Rating System Analysis in Quality Engineering Based on Fuzzy Similarity
Magdalena Diering, Krzysztof Dyczkowski and Adam Hamrol
The article describes a new measurement system analysis (MSA) methodology for qualitative characteristics. In this approach several features of the product are rated. Each feature can be expressed with imprecise data, in a nominal or an ordinal measurement scale. The goal of the methodology is to study raters ability to describe features in relation to the specification and customer requirements; also to gain information about the appraisers level of agreement in assessing products. The method is based on the novel fuzzy similarity coefficient SC, and Gwet’s AC1 coefficient. Proposed novel method is the basis for the authors to build an expert assessments supporting system based on fuzzy models. To illustrate the methodology, we give a practical example based on visual inspection in the production process of diagnostic catheters. The analysis let to decide if raters’ and expert’s knowledge (know-how) about the features and the product is a reliable source of information – it occured that appraisers are able to detect the product defects (if appears), but raters and the expert use different set of rules to classify the object as a good or defective one (it means that set of rules to make the final decision about the product needs to be improved).
Keywords: Qualitative characteristic, measurement system analysis, MSA, level of agreement, inter-rater reliability, similarity measure, cardinality of fuzzy sets, diagnostic catheters