Data-Driven Approach to Identify Acoustic Emission Source Motion and Positioning Effects in Laser Powder Bed Fusion with Frequency Analysis

Abstract

Acoustic signal analysis is a common technique for in-process monitoring of laser powder bed fusion (LPBF) to detect defects. However, the propagation of sound waves from the melting pool to sensors can introduce variability in the perceived acoustic signal due to the relative positioning between sensors and the melting pool. This study investigates this phenomenon using a data-driven approach. Sensors, both structural and airborne, not only capture process dynamics but also contextual factors in manufacturing, such as laser movement patterns, baseplate thickness, and printing positions. The frequency data derived from Fourier transforms of the acoustic signals per printing vector were employed as inputs for a 1D-Convolutional Neural Network (CNN) model that can classify these contextual factors. By leveraging Gradient-weighted Class Activation Mapping (Grad-CAM), frequencies crucial for printing positions and acoustic emission source dynamics were identified. The insights obtained from this research aid in developing robust acoustic monitoring systems for LPBF.