Nodal sensitivity and sensor location in hydraulic network

The quality of hydraulic model calibration depends on sensor measurements. More field information gives a better calibration, but data collection is a costly process. For this reason, ineffective measurements should be avoided and it is important to use a method for designing data collection that will give the most useful information. Some methods for sampling design of water distribution networks were proposed. The simplest method assumes the nodal sensitivity of the model outputs versus the unknown calibration parameters as a means to decide where to make measurements. Comparing this method with more sophisticated procedures, it seems effective and less costly. Its major problem is that computation of sensitivity coefficients requires some values to be assigned to the unknown calibration parameters. In this paper the dependence of the sensitivities upon the roughness values and the loading condition used for their computation is investigated. Considering two sample networks the results show the independence of nodal sensitivity distribution from both the assigned loading condition and pipe roughness. This implies that nodal sensor location can be obtained also by using a fairly inaccurate estimate of calibration parameters. The above observations suggest that nodal sensitivity distribution could represent a sort of network signature for design and management purposes