Architecting Very Large, Complex Bayesian Network Simulations for Practical Airworthiness Risk Assessment Applications

Abstract

While constructing practical Bayesian Networks (BN) of large, complex systems that are computationally tractable continues to be a significant challenge within the expanding BN modeling and simulation field, this paper advances methods that are changing the BN modeling paradigm from simplifying these models to reducing them by employing parent divorcing and generalized BN model development methods. We illustrate real world implementation of these concepts by reducing a large, complex cyber-physical risk assessment BN model into smaller, manageable ones able to be efficiently simulated, while identifying a newly structured approach toward appropriately weighting the integrating factors to be validated at the higher level. We apply d-connectedness techniques to maintain backpropagation and overall model integrity present in the larger, complex BN model, and use cause consequence and definitional/synthesis idioms to construct an Airworthiness BN risk assessment model, based on MIL-HDBK-516C, by implementing series and parallel networks of smaller models that simulate separately. Richer simulation results are enabled by extending the degree of model complexity able to be created and efficiently simulated within the practical capabilities of an existing commercial tool.