Machine learning in image-based metal additive manufacturing process monitoring and control: A review

Metal additive manufacturing (MAM) has transformed the fabrication of intricate, high-performing components for sectors such as aerospace, automotive, and healthcare. However, maintaining consistent quality remains a significant challenge due to the process’s intrinsic complexity and susceptibility to defects. Recent advances in machine learning (ML), particularly in combination with image-based monitoring and control, have demonstrated significant potential to address these limitations by enabling real-time defect detection, process optimization, and adaptive control. By leveraging techniques such as deep learning and computer vision, ML can extract actionable insights from the vast amounts of image data generated during MAM processes. This allows for the accurate identification of defects ranging from porosity and cracking to thermal distortions while simultaneously predicting anomalies and optimizing process parameters such as laser power, scanning speed, and feed rate. These developments pave the way for closed-loop control systems capable of dynamically adjusting process conditions to mitigate defects, improve part quality, and enhance overall process stability. However, significant challenges remain, including the need for high-quality labeled datasets, computationally efficient algorithms, and robust generalization across different materials, geometries, and process conditions. Addressing these challenges will require the integration of domain knowledge, physics-based models, and advanced ML techniques, alongside the establishment of standardized datasets and evaluation protocols. This review synthesizes current progress and identifies future research directions, emphasizing the transformative role of ML in advancing MAM toward fully autonomous, intelligent manufacturing systems.