Use of computer vision in μ-plasma metal additive manufacturing process for defect detection

Abstract

Deposition defects such as discontinuity, non-uniformity, waviness, under-dilution, over-dilution, porosity, balling, spattering, and poor surface quality in the metal additive manufacturing (MAM) processes significantly influence properties and performance of the fabricated product. Waviness and non-uniformity are characterized by unacceptable variations in deposition height and width respectively along the deposition length. The deposition defects reduce strength, toughness, fatigue life, and wear resistance thus making the product unsuitable for the intended application. They lead to metrological inaccuracies in the final product and can even lead to rejection of the entire thus necessitates their expensive post-processing. Therefore, real-time detection and elimination or minimization of the deposition defects is crucial to performance and service of the MAM products and for elimination of the expensive post-processing. This research proposes computer-vision based defect detection methodology to detect discontinuity, non-uniformity, waviness, under-dilution, and over-dilution in μ-Plasma Metal Additive Manufacturing (μ-PMAM) fabricated single-layer depositions of different biocompatible materials using hue saturation value (HSV) based color segmentation and centroid distance and the trained YOLO models. High-quality videos were recorded of single-layer depositions of Ti6Al4V, 63Co29Cr4Mo4Ti, and SS 316L materials using a high dynamic range (HDR) camera mounted on 5-axis CNC machine of μ-PMAM process for different parametric combinations.