An expert multivariate statistical control system for manufacturing processes

This dissertation developed an Expert Multivariate Statistical Control (EMSC) system architecture for the control of manufacturing processes with multiple process parameters. In the development of the EMSC, multiple tests of instability for multivariate process mean and dispersion, respectively, were developed to provide more effective detection of process instability and to assist in process diagnosis. The EMSC concept integrates univariate and multivariate Statistical Process Control (SPC), expert multivariable control, and fuzzy set theory to solve the problems found in traditional multivariate SPC functions of monitoring, analysis, and diagnosis.;In monitoring, data about all the process parameters is collected. Then, sample statistics for both univariate and multivariate procedures are calculated and displayed in their respective control charts for process analysis.;In analysis, the EMSC system accomplishes the instability detection by applying fuzzy set theory in both univariate and multivariate instability tests to assess a fuzzy instability index (in the interval of (0,1)) for each test. All the indices are combined to form an overall instability index for each subgroup at a certain time. If the process instability is detected, the interpretation of univariate sample statistic patterns is initiated by applying the univariate instability tests along with the interactive dialogue between the operator and the EMSC system.;In diagnosis, a knowledge base representing the fuzzy relationship between process symptoms and special causes of error is constructed by incorporating the knowledge of process experts and univariate and multivariate SPC expertise. The EMSC applies four steps of fuzzy relational products between observed process symptoms and the diagnosis knowledge base to infer the possible causes of error.;The Structured Analysis and Design Technique (SADT) was used to model the architecture of the EMSC system. An example has been used to demonstrate the EMSC concept.