NETWORK MODELS OF HIGH-DIMENSIONAL IMAGE PROFILES FOR CHARACTERIZATION AND DETECTION OF ADDITIVE MANUFACTURING DEFECTS
Open Access
Author:
Singh, Inder Pratap
Graduate Program:
Industrial Engineering
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
April 07, 2017
Committee Members:
Dr Hui Yang, Thesis Advisor/Co-Advisor Dr Saurabh Basu, Committee Member
Keywords:
image analysis process control defect detection quality control additive manufacturing wavelet analysis network model
Abstract:
The adeptness of additive manufacturing processes to produce parts of varying geometries and materials presents a challenging problem in terms of detecting defects and controlling the in-process quality of the part produced. To increase the in-process quality visibility, additive manufacturers are increasingly making use of high end imaging technology. Use of high dimensional images for in-process quality control poses significant challenges on traditional statistical process control practices for process monitoring. The second, mostly overlooked, problem is that of managing the large amount of data these high dimensional images bring with them. This thesis presents network models of high-dimensional image profiles for characterization and detection of additive manufacturing defects. Image profiles are first presented in the form of a dynamic network graph, which is then subsequently utilized to characterize complex additive manufacturing defects. Potts Hamiltonian approach is then used to detect network communities, which in turn form the basis of a multivariate control chart of the image stream. Statistics from this chart are used to detect the in-process defects occurring during the process. Experimental results on an image stream obtained from a laser sintering process shows that the methods proposed in this thesis can successfully detect defects and classify good and bad parts.
