Augmented Reality With Digital Metrology For Assembly
1 online resource (116 pages) : PDF
University of North Carolina at Charlotte
This dissertation aims to create and demonstrate a fast and inexpensive quantitative dimensional inspection system for industrial assembly line applications that can detect position errors of assembled components on a scale of 1 mm and larger. The researched and developed data acquisition and the computational pipeline is presented. Position error detection of 1 mm and higher is demonstrated on a 40-mm high and 8mm radius post welded onto a 30 mm by 30 mm steel plate.Data acquisition is performed using an open-source photogrammetry architecture to gather a 3D point cloud of the assembled part. The photogrammetry architecture involves a structure from motion (SfM) pipeline to obtain a sparse point cloud, and a depth map merging method is used to generate a dense point cloud. The component's position on the assembly is calculated by comparing the point cloud with the CAD model. A method using the iterative closest point (ICP) algorithm establishes a global coordinate system for the data to align with the CAD model. Once the global coordinate system is established and aligned, the position of individual parts is estimated relative to the reference. This method was able to identify a 1-mm and larger position error of a post, as described above, welded onto a steel place. A partial uncertainty evaluation shows that the position error can be estimated with uncertainty no better than several fractions of a millimeter. Uncertainty contributions can be divided up into three groups: i) random noise and possible bias in the acquired point cloud data, ii) the actual shape of the component compared to the ideal (CAD model) component shape, and iii) data processing choices. The investigations into these aspects showed that, with enough points in the reference, a down-sampled measured data set, and a low ICP threshold, the limiting factors in the uncertainty come from spatial bias due to 3D reconstruction and object surface roughness. Point-wise random noise in the measured 3D point cloud was also investigated with commercial software (PolyWorks) and open-source algorithms and found to contribute negligibly to the combined uncertainty compared to the uncertainty caused by the object shape and surface texture. These effects require more work to estimate a comprehensive combined uncertainty. The pipeline was made user-friendly by creating an augmented reality (AR) application. This application detected the component of interest in a live video feed and overlayed the position error information. Object detection was done by creating guide views of the CAD model.
Optical Science & Engineering
Morse, EdwardFalaggis, KostaSmith, Stuart
Thesis (Ph.D.)--University of North Carolina at Charlotte, 2022.
This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s). For additional information, see http://rightsstatements.org/page/InC/1.0/.
Copyright is held by the author unless otherwise indicated.