AI-controlled water distribution network, visualizing smart water management.

Smart Water: How AI Could Be Fixing Leaks Before They Start

"AI-powered iterative learning control offers a new approach to managing water pressure, minimizing leaks, and conserving resources in distribution networks."


Water scarcity is a growing global concern, exacerbated by significant water loss due to leaks in distribution networks. Traditional methods of pressure management have proven to be effective in reducing leaks and bursts, but a new approach is needed to integrate advanced technologies for smart and efficient water resource management. With pressure too low, end-users don't get the water they expect, in turn leading to pollutants entering the network.

Research into water distribution systems has largely focused on optimal pump scheduling, leakage detection, and contamination prevention. One promising area is the use of iterative learning control (ILC), a technique that leverages past data to refine control actions and improve system performance over time. By applying ILC to pressure control, we can meet pressure requirements at critical points in the network while minimizing unnecessary pressure levels.

Iterative Learning Control (ILC) offers a dynamic approach to refining control inputs. This technique is beneficial for managing systems such as mechanical robots or production lines and is now being applied to enhance smart pumping station control. Iterative learning uses logged pressure data to calculate the most efficient pressure, reducing reliance on internal models. It's particularly effective because water usage often follows predictable patterns.

AI Search Multiple angles on this topic

Leaks and the Cost of Uncertainty

In drinking water distribution systems, especially those operating intermittently, large volumes of water are wasted through leaks in the network, and service to users is not always satisfied within the time required to fill storage or with sufficient pressure. Demand at each node is an aleatory variable driven by the unpredictable behaviour of water users, making it one of the main sources of uncertainty in the design of this infrastructure. The growing volume of water demand datasets opens opportunities to better characterise this variability. In response, researchers are proposing online statistical hypothesis tests and monitoring frameworks to improve the operation of water distribution networks, including approaches that combine data-driven and model-based leakage detection.

Classical Analysis and Its Limits

Accepted practice in water distribution engineering centres on hydraulic network analysis and optimisation, with methods such as the Cross method, as adapted in the Gessler network approach, used to solve the flow conditions of looped systems. A commonly cited example is the Hanoi network in Vietnam, which consists of 34 lines, three cycles, and 32 nodes and is powered by gravity with a fixed head of 328 feet. Other design traditions incorporate informational entropy alongside hydraulic routines to evaluate alternative feasible flow configurations. However, these conventional demand-driven models have recognised limitations, notably in predicting the extent of contamination spread during a contamination event, which has prompted the development of new real-time connectivity analysis approaches.

From Basic Design to Resilience Thinking

Foundational understanding of water distribution systems is captured in comprehensive overviews covering the components, layout, and operation of these networks as core urban infrastructure. That foundational knowledge has gradually expanded into reliability thinking, with modern work developing multi-scenario simulation models that evaluate the post-disaster performance of networks in supplying both firefighting flow and original demand under seismic damage. The focus of research has therefore shifted from designing networks that merely function under normal conditions to understanding how they perform when damaged or stressed.

The AI Model: Reducing Water Loss

AI-controlled water distribution network, visualizing smart water management.

Water distribution networks can be modeled as a graph, with vertices representing pipe connections and edges representing the pipes themselves. Each vertex is associated with pressure, demand, and geodesic level, while each edge is characterized by pressure drop due to hydraulic resistance. This model allows us to analyze the network's behavior and develop control strategies to optimize pressure levels.

To simplify the analysis and control design, we can create a reduced-order model by partitioning the network into inlet vertices (where water enters the network) and non-inlet vertices (representing end-users). By making certain assumptions about the network, such as uniform head at all inlets and consistent consumption profiles at non-inlet vertices, we can derive a simplified expression for pressure at each non-inlet vertex. This expression relates pressure to total demand, inlet pressure, and a constant term that captures the network's physical characteristics.

