Inference of Melt Pool Visual Characteristics in Laser Additive Manufacturing Using Acoustic Signal Features and Robotic Motion Data

EasyChair Preprint 11476

Abstract

Laser additive manufacturing (LAM) demands robust real-time monitoring to identify defects and ensure product quality. Traditional methods, mainly reliant on coaxial cameras, lacks placement flexibility. Further complexities arise from the inherent variability in melt pool geometries and the high computational demands of image processing, hindering effective real-time monitoring. This research proposes a novel technique to directly infer melt pool visual characteristics in LAM by synergizing acoustic signal features with robotic tool-center-point (TCP) motion data. Acoustic monitoring has shown great promise in tasks typically reliant on vision sensors. In addition, the dynamics of the LAM process and defect occurrences are spatially dependent, primarily due to heat accumulation. By combining acoustic signals with spatial data from robot TCP motion, our method tracks melt pool variations with an R2 score above 0.7. An ablation study demonstrated that the proposed method outperforms the acoustic-only models. The findings suggest that the integration of a simple microphone sensor with robot motion information emerges as a flexible, cost-effective alternative for capturing dynamic melt pool behavior. It presents new prospects for closed-loop control in the LAM process.

Keyphrases: Laser Additive Manufacturing, Robot Operating Systems, acoustic signal processing, machine learning, real-time process monitoring

@booklet{EasyChair:11476,
  author    = {Lequn Chen and Youxiang Chew and Wenhe Feng and Seung Ki Moon},
  title     = {Inference of Melt Pool Visual Characteristics in Laser Additive Manufacturing Using Acoustic Signal Features and Robotic Motion Data},
  howpublished = {EasyChair Preprint 11476},
  year      = {EasyChair, 2023}}