Times are displayed in (UTC-07:00) Pacific Time (US & Canada)Change
Session: 07-01: NDE for Additive Manufacturing
Paper Number: 98319
98319 - Comparison of Flaw Detection Algorithms Using Simulated X-Ray Computed Tomography Ground Truth Data and Evaluation Metrics
X-ray computed tomography (XCT) is useful for detecting internal material discontinuities such as pores especially in additively manufactured parts. A flaw detection algorithm is applied to segment the detected pores into binary masks for quantitative analyses and automated inspections. Pore size and shape parameters can be derived from these binary masks, which may be used for part acceptance tests. To evaluate and compare accuracies of XCT flaw detection algorithms, we developed a phantom with known ground truth pores. Pores of varying sizes were distributed near the part surface and interior of the part to represent near surface pores and gas pores found in additively manufactured parts. The approach of using voxelated pores for simulation eliminated uncertainty associated with generating voxelated ground truth images. In this study, we compared three different algorithms and evaluated the segmentation results using various semantic and instance segmentation evaluation metrics at different user-defined parameter values. Pore size error plots and probability of detection (POD) curves were also estimated and compared. Two user-defined local thresholding parameters were optimized through a design of experiment (DOE) based on various metrics, and the optimization results are compared. We will also discuss our plans for developing future datasets for XCT defect detection challenge we plan to host.
Presenting Author: Felix H. Kim National Institute of Standards and Technology
Comparison of Flaw Detection Algorithms Using Simulated X-Ray Computed Tomography Ground Truth Data and Evaluation Metrics