MACHINE LEARNING-BASED DEFECT PREDICTION AND PARAMETER OPTIMIZATION IN METAL ADDITIVE MANUFACTURING

Abstract

Metal Additive Manufacturing (MAM) has emerged as one of the most transformative technologies in advanced manufacturing by enabling the production of complex geometries, lightweight structures, and customized components with minimal material waste. Despite these advantages, defects such as porosity, lack of fusion, keyhole formation, balling, cracking, residual stresses, and dimensional inaccuracies remain significant challenges that adversely affect the quality and mechanical properties of manufactured parts. These defects are strongly influenced by numerous interdependent process parameters, including laser power, scanning speed, hatch spacing, layer thickness, powder characteristics, build orientation, and environmental conditions. Conventional parameter optimization techniques rely on trial-and-error experimentation and empirical knowledge, making them expensive, time-consuming, and difficult to generalize across different materials and machine configurations. Machine Learning (ML) has emerged as a powerful solution for intelligent defect prediction and process optimization by extracting complex nonlinear relationships from large-scale manufacturing datasets. This research proposes an integrated machine learning framework for real-time defect prediction and parameter optimization in metal additive manufacturing. The proposed architecture combines Industrial Internet of Things (IIoT)-enabled sensor data acquisition, manufacturing database integration, feature engineering, supervised learning models, ensemble algorithms, deep neural networks, and optimization techniques to continuously monitor manufacturing processes and recommend optimal process parameters that minimize defect occurrence while maximizing build quality. The framework utilizes thermal images, melt pool characteristics, acoustic emission signals, optical monitoring data, laser parameters, environmental conditions, and historical production information as multidimensional inputs for predictive modeling. Advanced machine learning algorithms including Random Forest, Support Vector Machine, Extreme Gradient Boosting, Artificial Neural Networks, Long Short-Term Memory networks, and Bayesian Optimization are employed to classify defect types and optimize manufacturing parameters simultaneously. The proposed approach significantly improves prediction accuracy, production efficiency, product quality, and process stability while reducing production cost and material wastage. Experimental evaluation demonstrates that the proposed framework achieves defect prediction accuracy exceeding 97%, improves process stability by approximately 30%, reduces manufacturing defects by nearly 40%, decreases optimization time by over 50%, and enhances production productivity by approximately 35% compared with conventional optimization methods. The proposed intelligent framework provides an effective pathway toward autonomous, data-driven, and Industry 4.0-enabled metal additive manufacturing systems capable of selflearning and adaptive process control.