Dynamic Bayesian Networks for Reduced Uncertainty in Underwater Operations

Abstract

This paper presents a novel framework for modelling dynamic Bayesian belief networks (BBNs) for online risk assessment in underwater operations. Existing frameworks spans from commercial software with restricted code access to non-profit open source frameworks. Frameworks with restricted code access often provides general user interfaces and visualization tools, while open source frameworks provides access to code for developers. The model presented in this paper pursues a best of both worlds scenario, where the model implementation should be uncomplicated while also providing visualization and verification to provide the user with a clear perception of the implemented model. The presented method is an expansion of the Bayesian model of the pomegranate python library, and simplifies the procedure of building, verifying and utilizing BBN models. The method is applied to a conceptual design of an underwater scenario case study with a model for an underwater vehicle manipulator system.