Comparative Analysis of Deep Learning Approaches for Predicting Thermomechanical Behavior of Shape Memory Polymers

Thermomechanical constitutive modeling is crucial for understanding and designing shape memory polymers (SMPs) for advanced engineering applications. Traditional approaches are often time-consuming and computationally expensive which requires the development of more efficient and accurate methods. In this paper, we propose a Transformer-based deep learning model to predict the thermomechanical behavior of semicrys-talline two-way shape memory polymers (2W-SMPs) under thermomechanical cycles. By leveraging its ability to capture complex, time-dependent patterns, the framework accurately models the intricate relationships between polymer strain, time, temperature, and stress. A comparative analysis with other deep learning models demonstrates that the Transformer excels in accuracy and robustness by capturing intricate dependencies and non-linearities in the data. The test results demonstrate that our proposed Transformer model achieved a root mean squared error (RMSE) of 0.3365, outperforms the traditional deep learning models such as the feedforward neural network (FNN), convolutional neural network (CNN), and long short-term memory (LSTM) models which achieved RMSE values of 6.78, 17.35, and 13.85, respectively. This study highlights the potential of the Transformer model as a powerful tool for predicting material behavior and reducing the time and resources required for constitutive modeling.