인공지능을 활용한 폴리우레탄 폼 밀도 예측 및 제조 공정 최적화 연구

초록

Polyurethane (PU) foam is widely used across industries such as construction, automotive, and electronics due to its excellent mechanical, thermal, and cushioning properties. However, accurately predicting its physical characteristics particularly density poses a challenge because of the complex nonlinear interactions among raw material components such as Polyol, TDI, Water, Amine, Silicone Oil, Stannous Octoate, and MC. This study aims to evaluate the predictive performance of various artificial intelligence models, including ensemble tree-based methods (Gradient Boosting, Random Forest) and deep learning approaches (Autoencoder, Multilayer Perceptron), for estimating PU foam density. A dataset based on previous experimental formulations was analyzed through correlation and regression techniques to identify key influencing variables. The results indicate that Gradient Boosting achieved the highest predictive accuracy, closely followed by Random Forest, while Autoencoder and MLP showed moderate performance, likely due to limited data size. These findings highlight that ensemble models are better suited for small-scale datasets, whereas deep learning models may become more effective as the dataset grows. The research contributes to the development of intelligent formulation strategies for PU foam, improving process efficiency and material performance in industrial applications.