|M.Sc Student||Gideon Schramm|
|Subject||Methods for Detection Contamination in Drinking Water|
|Department||Department of Agricultural Engineering||Supervisor||Mr. Sinai Gideon (Deceased)|
A model was developed here that computes all feasible flow patterns - FFPs of a water distribution network for steady state. A flow pattern (FP) consists of all flow directions in a network that exists at the same time. The model consists of two main stages: in the 1st stage the model produces all possible FPs, in the 2nd stage it checks if conservation of fluid in all nodes and energy in all loops (Kirchoff's 1st and 2nd laws) are met in every FP.
Contaminant propagate in a network following flow direction, thus by identification of the flow pattern we can predict the contaminant propagation in the network from the monitoring station that detected the contamination spreading. Detecting exact injection point of the contaminant is a procedure that involves a search upstream of the monitoring station that detected a contamination incident. So if the flow pattern is known whiles that incident occurs, the pollution domain (pollutant ultimate spreading area) downstream of the monitoring station is known right away as well as the detection domain (nodes suspected as pollution injection points) upstream of the monitoring stations. The only input data for such searches are network layout and consumers/sources location.
Three real networks were studied: Central Arava regional network, Ashkelon region network both in Israel and Fairfield city, Ca, USA. The networks analyzed consist of 38 up to 111 nodes respectively and 39 to 123 links. The model application upon the example networks decreased the number of FPs several order of magnitude. The FFPs data base found was used to find all possible flow paths for each node. These paths were used later to define the detection and pollution domains. These domains were also used to produce an occurrence matrix of relative frequency of all pollution events possible!
A further implementation was made for finding the exact injection location of a pollution incidence in Fairfield network. The detection domain of monitoring stations produces a maximum suspected injection nodes list. The procedure further decreases this list until the pollutant injection node is detected by changing the current FFP to another and to make cross examination of the monitoring stations detection.
Although the model developed here is for steady state and conservative substances it represents a relatively novel approach for multiquality network analysis. It can inspire future works dealing with un steady state conditions and non conservative substances.