Demonstration of a process control methodology using multiple regression and tolerance analysis.

A methodology is presented that practicing engineers can use to improve the performance of systems with a large number of variables. The methodology is demonstrated in a plant that produces direct current motors and performance was measured by the final product reject rates. If the causes of poor system performance are not known, a system wide study is necessary that can determine the likely causes from a large number of potential variables (screening) and identify remedies that will result in acceptable system performance. The method must accomplish both as quickly as possible in a production environment without disrupting manufacturing operations. DOE based methods are inadequate because of the large number of variables involved, the time required to run a series of smaller experiments, the cost of system disruption to run experiments, and a production environment hostile to "tinkering". Methods that use observational data, such as multiple linear regression analysis (MLRA), are not often used because they indicate the problem, but not the remedy. A remedy consists of a change to the process, the specification, or the measurement system. The problem of developing remedies from observational data has not been extensively addressed. A methodology was developed that progressively screens out variables and identifies the processes most likely to be the cause of poor system performance. The speed at which this is accomplished is enhanced by collecting observational data from many variables simultaneously. The research involved applying step-wise MLRA as a first level screening tool on a plant wide scale, followed by process capability to distinguish between design and process problems and tolerance analysis to develop and evaluate potential remedies. Some results are: (1) Successful demonstration of MLRA as a screening device. (2) A conceptual framework that addresses consistency issues between measurement systems, product and process specifications, and process control. (3) A methodology that (a) identifies from a large number of variables the likely causes of poor system performance, (b) distinguishes between process and design causes, (c) evaluates the internal consistency of product and process specifications, (d) determines and evaluates quantitative specification or process factor values as remedies.PhDIndustrial and Operations EngineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/103077/1/9303741.pdfDescription of 9303741.pdf : Restricted to UM users only