Computer Vision System for Analyzing Temperature Variations in Metal Additive Manufacturing

This paper discusses improving the quality control process in MAM, which builds complex 3D metal parts layer-by-layer using metal powder. A key challenge in MAM is managing temperature variations during the printing process, as poor thermal control can cause defects such as warping, cracks, and residual stress. To address this, the research introduces a computer vision-based system that captures thermal images throughout the metal printing stages. These images form a dataset representing various temperature conditions, which is then pre-processed using min–max normalization for enhanced analysis. For temperature prediction, a novel ML method ESW-XGBoost is proposed by fine-tuning its hyperparameters, resulting which is more accurate in temperature predictions. The XGBoost methods were also implemented on the dataset, trained the models accordingly, and obtained the corresponding results. Comparative experiments reveal that the ESW-XGBoost model significantly outperforms XGBoost, achieving high prediction performance with 96.87% accuracy, 94.54% precision, 95.73% recall, and a 96.48% F1-score. The ESW-XGBoost algorithm, when combined with computer vision technology, offers a powerful tool for real-time temperature monitoring and process optimization in MAM, enhancing manufacturing outcomes.