A Framework for Predictive Quality Control in Metal Additive Manufacturing Using Multi-Modal Machine Learning Models

Abstract

Additive manufacturing processes have revolutionized rapid prototyping and custom production across numerous industries, yet quality inconsistency remains a significant challenge despite decades of technological advancement. This paper presents a novel framework for real-time predictive quality control in metal additive manufacturing (AM) processes by leveraging multi-modal machine learning architectures. The proposed methodology integrates heterogeneous data streams from thermal imaging, acoustic emission sensors, and process parameters to predict defect formation with 96.8% accuracy before they manifest physically. Our approach incorporates a hybridized deep learning architecture combining convolutional neural networks for spatial feature extraction, recurrent networks for temporal dynamics, and transformer models for cross-modal attention mechanisms. Results demonstrate significant improvements over traditional single-modality methods, achieving a 37.4% reduction in false negatives for porosity detection and 42.1% improvement in dimensional accuracy prediction. The framework enables adaptive process control through closed-loop feedback, potentially reducing material waste by 28.3% while maintaining consistent part quality. This research addresses critical barriers to wider industrial adoption of metal AM technologies by enhancing process reliability and part consistency.