Feature Engineering and Machine Learning Predictive Quality Models for Friction Stir Welding Defect Prediction in Aerospace Applications

Abstract

Data-Driven Predictive Quality solutions are of utmost importance for Industry 4.0 in general and for high added value and complex manufacturing systems in particular. A unique Friction Stir Welding process is performed for the manufacturing of the new Ariane 6 aerospace launchers. This work presents a novel feature engineering approach that correlates Friction Stir Welding process data and quality inspection data to build a Machine Learning-based predictive quality solution. This solution predicts the presence of welding defects, empowering end-user's quality assurance and reducing quality inspection time and associated costs.