Design Guidance Using Simulation-Based Bayesian Belief Networks

In our work, the task of complex computer-based system design optimization involves exploration of a number of possible candidate designs matching the optimisation criteria. However, the process by which the possible candidate designs are generated and rated is fundamental to an optimal outcome. It is dependent upon the set of system characteristics deemed relevant by the designer given the systems requirements. We propose a method which is aimed at providing the designer with guidance based upon description of the possible causal relationships between various system characteristics and qualities. This guidance information is obtained by employing principles of multiparadigm simulation to generate a set of data which is then processed by an algorithm to generate a Bayesian Belief Network representation of causalities present in the source system. Furthermore, we address the issues and tools associated with application of the proposed method by presenting a detailed simulation and network generation effort undertaken as part of a significant industrial case study.