Predicting glass transition temperature with tBA-co-DEGDA compositions and 4D printing factors based on design of experiment and machine learning

Abstract

Previous research using a tBA-co-DEGDA photo-resin has shown promising results of more than 20 shape memory cycles. However, due to the cost-extensive material characterization, their work does not consider all possible material ratios and their effects on desired properties such as glass transition temperature (Tg). Therefore, this study identifies the relationship between Tg and 4D printing process parameters to propose a machine learning model for predicting Tg. The effect of individual process parameters on Tg has been analyzed following the OFAT design of experiments methodology, and a model for predicting Tg was created following Graeco-Latin Square design of experiments methodology and machine learning. Various algorithms such as SVM, elastic net, artificial neural network, gradient boosting and random forest were evaluated. Amongst the various algorithms, SVM results in the highest accuracy of Tg prediction with a mean absolute error of 0.94 and a mean squared deviation of 1.57, which is 0.26,0.1 times smaller than gradient boosting algorithms.