Privacy Preserved Melt Pool Data Sharing in Metal Additive Manufacturing Processes

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Collaborative process-defect modeling can significantly advance the development of generalized, reliable machine learning models for anomaly detection and defect prediction in metal additive manufacturing processes. However, sharing data, such as melt pool images, raises concerns over data privacy and Intellectual Property (IP) protection. This study addresses these challenges by proposing two privacy-preserving mechanisms for sharing melt pool images: (1) binary segmented melt pool images and (2) binary segmented images with random rotation. To prevent scan pattern leakage, melt pool images are shared in randomized order under both mechanisms, ensuring spatial analysis capabilities without compromising privacy. To evaluate these mechanisms, two privacy attack models, defined as Scan Pattern Recognition Attacks (SPRA), were developed: SPRA-PP (plume-position-based) and SPRA-MF (motion-feature-based). While SPRA-PP identifies scan directions from the melt pool images in over 92% of cases, demonstrating privacy leakage from raw melt pool image data sharing, it fails when the first privacy mechanism is implemented. Conversely, SPRA-MF defeats the first privacy mechanism and identifies the scan direction in over 98% of cases. However, the second, more conservative privacy mechanism is effective against both attacks, with SPRA-MF's success rate dropping from 98% to approximately 60% when the second privacy mechanism is implemented. Our findings also reveal an inverse relationship between data utility and privacy retention, with the second mechanism maintaining above 80% of data utility. The proposed data-sharing approach can facilitate secure and collaborative LPBF research, advancing additive manufacturing by balancing data privacy with innovation.