Risk-based inspections enhanced with Bayesian networks

Abstract

The present paper aims to assess the usefulness of the integration of a qualitative risk-based inspection (RBI) procedure with the modelling of Bayesian belief networks (BBNs). In such a way, qualitative RBI, usually performed following standard procedures, might still be applied, even with the lack of some necessary data. Another benefit of the proposed method is the capability to include new factors in the evaluation, overtaking the rigid structure of the procedure. The qualitative RBI was chosen for its simple and clear structure, while the choice of the BBN formalism was made because of its flexibility and power to represent discontinuous variables. Bayesian formalism, moreover, allows a quick and easy extension of the model with additional variables. As a result, a BBN modelling a qualitative RBI was constructed. In the network, all of the RBI variables were included, as well as new external variables. An application for a chemical plant was tested in order to prove that BBNs can profitably model qualitative RBIs and can go beyond its rigid structure with the addition of new variables.