Optimal inspection policies for sewer networks under resource constraints using partially observable Markov decision processes

Inspection strategies are an important part of asset management. Sewer networks use these strategies to optimize the expected state of the network and minimize the downtime. These networks are an important part of modern infrastructure and reduce risks of waterborne diseases. Therefore, the scheduling strategies have had to be optimized further. Without a self-announcing failure, difficult degradation patterns, and a long-expected lifespan, optimizing the maintenance strategy has been difficult. This paper presents a model based on the theory of the partially observable Markov decision process (POMDP), where the maintenance schedule is optimized. The schedule considers a predefined horizon and includes both rehabilitation and inspection. The value of inspection is modeled through posterior probabilities of the expected state space. The model is solved using mixed integer linear programming (MILP). The resulting schedules of the model are compared to simpler scheduling strategies. The differences in expected costs are used to determine if this model is able to construct an appropriate and efficient maintenance schedule for a single pipe.