In-situ surface porosity prediction in DED (directed energy deposition) printed SS316L parts using multimodal sensor fusion

Real-time defect detection is now possible: Researchers achieved 87% accuracy in detecting porosity defects during 3D metal printing as they occur, at a resolution of 0.5mm—a significant advancement over previous post-production inspection methods that could only identify flaws after printing was complete.

Multiple sensor types work better together: The study combined acoustic emission, force, vibration, and temperature data to catch defects more effectively than any single sensor could, demonstrating that multimodal sensor fusion improves reliability in manufacturing quality control.

AI explanations matter for manufacturing trust: By using LIME (an explainable AI method), researchers could identify why the system flagged certain defects—specifically linking high-frequency acoustic signals to two known failure modes (spatter events and insufficient fusion)—rather than treating the AI as a "black box," which is crucial for manufacturers to trust and act on AI recommendations.

Acoustic signals during printing outperform post-production data: Sensor data collected while parts were being printed proved 33% more sensitive at detecting porosity than data gathered during the subsequent machining phase, suggesting that catching problems in real-time is more effective than relying on later inspection stages.

This approach could reduce waste and improve safety: Real-time porosity detection enables immediate correction during manufacturing, potentially preventing defective metal parts (like those used in aerospace or medical devices) from reaching customers and reducing material waste.