Real-time monitoring of water distribution networks enables water quality (WQ) control systems to trace the evolution of disinfectants and contaminants within the network. To that end, WQ sensors are typically employed within the network to achieve an observable system where sensor data provides network-wide situational awareness. However, the cost and practicality of WQ sensors installation requires the sensors to be optimally placed at certain locations of the network. Prior research has approached solving the optimal geographic placement of WQ sensors from two different standpoints. The first considers transforming public health metrics related to contamination events into the optimal problem formulation. The second takes on a computational approach that formulates the problem by considering network-wide observability-based metrics (i.e., utilizing the sensor data to recover or compute the overall WQ state of the network). However, such methods have typically adopted the use of simplified single-species decay and reaction models of chlorine. Consequently, the resulting sensor placements are not robust to changes in the network’s hydraulic profile or advanced WQ models that capture far more than chlorine decay. To that end, in this work we introduce an optimal sensor placement framework that addresses the literature’s limitations. The underlying water network model is based on a multi-species reaction dynamics representation; it enables contaminant reactivity modeling. The proposed sensor placement framework offers the following: (i) a robust solution towards fluctuations in water demand patterns; (ii) a scalable algorithm that enables its applicability to large-scale networks. A comprehensive case study is provided on benchmark water networks with varying hydraulic conditions and network setups. The sensor placement framework is solved by considering several system observability metrics. This offers different insights that water system operators can consider.
