Aero-engine blade error distributions predictions using novel machine learning models

Abstract

The milling of thin-walled aero-engine blades presents substantial challenges due to their complex geometry. Accurate prediction of milling error distributions is critical for ensuring machining stability and precision. However, traditional mechanistic models suffer from high complexity and limited adaptability, and most existing data-driven methods often predict only single error values without capturing the full distribution of errors. To bridge this gap, this study proposes a novel Gaussian Mixture Model-based Error Distribution Prediction (GMM-EDP) framework that models the probability distribution of milling errors rather than just point estimates, which, to our knowledge, has not been done in prior studies. Two high-quality experimental datasets were generated after milling 44 blades and 34 impellers, incorporating key machining parameters as inputs. The GMM-EDP framework uses a Gaussian mixture model to characterize complex error distributions and a multi-output machine learning model to predict distributional features. Comprehensive evaluation using Jensen–Shannon divergence, Hellinger distance, total variation distance, and root mean square error (RMSE) demonstrates the framework’s accuracy and robustness. The proposed approach shows excellent generalization across different machining conditions. Results confirm that the GMM-EDP framework not only significantly improves the precision of milling error predictions but also provides deeper insights into machining consistency and uncertainty, which are critical for optimizing process parameters and improving the quality and reliability of thin-walled blade production.