Theoretical and Conceptual Framework - Pattern Recognition for Reliability Assessment of Water Distribution Networks - page 28

and more stable simulations of real-life networks. The questions that can be raised in relation
to the tools for reliability assessment are:
(C.1) Are the existing numerical approaches appropriate for reliability analyses?
Applying hydraulic solver based on demand-driven approach makes the simulation
approaches practical i.e. applicable on a system of any size and complexity, provided that the
reliable data for the model building and the burst frequency can be extracted from the water
distribution company records. That is why the adaptation of this approach into a 'quasi
pressure-driven' approach, such as the one described in the works of Ozger and Mays (2003),
and Yoo et al. (2005) seems to be a good compromise between the reality and robustness of
the numerical algorithm. Nevertheless, the concept of emitter coefficients available in
EPANET software becomes more frequently used for pressure-driven demand calculations,
being pretty robust while simpler to code.
The commercial introduction of genetic algorithms (GA) opens interesting opportunities for
optimisation of network performance based on economic grounds that also include reliability
aspects. For example, particular design layouts can potentially be optimised to minimise the
reduction of demand during the failures in the system. Such a software tool would certainly
be instrumental in developing the methods suggested under (B.2).
(C.2) What is the real nature of pressure-demand dependency and its implication for demand
patterns?
In all kinds of pressure-driven demand considerations, the real nature of the dependency
between these two parameters is difficult to generalise. It is likely that a specific way of water
use and type of water appliances and outlets typically installed in one area contributes to this
relation. Secondly, the diurnal demand pattern applied in regular situations will almost surely
be distorted in the times of failures. It is very difficult to predict how much of this will be
caused by lowering of the pressures and how much by changing usual habits in water use,
which makes running of extended period simulations for reliability analyses rather
problematic. Consequently, the results like the one shown in Figure 2.6, would need to be
more rigorously scrutinised. In both of these cases, a thorough collection and analyses of the
relevant field data could be helpful to discover the pressure-demand patterns.
(C.3) Is accurate calibration of the models studying network reliability possible?
One of the difficulties in the reliability analyses is that the model is difficult to calibrate using
the field data. Water companies will hardly inflict a failure in order to be able to test its
consequences on the level of service; any failure will happen spontaneously i.e. mostly
unpredictably. Unfortunately, those networks that are mostly suffering from general
negligence are the same networks where poor data collection is typical problem. A thorough
monitoring of the system performance during any calamity is vital source of information for
computer modelling of network reliability.
(C.4) What steps of the reliability assessment process could become standard features of
commercial network modelling software?
In the absence of generally accepted reliability method, most of the applications from the
literature show that the simulation approaches are computationally intensive and automation
of the calculation process is a must, especially if networks above a few hundred nodes would
be considered. With the current speed of state-of-the-art PCs this is however not seen as a
serious constraint. Nevertheless, the commercial water distribution software hardly posses