AI-Powered Defect Detection System Achieves 99.4% Accuracy in Metal 3D Printing

Scientists at Oak Ridge National Laboratory (ORNL) published results demonstrating an AI-powered defect detection system achieving 99.4% accuracy in identifying internal porosity and lack-of-fusion defects during metal additive manufacturing builds, a significant advancement for quality assurance in critical aerospace and medical applications.

Technology Approach

The system uses a convolutional neural network (CNN) trained on over 2 million melt pool images from DMLS (Direct Metal Laser Sintering) and EBM (Electron Beam Melting) builds. The AI analyzes thermal signature patterns during printing to detect anomalies as they occur, enabling real-time build adjustment or termination before defects propagate.

Real-Time Monitoring Capabilities

Integrated with standard industrial melt pool monitoring cameras, the system processes image data at 1000 frames per second, identifying defects smaller than 50 microns in diameter. The AI distinguishes between acceptable process variation and critical defects that would compromise part integrity.

Industry Applications

The technology targets high-value metal additive manufacturing for aerospace turbine components, medical implants, and fuel system parts where defect-free microstructures are essential. Current manual inspection methods typically detect only 60-70% of internal defects using post-build CT scanning.

Commercialization Path

ORNL has licensed the technology to three major metal 3D printing equipment manufacturers for integration into next-generation systems. Commercial deployment is expected beginning Q4 2026, with the AI models continuously updated through federated learning across installed equipment bases.