Machining quality prediction of aero-engine blades under data-constrained conditions using a gated attention-enhanced Kolmogorov–Arnold network

Predicting aero-engine blade machining quality under data scarcity remains challenging owing to the strong coupling of process variables. We present GA-KAN, which combines localized B-spline feature representations with gated attention to adaptively capture inter-feature relationships. Residual connections and a hybrid regularization scheme (and attention-aware regularization) improve optimization stability and generalization. We evaluate GA-KAN on a proprietary blade milling dataset and a public CNC turning dataset. GA-KAN outperforms competitive baselines, with the average RMSE reduced by 3.3% for roughness and residual-stress targets on the blade milling dataset and by 4.7% for the four roughness indicators on the turning dataset. These results demonstrate GA-KAN’s effectiveness and support its practical use in precision blade machining and other data-constrained manufacturing settings.