Study On Variable Measurement System Capability Assessment

In some real variable measurement processes, quality characteristics of a productor process are two or more related variables. It’s an urgent need to address theproblem how to assess the bivariate or multivariate measurement system capability.Measurement system capability assessment for simple linear profile is to evaluate themeasurement system for the product or process quality characterized by a simplelinear profile, and is a hot research point in quality engineering.In this paper, based on the practice, for specific problems, the theoretical methodsand techniques for comparing capability of two measurement systems, assessingbivariate, multivariate and simple linear profile measurement system capability aresystematically studied using generalized inference, projection pursuit and avector-based methods. These studies have a very important significance formanufacturers to analyze the measurement system capability and to continuouslyimprove the product or process quality.This study firstly proposed a method based on generalized confidence intervals tocompare the capability of two measurement systems that measure the same qualitycharacteristic. Generalized confidence intervals for the indices are built usinggeneralized inference. Computer simulation is conducted to evaluate the validity ofthe method. The results reveal that the proposed approach based on generalizedconfidence intervals can effectively compare capability of two measurement systems.Secondly, as for the measured quality characterized as bivariate in themeasurement process of manufacturing enterprises, the evaluation indices of bivariatemeasurement system capability are built in this paper. Due to measurement values,true values and measurement errors of the quality characteristics followed bivariatenormal distribution, the univariate measurement system capability indices can beextended to bivariate measurement system, and bivariate measurement systemcapability indices are constructed based on the ratio of the ellipse areas, according tothe statistical properties of the bivariate normal distribution.Next, two methods for multivariate measurement system capability evaluation areproposed in this paper. One of the methods is to build multivariate measurementsystem capability indices based on the vector, according to multivariate normality oftrue values, measured values and measurement errors of quality characteristics. Another approach is to apply dimensional reduction methods, such as principalcomponent analysis or projection pursuit method to reduce dimensionality of themultivariate related quality characteristics, and then assess the measurement systemcapability using common measurement system capability indices.Finally, this study proposed a method to assess measurement system capability forsimple linear profile. Each simple linear profile can be fitted through the least squaresmethod, and then select two predicted values of the profiles as two related variables.The bivariate evaluation indices are constructed by bivariate analysis method to assessthe simple linear profile measurement system. How to apply the proposed method toassess simple linear profile measurement system is illustrated through an example ofpressure spring measuremeng system.