Development of an Online Defect Detection System for Additive Manufacturing Based on Multi-Sensor Fusion

St. John’s Northwestern Academies, Wisconsin, 53018, United States

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Additive Manufacturing (AM), especially metal Selective Laser Melting (SLM), is essential for high-end industries like aerospace (e.g., turbine blades) and medical implants (e.g., hip implants), but in-situ defect detection remains a key bottleneck limiting its industrialization. Existing single-sensor systems fail to cover surface-subsurface-internal defects comprehensively, while traditional multi-sensor fusion suffers from information loss and poor real-time performance. This study develops an online system integrating visible light, infrared, and ultrasonic sensors, proposing a cross-modal attention-enhanced NSCT-PCNN fusion algorithm to optimize multi-modal feature fusion and an improved YOLOv3 (with an additional small-scale branch and CIoU loss) to boost detection precision. Experiments on the EOS M290 machine using Ti-6Al-4V material show the system achieves 95.8% mAP, 92.3% small defect detection rate, and 21.5 fps speed. Closed-loop control based on the system reduces the defect rate of components from 18.5% to 5.2%, and tensile strength increases by 12.3%. Future work will upgrade ultrasonic hardware and validate the system’s adaptability to multi-material AM processes.