Harnessing Expert Knowledge: Defining Bayesian Network Model Priors From Expert Knowledge Only—Prior Elicitation for the Vibration Qualification Problem

Abstract

As systems become more complex, systems engineers rely on experts to inform decisions. This challenges systems engineers as they strive to plan activities such as, qualification in an environment where technical constraints are coupled with the traditional cost, risk, and schedule constraints. Bayesian network (BN) models provide a framework to aid systems engineers in planning qualification efforts with complex constraints by harnessing expert knowledge and incorporating technical factors. By quantifying causal factors, a BN model can provide data about the risk of implementing a decision supplemented with information on driving factors. This allows a systems engineer to make informed decisions and examine “what-if” scenarios. This paper discusses a novel process developed to define the probability distributions, or priors, in a BN model based primarily on expert knowledge supplemented with extremely limited data (25 datasets or less). The model was developed to aid qualification decisions—specifically to predict the suitability of six degrees of freedom vibration testing for qualification. The process defined the model priors primarily with expert knowledge in an unbiased manner. Validation of the model provided evidence that this novel process might be an effective tool in harnessing expert knowledge for a BN model.