Generative model-based super-resolution framework for improved defect detection in LPBF process

Abstract

Recent technological advancements have positioned metal additive manufacturing, particularly laser powder bed fusion (LPBF), as a transformative technology in high-value industries. Despite its potential, challenges in process reliability and reproducibility continue to impede its widespread industrial adoption, underscoring the need for robust in-situ monitoring systems for quality assurance. Although high speed camera-based monitoring systems show promise, the inherent constraints of the LPBF environment and prohibitive costs of advanced sensors often require the use of lower-specification cameras. This limitation in image acquisition significantly impacts the data resolution and thus its applications, particularly in artificial intelligence-based defect detection.

This study proposes a novel generative model-based framework comprising enhanced super-resolution and defect detection models. The proposed super-resolution model improves layer-wise images collected from the built-in camera of a commercial LPBF machine by leveraging defect-specific textural features during training. Using the improved images, a subsequent defect detection demonstrates robust performance in identifying process anomalies. Comprehensive quantitative and qualitative analyses are performed to validate the effectiveness of the proposed approach. The results demonstrate substantial improvements in both image resolution and defect detection accuracy, suggesting promising implications for advanced process monitoring and control within LPBF systems.