides a blueprint for computational design in various research areas beyond solid mechanics, such as polymer chemistry, fluid dynamics, meteorology, and robotics.

Presenting Author: Wan Shou University of Arkansas

Presenting Author Biography: Dr. Wan Shou is an assistant professor in the Department of Mechanical Engineering at the University of Arkansas (UArk). Before joining UArk, he was a postdoc at MIT from 2019 to 2021. Dr. Shou received his Ph.D. degree in mechanical engineering from Missouri University of Science and Technology in 2018. His research interests cover advanced materials, innovative manufacturing processes, functional applications (electronics, clean water/energy, and robotics), artificial intelligence, and the understanding of materials transformation during manufacturing.

Authors: