Layering for leakage
More sophisticated use of existing data can deliver network efficiencies. Frank Rogalla of Aqualia invites TaKaDu's Guy Horowitz to share his expertise on how this can be put in place
The water network across England and Wales has more than 335,000km of mains, connected to almost 24m properties. Distribution pipes are made from a wide range of materials – cast iron, ductile iron, lead, modern plastics – all of which can leak.
Even though Ofwat reported in its 2009-10 Service & Delivery report that leakage is down by 35% since its peak in the mid-1990s, it seems almost impossible to reduce leakage levels to zero. The latest 2009 price review by Ofwat accepted UK water companies’ plans to reduce leakage to 3,245Ml/d by 2014-15.
While the annual improvement in the reduction of water losses peaked at 2% in 2008, it settles at ten times less in 2014 – stabilising in an overall loss of just under 10l/d/m of mains nationwide. In addition, trials with smart meters revealed that domestic supply pipes are leaking significantly, estimated to be around 25% of total losses. More than 4% of properties have leaks on the customer side of the meter, and while companies achieve leakage reductions on their distribution pipes, losses n the overall pipework can sometimes increase.
To control their networks, water utilities throughout the world have installed increasingly sophisticated SCADA systems, collecting near real-time data from sensors which measure flow and pressure at key network points, as well as other data, such as reservoir levels, various water quality indicators and valve positions. Most utilities use these systems for ongoing operational needs, making little use of the huge wealth of archived past data accumulated in their systems.
To reduce the inefficiency of water utility operations, there is now an opportunity for a layer of applications built on top of the increasingly sophisticated data collection arrays. Such applications may improve core utility activities such as water loss control and infrastructure monitoring, pressure management, and network design and planning.
Together, these emerging solutions are creating the smart water network, more efficient and better controlled, by virtue of relying more directly on water network data. With the increased instrumentation and telemetry of water distribution networks, a new layer of smart data applications has become possible, including a “smart” structure of data-driven components which helps operate the “dumb” or data-less physical layer of pipes, pumps, reservoirs, and valves.
A key component of the smart water network is water infrastructure monitoring. There are always complex events unfolding in the water network (such as leaks), often unseen, but with a variety of immediate effects and cumulative effects. Collecting and using comprehensive data about water network operations offers the promise of better operations through better knowledge and tighter control of the network’s extensive and complex assets.
Just as communications protocols or computer systems are often described as layers with distinct functions, with interaction typically occurring between adjacent layers, one can suggest a layer model for water networks. Each of the layers in the model can be made more intelligent, and the network as a whole can become smarter by applying the right solutions to the relevant layers.
Layer 1: Physical
This is comprised of the equipment enabling the distribution and delivery of water along the network, mainly the ‘wet’ components which deal with water. Pipes, pumps, valves, pressure reducing valves (PRVs), reservoirs and delivery endpoints are all part of this layer. These are data-less elements, that typically perform mechanical, hydraulic or chemical functions. While the physical layer does not have data interfaces, it can be controlled using data collected in the next layer – sensing and control.
Layer 2: Sensing and control
This is comprised of equipment and instruments that measure parameters of water delivery and distribution (such as flow, pressure, water quality parameters, reservoir levels, water temperature, acoustic information and more) and devices enabling the remote operation of the network (such as remote-controllable pumps, valves, and pressure-reducers). The sensing and control layer is the only interface between the network operator’s data systems on one side, and the physical layer on the other side, enabling the connection of the ‘smarts’ to the physical network.
Elements of this layer typically have one ‘wet’ end or aspect with direct contact or relation to water, and one ‘dry’ data interface, such as a valve controller input, or a sensor’s data output.
Layer 3: Collection and communications
This is responsible for discrete data point gathering, transmission and storage, or similarly, the timely distribution of automation and control signals to endpoints throughout the network. Data loggers, SCADA, AMI, and similar data transfer devices are all part of this first ‘dry’ layer: it only moves data between layers.
Layer 4: The data management and display layer
This is where data from different sources comes together, to be used by operators, preprocessed, stored, transferred and accessed by central systems. Human commands or instructions from higher-level systems are interpreted into concrete device settings (for example, changing to a named network
configuration may imply switching several pumps on or off, changing valve states).
Obviously, this interfaces with the underlying communications infrastructure on one side, and with an operator or with other central data systems on the other side. This layer is often characterised by a collection of generic solutions, for example, to validate, archive, and display numerical data, with a fair amount of IT integration behind the scenes.
The dashboard applications provided with many SCADA systems fall into this layer, with some data validation and the graphic display of multiple data streams and so on. Other components in this layer include data repositories, GIS or network schematic visualisation tools, control room systems with simple alert rules, graphical control interfaces, water balance applications and fixed-rule feedback automation. In addition to water network data, this layer can also aggregate data from other operational systems, such as information from workforce and asset management systems. This layer is typically where operators first use the Smart Water Network, interfacing with data sources from
a more generic enterprise data structure, or blending in data from external sensors.
Layer 5: Data fusion and analysis
This brings together raw input from the data handling and discovers the knowledge which was not previously obvious from the raw data. The results may be displayed to an operator, passed on to further analysis within the layer, and trigger automatic action by means of layer 4, or directly via layer 3. The value of this information comes from sifting through the flood of data from multiple samples, data sources, and even data types, to extract high value information, in the form of alerts on problems, automated responses to system changes, high-level summaries, network forecasts and so on.
This layer includes hydraulic modelling systems, network infrastructure monitoring, smart pressure management, pumping or energy optimisation systems, or decision support systems. It is most distant from the physical layer, requires the most sophisticated intermediary data systems and is obviously the least developed in today’s smart water network.
Computer science and statistics offer several approaches applicable to analysing water networks based on historic and online data. One example of a commercial system based on such new methods provides some initial insights from its deployment and use in the water utility of Jerusalem, since mid 2009, using existing sensors only.
It demonstrates the abilities of an algorithmic-statistical approach in monitoring water networks, and its capability to detect actionable events: leaks, thefts, meter drifts and other network malfunctions. The overall approach of the TaKaDu system is one of detecting statistical anomalies in data, compared to what the system observes and models as “routine behaviour”.
This relies to some extent on existing computer science methodologies in anomaly detection (see Chandola et al 2009, for an overview of this field), and is a rather different approach than hydraulic modelling systems, where very explicit assumptions and precise physical calculations are required for any sort of meaningful result.
The benefits of this rather generic approach include ease of deployment (uses existing sensor array, no intensive modelling required of the utility), a wide and flexible range of possible input and output data types, and the ability to detect even anomalies of types unforeseen when designing the system.