Methods And Application Of Attribute Measurement System Analysis

Measurement System Analysis (MSA) is of vital significance in a Six Sigma project. Compared with Variable MSA, Attribute MSA has received relatively little attention, and also some key problems need to be solved.Therefore, Attribute MSA should be specially investigatedIn this PhD thesis, a comprehensive literature review on the research of attribute measurement system is firstly made, and then a detailed discussion on the evaluation and application of attribute MSA is given based on level of measurement, including nominal measurement system, ordinal measurement system, and binary measurement system.With regard to nominal measurement system, the model and detailed calculation method for several kinds of agreement index are discussed, and a comparative study is made using cross-classification table. A case study on the MSA in the sealing process of thermostat has validated the conclusions that are previous drawn.With regard to ordinal measurement system, a detailed analysis on the square weighted Kappa Statistic, linear weighted Kappa Statistic, six kinds of Intraclass Coefficients, and Kendall's Coefficients is first presented, and then the agreement characteristic of weighted Kappa Statistic, Intraclass Coefficients, and Kendall's Coefficients to ordinal measurement system is proved through a case study of fabric quality.With regard to binary measurement system, the determination of sample size is firstly discussed based on agreement analysis, including the number of parts, appraisers and trails. A new method of choosing the number of parts is presented through the investigation of the influence of the number of parts on the confidence interval of proportion of agreement. As for the number of appraisers and trails, based on the majority voting rule, a more feasible method to evaluate agreement within and between appraisers is proposed considering that the current standard is too rigorous.Finally, the influence of inspection error on the screening problem in the binary measurement system is investigated andthe formula of calculating the cut-off limit of screening variable is introduced.