Quality prediction in complex product development processes

A probabilistic approach to predict the expected quality of complex during all stages of the product development process is presented. The problem domain is modeled with the aid of a Bayesian belief network, wherein the nodes represent random variables that stand for the parameters, which have a significant impact on the product quality, and the links represent the dependency relations among these prarameters. The model is built up hierarchically, i.e. at the top level the dependencies among the main parameter classes that have an impact on the quality of the final product are modeled. Each of these parameter classes can possibly be influenced by other parameters on a lower level of abstraction. Each parameter on this level can in turn be refined the same way and so on. Depending on how detailed knowledge is available and how advanced the development is, a qualitative estimation of the expected product quality can be carried out at any stage of the design process.