Predicting the Activation Temperature of 4D-Printed Parts through Machine Learning Algorithms with a Sparse Dataset-Supplementary Information.docx

This study aimed to predict the glass transition temperature (T_g) of 4D-printed shape memory polymers (SMPs) using machine learning while considering 4D printing process parameters and resin composition ratios with a limited dataset. The T_g plays a crucial role in shape recovery, as it represents the activation temperature of 4D-printed SMPs, thus necessitating accurate predictions. However, while previous studies have applied machine learning models to predict strain and recovery stress, the effects of fabrication parameters on T_g have not been sufficiently explored. To address this gap, this study developed a predictive model, integrating both resin composition ratios and post-processing parameters, to estimate T_g. Five machine learning models were trained and evaluated using data from Graeco-Latin square design and one-factor-at-a-time experimental design. Among these, support vector regression demonstrated the highest prediction accuracy (MAE: 1.06, MSE: 2.62), making it the most suitable model. Additionally, the trend of the predicted T_g aligned closely with findings from previous studies, further validating the approach. It proposes a predictive model that enables the efficient optimization of 4D printing materials using a small experimental dataset. By leveraging this approach, both experimental time and cost can be reduced significantly while permitting precise material design and process control.