Data Fusion for Additive Manufacturing Process Inspection

Open Access
Morgan, Jacob
Graduate Program:
Electrical Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
March 13, 2019
Committee Members:
  • Richard L. Tutwiler, Thesis Advisor
  • William Evan Higgins, Committee Member
  • Edward William Reutzel, Committee Member
  • data fusion
  • additive manufacturing
  • computer vision
In-situ monitoring of the powder bed fusion additive manufacturing (PBFAM) process is a rapidly expanding area of interest because it offers insight into process physics and is potentially a lower cost alternative to current post-build nondestructive inspection. Ultimately, sensor data may be used as feedback for a real-time fault remediation system. However, it is unclear what defects look like in the sensor data and multiple modalities cannot be used together because they are in arbitrary frames of reference. The goal of this thesis is to present a framework for automatically registering in-situ sensor data to post-build inspection data. This will enable defects found in the post-build inspection to be mapped to the sensor data to serve as a ground truth for developing automatic defect recognition (ADR) algorithms. In this work, high resolution images and multispectral point sensor data collected during the build are registered to a post-build computed tomography (CT). These sensing modalities can be thought of as 2D raster data, 2D point cloud data, and 3D raster data respectively. A unique optimization approach for registering each modality to a common frame of reference is presented. The process is automated so that large datasets may be generated for use in developing future ADR algorithms. Voids that are clearly visible in the CT can be mapped into the in-situ sensing modalities.