Conclusions - Pattern Recognition for Reliability Assessment of Water Distribution Networks - page 274

How the change in node connectivity affects the levels of service? The results showed that the
connectivity measures alone can hardly be correlated to the reliability measures that are also
based on pipe geometry and hydraulic performance. The approach based on graph theory has
shown more consistent results of more simple measures such as the graph density (GD),
which correlate better than the more complicated ones e.g. the betweenness centrality (BS).
The difference between undirected and directed graphs, the latter coinciding with flow
directions in the pipes, analysed in this research has been insignificant. Although used mostly
for social networks, NodeXL has been user friendly add-on although lacking direct link with
EPANET software. The range of features enabling good visual presentation of graphs gives
impression that this MS Excel template can be used for further water distribution analysis, for
instance for network zoning, district metering areas, etc.
What is the effect of selected supply schemes, with or without balancing tanks in the system?
This question was only indirectly addressed within the economic analyses done for various
scenarios i.e. reliability levels. As it could have been expected, the supplying schemes with
more pumping have higher operational and lower investment costs. Those schemes with
balancing tank appear to be more reliable and less costly than those with direct pumping,
provided that the tank is properly designed and operated in combination with pumping
station. Because all hydraulic simulations were snap-shots, the balancing volume was just
roughly estimated, instead of being optimised. An extended period simulation would have
yielded more useful results for this purpose.
What is the link between the pressure levels and the effects of potential failure in the system?
The results point fairly good matching with the reliability i.e. resilience measures. The logical
response of the indices occurs after the demand and diameter increase, or having various
degree of network connectivity applied in the model. The pipe correlations of geometric and
hydraulic parameters show less consistent patterns. Different network categories have been
showing different level of correlations of particular parameters. GA-optimised networks
show generally better correlations between the volumes and flows, while the networks with
more buffers have better correlations of the hydraulic parameters involving friction losses. In
most of the cases, the results have also shown the trend of improved correlations after
gradually increasing the level of demand, but no firm conclusions could have been drawn.
Is there a link between the reliability and energy balance in the network? The network
pressure index values ( NPI ) seem to be generally more susceptible to nodal elevations (i.e.
the pressures) than are the pressure buffer index values ( PBI ) values. Oppositely, the PBI
values are more susceptible to the demand growth (i.e. the friction losses) than are the NPI
values. Additional buffer achieved by the increase of pipe diameters shows different effect on
the improvement of network resilience. The different form of the curves for the same
reliability index results from the presentation in different ranges of ADF avg .
What are the differences between the most economic design and the most reliable design?
Contrary to the assumption that 'the cheapest' means 'the least reliable', in the economic
scenario of high O&M costs and unfavourable loans leading to substantial annuities, the least
cost design scenario is likely to be more reliable than the one with increased costs for O&M.
In addition, the reliability of networks laid in topographies which allow substantial use of
gravity for water conveyance, appears to be more dependent on network capacity i.e. pipe
volume.