Deep learning for predicting the thermomechanical behavior of shape memory polymers

Thermomechanical constitutive modeling is essential for shape memory polymers (SMPs) to be used in engineering structures and devices. However, the classical method of deriving constitutive models is difficult, time consuming, and relies heavily on trial and error. In this work, we aim to decrease the time and resources needed to develop new thermomechanical models for SMPs. The method proposed in this work uses deep learning (DL) to predict the thermomechanical behavior of SMPs under thermomechanical cycles. Particularly, a semicrystalline two-way shape memory polymer (2W-SMP) is selected as an example. Predicting such behavior will give insight on the SMP properties and help validate its characteristics. In this paper, we have compared several DL models to find which one can predict the experimental thermomechanical behavior with the highest accuracy within a reasonable time frame. The results reveal that the fully connected neural network (FCNN) and the convolutional neural network (CNN) were the most accurate DL models. Overall, using one of the selected DL models, we can predict the results of new iterations of the experiment without spending as much time and resources.