Abstract
A strategy is proposed for the identification and quantification of sources of variation in a manufacturing process. The strategy involves six steps: identification and selection of factors, model selection, design of the experiments, performing the experiments, estimation of sources of variation, and finally, interpretation of the results. This strategy helps in finding those factors that contribute mostly to the total variation apparent in analysis results due to the production process itself, sampling, and analysis of samples. The strategy is then applied to a case study in which sources of variation in steel analysis are identified and quantified. The study develops mixed (random and fixed) effect models for the three phases of steel manufacturing—stirring, tundish, and mold. The models show that differences between spectrometers can have an important influence on the total variation apparent in the final analysis results.
Acknowledgment
The authors are indebted to Mart Schouten for analyzing the samples. The authors thank Bert Snoeyer and the Research and Development Department for their cooperation in sampling the process.
Notes
aIn general, the maximum likelihood estimators do not take into account the loss in degrees of freedom resulting from the estimation of the fixed effects, and, hence, they become biased. The restricted maximum likelihood estimators take into account this loss of degrees of freedom.