High-throughput prediction of stress–strain curves of thermoplastic elastomer model block copolymers by combining hierarchical simulation and deep learning

Abstract

Abstract: We achieved high-throughput prediction of the stress–strain (S–S) curves of thermoplastic elastomers by combining hierarchical simulation and deep learning. ABA triblock copolymer with a phase-separated structure was used as a thermoplastic elastomer model. The S–S curves of the ABA triblock copolymers were calculated from the hierarchical simulation of self-consistent field theory calculations and coarse-grained molecular dynamics simulations. Because such hierarchical simulations require considerable computational resources, we applied a deep learning technique to accelerate the prediction. Sets of phase-separated structures and the S–S curves obtained from the hierarchical simulation were used to train a 3D convolutional neural network. Using the trained network, we confirmed that the predicted S–S curves of the untrained structures accurately reproduced the simulation results. These results will enable us to design novel polymers and phase-separated structures with desired S–S curves by high-throughput screening of a wide variety of structures. Graphic abstract: [Figure not available: see fulltext.]