Quality prediction, interpretation, and optimization of catalytic infrared-cured thermosetting powder coatings via TabPFN and NSGA-II framework

Catalytic infrared (CIR) radiation curing has shown excellent performance in curing thermosetting powder coatings. However, the effect of catalytic infrared curing parameters on coating quality remains unclear. Moreover, predictive models for CIR-cured coating quality are still lacking, and the absence of CIR curing parameters in coating manuals affects the precise control of the curing process. Therefore, this study proposed a multiobjective optimization framework based on the nondominated sorting genetic algorithm II (NSGA-II) and Tabular Prior-data Fitted Network (TabPFN), a novel algorithm designed for small-sample learning. Experimental results and Shapley additive explanations (SHAP) analysis were used to reveal the relationship between the curing parameters and coating curing quality. This study investigated a polyester coating and two hybrid coatings. A total of 150 experimental sets were designed using the Latin hypercube sampling method, and each dataset underwent five tests: adhesion strength, hardness, impact resistance, color difference (ΔE), and gloss. The TabPFN model was compared with seven baseline models: multiple linear regression, support vector machine, random forest, extra tree, decision tree, extreme gradient boosting, and artificial neural network. The results indicated that the TabPFN model achieved comparable accuracy to baseline models. It required no hyperparameter tuning and was trained in 26 seconds, making it highly suitable for engineering deployment. Experimental results and SHAP analysis revealed that adjusting curing distance yields considerable gains during single-sided heating. Furthermore, all coatings responded similarly: Increasing radiation intensity lowered gloss while enhancing adhesion, impact resistance, and ΔE. The TabPFN-NSGA-II was combined with the technique for order of preference by similarity to ideal solution to discover optimal curing parameters. Experimental validation confirmed the effectiveness of the proposed framework, with a mean relative error of 8.10% or less. Compared to hot-air curing, CIR curing reduced curing time by 47–77% while improving the mechanical properties of coatings. However, high radiation intensity negatively affected the coating appearance.