Macroscopic examination of welds: Interlaboratory comparison of nominal data

Abstract

A statistical technique is developed for interlaboratory comparisons of the macroscopic examination of weld imperfections caused by failures in the welding process. This technique allows comparisons of nominal data with more than two categories which are influenced by two factors (variables). The development is based on two-way categorical analysis of variation (CATANOVA). Decomposition of the between-laboratory variation for a cross-balanced design by these two factors (e.g., laboratory and technician experience) and their interaction, as well as by the categories of the response variable, is proposed. The developed technique was applied to an interlaboratory comparison using 12 images/photographs of welds with five categories of imperfections as the test items for macroscopic examination. The items were distributed to three participating laboratories and examined by both experienced technicians and novices. Analysis of the obtained data showed a consensus between laboratories and between technicians and no interaction between them. It was found also which categories of imperfections were more difficult to identify than others. This technique is applicable for other nominal properties, and can be adjusted for proficiency testing (PT) of laboratories participated in the comparison.