ABSTRACT
Maintenance of road drainage systems requires efficient use of public funding and resources. Current periodic gully pot cleansings can be improved through risk-based decision-making tools that use inspection and historical data. This paper proposes a framework to define gully cleansing regimes making use of stochastic Petri nets and Machine Learning. The stochastic Petri net model is used to predict gullies’ sediment level ranges over time based on historical data. Machine learning techniques are used to classify gullies and overcome data absence. The probabilistic distributions used to predict sediment levels are also utilized to derive the probability of reaching a limit state given an assigned cleansing frequency. The framework enables a risk-based analysis to optimize maintenance costs. The results show that the stochastic Petri net model is suitable for representing sediment level condition states and the risk-based tool is useful for maintenance decision-making.
