Optimization of the elastic properties of block copolymers using coarse-grained simulation and an artificial neural network

Block copolymers consisting of immiscible glassy and rubbery blocks have microphase-separated structures that result in various elastic properties depending on the polymer structures. However, because the complete simulation approach to surveying a wide variety of polymer and microphase-separated structures is time-consuming, a more efficient approach is required for the design of materials with desired properties. In the present study, we used coarse-grained molecular dynamics (CGMD) simulations and an artificial neural network (ANN) to design polymer structures with the desired stress-strain properties. CGMD simulations were conducted to obtain stress-strain curves of linear diblock and triblock copolymers of various chain lengths, block volume fractions, and asymmetricities. We trained the ANN for regression between the polymer structures and the stress-strain curves. Then, using the trained ANN, we performed Bayesian optimization to obtain a polymer structure with an arbitrary target stress-strain curve. CGMD simulations of the optimized polymer structure produced a stress-strain curve that agreed with the curve predicted by the ANN. Therefore, simulation and use of an ANN are potentially useful strategies for the design of polymer structures with desired properties.

Automated summary

This study presents the design of polymer structures with desired stress-strain properties using coarse-grained molecular dynamics simulations and artificial neural networks. Through simulation and optimization, a polymer structure that agrees with the predicted curve is obtained, providing an efficient method for material design.