Volumetric Data Analysis for Reverse Engineering and Solid Additive Manufacturing: A Framework for Geometric Metrological AnalysisGeng, Zhaohui (2021) Volumetric Data Analysis for Reverse Engineering and Solid Additive Manufacturing: A Framework for Geometric Metrological Analysis. Doctoral Dissertation, University of Pittsburgh. (Unpublished)
AbstractPoor geometric quality is one of the main constraints that hinders the wide adoption of reverse engineering (RE) and additive manufacturing (AM). RE models from a single scan will most likely generate inaccurate representations of the original design due to the uncertainties existing in individual parts and scanning procedures. On the other hand, metrological methodologies for AM significantly differ from those for the traditional manufacturing processes. Conventional statistical methodologies overlook these three-dimensional (3D) feature-independent processing techniques. In this dissertation, we develop a novel statistical data analysis framework---volumetric data analysis (VDA)---to deal with the uniqueness of both technologies. In general, this framework also addresses the rising analytical needs of 3D geometric data. Through VDA, we can simultaneously analyze the measured points on the outer surfaces and their relationships to acquire manufacturing knowledge. The main goal of this dissertation is to apply the proposed framework in multiple RE and AM applications related to their geometric quality characteristics. First, we demonstrate a novel estimator to increase the precision of RE-generated models. We built a Bayesian model with prior domain knowledge to model the landmarks’ uncertainty. We also proposed a bi-objective optimization model to answer the RE process-planning questions, e.g., how many scans and parts are required to achieve the precision requirements. The second major contribution is a study of tolerance estimation procedure for the re-manufacturing of legacy parts. We propose a systematic geometric inspection methodology for the RE and AM systems. Moreover, based on the domain knowledge in production-process design and planning, we developed methods to estimate empirical tolerances from a small batch of legacy parts. The third major contribution of this dissertation is to design an automated variance modeling algorithm for 3D scanners. The algorithm utilizes a physical object’s local geometric descriptors and Bayesian extreme learning machines to predict the landmarks’ variances. Lastly, we introduce the VDA framework to AM-oriented experimental analysis. Specifically, we propose a high-dimensional hypothesis testing procedure to statistically compare the geometric production accuracy under two AM process settings. We present new visualization tools for deviation diagnostics to aid in interpreting and comparing the process outputs. Share
Details
MetricsMonthly Views for the past 3 yearsPlum AnalyticsActions (login required)
|
![[feed]](/images/feed-icon-32x32.png){kind=icon}