Modeling the nonlinear responses of soft network materials assisted by masked-fusion artificial neural network

Soft network materials (SNMs) have attracted intensive attentions due to the merits of lightweight, high stretchability, and highly tunable biomechanical properties and responses. In this context, the SNMs have been widely used in the fields of flexible electronics, health monitoring, and tissue engineering etc. in the last decade. To assist the modeling of nonlinear J-shaped stress-strain curve of SNMs with complex curved filaments, a phenomenological framework assisted by machine learning (ML) approach was developed in the last year. However, the overfitting problem affected the computational efficiency and numerical stability. Herein, a novel masked-fusion artificial neural network (ANN) has proposed to replace the conventional feed-forward ANN in ML approach in this research. Compared to the conventional strategy, the masked-fusion ANN has fewer hidden layer which leads to smaller test loss and less computational burden. The risk of overfitting can also be avoided. Besides, the predicted J-shaped stress-strain curves of randomly generated SNMs from masked-fusion ANN are closer to those obtained from finite element (FE) analyses compared to the conventional ANN.