Assignment # 5

Article Reference:
Jonathan W. Berry, Lisa Fleischer, William E. Hart, Cynthia A. Philips, Jean Paul Watson, Sensor Placement in Municipal Water Networks, 2005, Journal of Water Resources Planning and Management

Summary:
The paper presents a model for optimizing the locations of the sensors in the municipal water networks for detecting the injected contaminants using the mixed integer program.

As already studied, the importance of estimating the water quality in the distribution system is essential, therefore placement of the sensor in the water system is of greater importance. The recent concerns of the terrorist attacks may affect the public belief on the water supplies. Therefore EPA is working with various community water systems towards water safety and security, real-time early warning systems (EWS) is one of the methods to monitor and identify any unusual threat the water supply system thereby getting appropriate time to handle the adverse impacts. EWS still uses conventional water quality monitoring which considers the changes in a specific water quality parameter to infer the presence of a contaminant. There are many issues regarding usage of EWS which needs to be researched yet.

The present study is carried out to deploy a sensor within a water distribution system. An effective deployment of sensor would take into consideration the maximum coverage of the network for the monitoring of the contamination. To optimize the location of the sensor there main objective would be to minimize the cost of the deployment and maximize the coverage of the sensor. But such sensors would not provide warning of the initial contamination. Present paper focuses on the optimization of the sensor placement based on minimizing the fraction of population being affected by the contaminants. Contamination across the network is modeled as the probability distribution across the junctions. Paper takes two synthetic data sets and one real time example to illustrate the placement of sensors.

Authors have made few simplified assumptions which ignore temporal effects, concentration effects and health impacts therefore this model may be more suited to the system where large volume of contaminant flow rapidly through the network. In the studies carried out earlier they have modeled the placement of sensors based on demand coverage, volume of the contamination and contamination time, the present study models the system considering the fraction of the population who are going to be effected by the contamination in the system. System is modeled as a single point attack. As the location of the attack is just an assumption therefore the sensors are places based on set of weighted attack scenarios. Demand in the system is assumed to be of a fixed set of pattern and the water flow analysis is carried out using EPANET. As we have seen in Lee and Deininger each node is considered as the fraction of the total demand, here we consider each node to be the fraction of population being affected. The population considered is not based on the demand. All the attack scenarios are defined as the probability distribution which is based on the expert opinions.

The model is optimized based on mixed integer programming (MIP) which minimizes the total population being affected by the contamination. Different constraints considered are (1) if the node is attacked directly then it is contaminated, (2) a single sensor covers the flow in both the directions (3) About contamination propagation (4) limit of the total number of sensors.

The authors carried out the optimization of sensor placement on the two data set examples from EPANET and the real world data set. The study is carried by dividing the total system into different categories, considering different attack scenarios, flow patterns and limiting the availability of the sensors. The study was majorly concentrating on how the uncertainties in data, population densities and attack risks affect the optimal solutions and optimal location of the sensors in the system. For studying the affect of the uncertainties in the solution, they have modeled the problem using different set of noise levels where they have considered different values for each element of population density and the risk probabilities, which sum up to total population and risk probability summing to 1. The optimal sensor configuration is based on the distance measure between two sensors; this is compared with the baseline (no noise level). The distance is defined as the number of nodes which are covered with shortest path between the two pipes in the network. The authors illustrated how the uncertainties do not have much of an effect on the optimal sensor placement and the configuration.


Discussion:
With the previous paper which discusses on the methodology of how we can optimize the location of the sensor, this paper gives more insight into the topic by explaining different assumptions, variable and constraints involves, and sensitivity analysis carried out which might affect the solution. The assumption of considering only one set of flow pattern according to me might not be the real world problem. As different distribution systems have different flow patterns and different source of supply. It would be interesting to do research on how multiple attack points be modeled and optimized. In the present study they have considered the cost or total number of sensors, as the one of the constraint, considering about how the optimal solution may alter if we considers time as the main constraint, which means we model the problem in a way that we require the contamination detection or source identification within a given time limit.