Mobile Sensor Coordination and Bayesian Inference for the Localization of Contamination Sources in Water Distribution Systems

Localizing contamination sources in water distribution systems (WDSs) is crucial for determining mitigation strategies to maintain quality of water and prevent the propagation of pollutants to consumers. Various strategies have been proposed to solve the source localization (SL) problem by integrating sensor data and models, including deterministic and probabilistic optimization approaches. Regardless of the approach, SL has been primarily studied in the presence of online sensors installed in fixed locations within a WDS. In this work, we study the SL problem in the presence of mobile robots that are equipped with sensors that collect measurements as they travel through the WDS. The trajectory of the mobile robots in a WDS is controlled by determining the sequence of actions guiding the movement of the robots to maximize information gain and reduce uncertainty. We refer to this as the mobile sensor coordination (MSC) problem. We evaluate the performance of infotaxis, locally- and globally-greedy MSC strategies running alongside with probabilistic SL. We propose a time-dependent probabilistic model to estimate the source probability, which is updated based on mobile sensor measurements using Bayesian inference. The probability is updated as sensors move throughout the WDSs following an MSC strategy. Sensor data collection and sensor positioning are executed online in a centralized fashion. We compare the performance of the methods in terms of accuracy, running times, and localization times by generating various uncertain scenarios and varying the number of mobile sensors in each scenario. We test the proposed methods using a collection of WDSs. The results show that effective SL requires appropriate selection of an MSC strategy depending on the WDS properties and sets the foundation for performance evaluation of MSC in WDSs.