Available at: https://digitalcommons.calpoly.edu/theses/3139
Date of Award
6-2025
Degree Name
MS in Mechanical Engineering
Department/Program
Mechanical Engineering
College
College of Engineering
Advisor
Ramanan Sritharan
Advisor Department
Mechanical Engineering
Advisor College
College of Engineering
Abstract
This thesis develops and trains a convolutional neural network (CNN) to classify defects in metal parts produced by selective laser melting (SLM) using acoustic emission (AE) data. The model successfully identified porosity and standard build layers with high accuracy but consistently struggled to classify lack of fusion (LOF) defects, regardless of input type, normalization, or class count. These results suggest that while AE signals contain distinguishable features for certain defect types, LOF signals may be too subtle or inconsistent for reliable detection. The study addresses the ongoing challenge of in-situ quality monitoring in metal additive manufacturing, where defects like cracks, porosity, and warping can compromise part reliability. AE sensing offers a low-cost, real-time method for defect detection, and artificial intelligence presents a powerful tool for pattern recognition within this data. Fast Fourier Transforms (FFT) and wavelet-based spectrograms were explored as CNN inputs, with testing performed on labeled AE images from SLM builds using 3 defined laser parameters. Validation of predicted defect types was carried out using metallography, microscopy and relative density measurements. The findings support the potential of AI-based AE monitoring for detecting porosity issues in SLM, while highlighting the need for further refinement in detecting LOF-related defects.
Included in
Acoustics, Dynamics, and Controls Commons, Manufacturing Commons, Other Materials Science and Engineering Commons, Structures and Materials Commons