Delay Modelling in Complex Industrial Systems via Bayesian Networks

Abstract

In most systems, delays may severely degrade performance. That is why they are monitored, controlled, and minimised as much as possible. However, doing so in complex industrial systems is particularly challenging, due to their intrinsically non-linear dynamics. This paper proposes an approach to model and analyse delay in such complex systems, seen as an instance of system of systems (SoS), by building on developments in Bayesian Networks (BN). The proposed BN considers the availability of different subsystems as an important cause of system and SoS delays. By conditioning the availability of systems to operational choices, such as maintenance strategies, it becomes possible to analyse their impact on the total SoS delay. A container terminal is used as an example case to implement and test the proposed BN approach. Multiple simulations were generated based on a combination of different external factors (e.g. weather conditions), as well as internal choices (e.g. maintenance strategies). Compared to other simulation methods, like discrete event simulation, this approach is less data-intensive and computationally demanding, allowing for the rapid evaluation of multiple operational choices and their impact on the system and SoS delays. For instance, we observed that an expected reduction of up to 13% in the container terminal delay could be achieved by implementing a weekly preventive maintenance policy, due to an improvement in the availability of all systems.