Recurrent Neural Network-based Modeling Approach for 4D Printing Design : Long sequence prediction based on short sequences

This study presents a method for predicting the deformation of hydrogel models fabricated by four-dimensional printing technology using deep learning. In this method, a large number of hydrogel models with the same shape but different distributions of length and expansion ratio are first created using the modeling software Abaqus. Then, deformation simulations are performed based on finite element analysis using the thermal expansion method, and the simulation results are processed and organized to create a data set. Afterward, a recurrent neural network (RNN) was created and trained using a composite dataset consisting of multiple hydrogel model sequences of different lengths to enable the RNN to learn model deformation features and, accordingly, predict the deformation of various longer sequences of hydrogels in a very short period of time while recording the change in accuracy of the prediction for each length. This research provides a new approach to modeling and inverse design for 4D printing technology.