Abstract
The paper presents a classification methodology for determining geometrically good and defective bead segments for the Wire Arc Additive Manufacturing (WAAM) process based on optimally determined thresholds of three parameter metrics: namely, the range of curvature, range of sound power, and maximum height. We show how a heuristic search can be performed to systematically identify the suitable thresholds for the various parameter metrics to classify bead segments. The heuristic search for each parameter metric is performed by finding a threshold value that minimizes overlapping area between the Kernel Density Estimation (KDE)s formed by the distribution of the good and defective bead segments. We combine these thresholds to yield a three-dimensional threshold window for identifying geometrically defective bead segments. This approach is helpful for dataset labeling that overcomes the visual inspection labeling used in the literature. Experiments are conducted on the Inconel 718 material, with 660 bead segment data. We verify the results of our threshold selection using real bead scans and evaluate its mean squared error (MSE) with the parabolic bead model commonly used in literature to model bead profiles. We find that our proposed approach can separate the starting and ending segments of the beads well and identify geometrically defective bead segments between them. We also observe that the MSE of geometrically defective bead segments' is much higher than the good segments, demonstrating the validity of our approach to classify geometrically defective bead segments.