Predictive maintenance for critical infrastructure

Abstract

The sustainability of many critical systems, such as water transmission networks or electrical grid, requires predictive maintenance strategies to prevent malfunction of components. These strategies typically use a troubleshooting model to suggest the components that are most beneficial to replace. This paper suggests a new dimension, which considers not only replacement costs and failure probabilities of components, but also adjacency of the components being replaced. We propose a model in which replacing adjacent components is often beneficial, because they can be replaced in a single replacement action. This helps minimizing costs known as overhead costs, which include the cost of sending a team to perform the replacement, the disruption to service during the replacement, and more. We propose several algorithms and AI techniques to suggest economical replacement methods. Evaluation on a real-world water transmission network shows that near-optimal solutions return a solution very fast, which is very close in terms of expected cost to the optimal solution.