Heterogeneous Measurement System for Data Mining Robotic GMAW Weld Quality

Abstract

During robotic welding, several streams of heterogeneous data can be collected. To gain a systemic understanding of the welding process, these data streams have to be combined precisely and accurately, especially if our goal is to develop online weld quality assessments. Establishing correspondence among temporal and spatially based data is a nontrivial effort. This article presents a data collection system using a novel methodology for establishing correspondence across multiple data sources of robotic gas metal arc welding for objective quality assessment. First, correspondence between the weld process data and the resulting weld required time synchronization and spatial alignment. Second, an objective weld quality extraction technique that assigns quantitative measures at a resolution of 1 mm of linear weld travel was developed to evaluate weld quality. Specifically, in addition to developing a method for objective weld profile assessment, we developed an objective analysis of radiographic data for the occurrence of subsurface porosity to assess defects and demonstrate how to objectively quantify the occurrence of surface porosity. While some aspects of this paper have been addressed individually and separately by other research, this paper presents an integrated approach to these operations for a wide variety of weld data types and develops objective weld quality metrics that can be used for machine learning of weld quality for robotic welding.