  • Reduced Leaks: By maintaining optimal pressure, the likelihood of leaks due to excessive pressure is significantly reduced.
  • Energy Savings: Lowering the amount of energy used by pumps.
  • Improved Comfort: Stable pressure for all consumers.
  • Avoiding increased risks of pollutants.
AI Search Multiple angles on this topic

A Renewed Research Agenda

Research on water distribution networks has recently been characterised by a deep renewal and development, driven by technical progress in control systems and computational resources. A substantial body of that work applies evolutionary optimisation, with genetic algorithm approaches - rooted in the natural selection and genetics concepts introduced in stochastic methods by Goldberg and Kuo - being used to design and optimise networks, and early applications such as Simpson et al. producing solutions close to the ideal. Reviews now provide in-depth and organised assessments of these optimisation studies, while bibliometric and systematic reviews map the field of network resilience research between 2000 and 2021. Looking ahead, reviewers note that advances in the Internet of Things and Low-Power Wide-Area Network technology will accelerate the adoption of smart meters in water systems.

What the Networks Can't Withstand

Water distribution networks are critical urban infrastructure, but as they expand and age, the risk of pipeline ruptures and leaks grows, and predicting these risks is essential for preventing accidents, improving management, and protecting public safety. Networks remain vulnerable to a range of threats, including leaks, pipe breaks, and contaminant intrusions that compromise the delivery of water of adequate quantity and quality. Practical methods have been developed to assess the impact of various pipe failure conditions without undertaking a full network analysis or simulation for each individual component failure. Extreme events can also overwhelm these systems, as seen during wildfires, where structural damage drains grids and contributes to water pressure failures.

Comparing Networks, Configurations, and Meters

Comparison in this field operates at several levels. Correct network segmentation is necessary to perform proper maintenance activities, and studies evaluate segmentation alternatives against criteria using decision matrices to guide choices. At the network-design level, research on steady motion in looped pipe networks highlights how leakage from pipes, which may take different opening shapes, creates problems for human health and the environment, informing comparisons of layout configurations. At the instrumentation level, brand comparison guides for magnetic flow meters benchmark products from manufacturers such as Endress+Hauser, KROHNE, Siemens, ABB, and Badger Meter for use in water treatment and chemical processing.

The control objective is to maintain a minimum pressure requirement at the measured vertices. The iterative learning control (ILC) algorithm adjusts the inlet pressures based on past performance to achieve this objective. By iteratively refining the control actions, the system learns to compensate for disturbances and uncertainties, ensuring that pressure requirements are met while minimizing unnecessary pressure levels. The controller will continuously adjust the inlet pressure to reduce pipe stress and overall energy consumption. Furthermore, the proposed control gives pressure set points to all inlets in the network instead of flow set-points thus reducing the need for flow measurements which are typically more expensive.

The Future of Smart Water Networks

The ILC-type control structure offers a promising approach to pressure control in water distribution networks, enabling reduced leaks, energy savings, and improved system performance. By leveraging AI and machine learning, we can create smarter and more resilient water systems that are better equipped to meet the challenges of water scarcity and environmental sustainability. Future research should focus on estimating inlet node elevation, handling sensor dropouts, and extending the approach to networks with elevated storage to push the boundaries of innovation in smart water management.

AI Search Multiple angles on this topic

Where the Industry's Priorities Converge

Water providers consistently rank asset management, revenue protection, operational efficiency, regulatory compliance, and customer service among their top concerns, and these priorities frame how new technologies are evaluated. The technical foundation beneath these concerns is pipe network analysis, which determines pipe flow rates and pressure heads at the outflow points of the network. Distribution networks, more broadly, are understood as the systems that route goods from producers to end customers. Seen together, this suggests that leak detection and AI-driven monitoring are assessed not as standalone technology but as tools that serve the operational and commercial goals utilities already manage.

The Road to Two-Way Smart Water

The future direction points toward the Smart Water Grid, described as a two-way water distribution network involving intelligent monitoring and response systems that rely on real-time information, sensors, and IoT technology-driven devices. Expanding cities require reliable water distribution networks supported by modern metering systems, and government initiatives focused on smart cities and digital utilities are encouraging the adoption of advanced water meters. Adjacent markets show a similar innovation trajectory, with research and development activity in water-related products increasing, including a reported annual rise of about 20% in patent activity and a growing focus on AI-driven personalisation and health monitoring features.

