Incorporating Asset Interdependency in Risk Assessment Modelling: A Bayesian Neural Network Approach

Abstract

Risk assessment is essential for asset management, particularly in accounting for interdependencies between assets. This paper introduces a framework that analyses risk at both asset and system levels. A Multinomial Regression (MR) approach is employed to predict the probability of performance for individual assets, while Bayesian Neural Network (BNN) are used to model interdependencies between assets. The BNN method is well-suited to handling Boolean, categorical, and numerical inputs, and it effectively captures uncertainty in performance predictions. The framework is validated using data from the tracks and drainage systems of four railway routes in the UK. The results demonstrate that this approach is a reliable tool for asset performance evaluation and uncertainty quantification, offering valuable insights for improving asset management practices.