Combining process knowledge for continuous quality improvement

Quality improvement efforts often make use of various mathematical models that describe the relationships between quality characteristics and process factors. Such models typically come from a variety of sources: experiments, theory, on-line data analysis, expertise, and other process documents. These sources of knowledge are often distinct and separate, often yielding models with slightly different predictions, having different precision and validity. In this paper we explore alternatives in which different mathematical models can be integrated together into a single prediction that takes into account both model validity and model variability. Some guidelines for establishing and quantifying model validity are presented. The approach is demonstrated within the context of predicting surface finish in a machining process.