Statistical Modeling of a Fluorescent Tube Coating Process

Abstract

The coating of a fluorescent lamp with fluorescent paint is an example of a complex industrial process. Improved control of this process could lead to reduction in the cost of producing a lamp. Modeling the process is necessary for improved control. As a first step. a study of the coating process at the Westinghouse Fairmont Works in Fairmont, West Virginia has been made. The study included two criterion, or dependent variables and 12 predictor, or independent variables. Analysis of the study data has produced a linear regression model with five independent variables which accounts for 58% of the variation in coating thickness. Also, a set of linear classification functions has been found which correctly classify 92% of visuial defects from 12 input variables, using the training data. These preliminary models have been used to design a controlled experiment. The controlled experiment will allow the significance of seven independent variables in determining optical density and visual defects to be established conclusively.