REAL-TIME PRINT TRACKING IN METAL ADDITIVE MANUFACTURING USING ACOUSTIC EMISSION SENSORS AND VISION TRANSFORMER ALGORITHMS

Directed Energy Deposition (DED) is an additive manufacturing (AM) method with applications in the aerospace, automotive, and healthcare sectors. In such complex and high-stakes applications, accurate and reliable monitoring is indispensable for assuring fabrication quality. Conventional monitoring systems using mechanical encoders and optical devices have limitations such as wear susceptibility and line-of-sight issues respectively, thereby necessitating alternative monitoring systems. One crucial aspect often overlooked in conventional monitoring systems is the real-time printing process tracking. Achieving high accuracy in tracking is paramount for identifying and mitigating defects in real-time, ultimately leading to improvements in fabrication quality. To address this problem, this research employs acoustic emission sensors for real-time monitoring in Laser DED. These sensors augment existing monitoring methods to improve both reliability and part quality. We developed and tested two machine learning models to study acoustic data correlation with the print process tracking. The first model, based on a Hybrid Convolutional Auto Encoder (HCAE), achieved over 94% accuracy in the print head spatial localization. The second, a Transformer-based model, excelled with a 98.5% accuracy rate and computational efficiency in process tracking. Our findings promise enhanced printing process tracking and pave the way for advanced AI algorithms incorporation into AM quality monitoring. The AI-enabled methods developed can be generalized to other manufacturing applications such as Laser Powder Bed Fusion.