Laying Pipes, Integrating Systems

Distribution pipes are generally laid below road pavements, so their layouts tend to follow the layouts of roads, and four different types of pipe networks can be used, singly or in combination, depending on the location. Implementing integrated water management is not straightforward, with one of the primary hurdles being the need to ensure data accuracy and real-time synchronization across multiple interconnected components. Contemporary systems also face stresses they were not designed for, as wildfire events have shown, where power loss impacts water production and pumping. Even at the household level, distribution problems such as long wait times, pressure drops, and heat loss persist.

Real Networks, Real Constraints

Population and economic growth are placing increased demands on water distribution networks in both developed and developing countries, with consequences for economic development, health, and wellbeing. Implementing these improvements in the real world is constrained by computation, since for complex optimisation problems such as the design of real-world networks, the number of function evaluations or simulations required can be prohibitive. A hydraulic model, the workhorse of such analysis, is a mathematical representation of a distribution system that calculates how water moves through pipes, pumps, valves, reservoirs, and customer connections - tying every technical advance back to the practical job of getting safe water to people.

About this Article -

Written with AI assistance from published research, and reviewed by the Mystum team. See our About page for more information.

This article is based on research published under:

DOI-LINK: 10.1109/ccta.2018.8511513, Alternate LINK

Title: Iterative Learning Pressure Control In Water Distribution Networks

Journal: 2018 IEEE Conference on Control Technology and Applications (CCTA)

Publisher: IEEE

Authors: Tom Norgaard Jensen, Carsten Skovmose Kallesoe, Jan Dimon Bendtsen, Rafal Wisniewski

Published: 2018-08-01

Everything You Need To Know

1

How does Iterative Learning Control (ILC) use data to optimize pressure in water distribution networks?

Iterative Learning Control (ILC) uses past data to refine control actions. In the context of water distribution networks, logged pressure data is analyzed to calculate the most efficient pressure, reducing reliance on complex internal models. This is particularly effective because water usage follows predictable patterns. By iteratively adjusting the inlet pressures based on past performance, the system compensates for disturbances, minimizes unnecessary pressure levels, and maintains minimum pressure requirements at measured vertices.

2

What is the main objective of using Iterative Learning Control (ILC) in managing water pressure, and what benefits does it offer?

The primary goal of using Iterative Learning Control (ILC) in water distribution networks is to maintain a minimum pressure requirement at critical points while minimizing unnecessary pressure levels. The Iterative Learning Control (ILC) algorithm adjusts the inlet pressures based on past performance to achieve this objective, reducing leaks, optimizing energy consumption, and ensuring stable pressure for all consumers. This approach also helps in avoiding the increased risks of pollutants entering the system due to low pressure.

3

How is a water distribution network modeled for analysis, and what do the different components represent?

In a water distribution network, the network is modeled as a graph. Vertices represent pipe connections, each associated with pressure, demand, and geodesic level. Edges represent the pipes, characterized by pressure drop due to hydraulic resistance. This model helps analyze the network's behavior and develop control strategies to optimize pressure levels. For simplification, the network can be partitioned into inlet vertices (where water enters) and non-inlet vertices (representing end-users), enabling a simplified expression for pressure at each non-inlet vertex related to total demand, inlet pressure, and a constant term reflecting the network's physical characteristics.

4

What are the limitations of traditional methods in managing water pressure, and how does Iterative Learning Control (ILC) address these?

Iterative Learning Control (ILC) addresses the limitations of traditional methods by leveraging past data to refine control actions, improving system performance over time. Traditional methods primarily focus on pump scheduling, leakage detection, and contamination prevention, whereas Iterative Learning Control (ILC) offers a dynamic approach to refining control inputs, enabling smart pumping station control and enhancing overall water resource management by reducing leaks and bursts through optimized pressure management.

5

What are the next steps in enhancing Iterative Learning Control (ILC) for smart water networks, and what impact will these advancements have?

Future research should focus on estimating inlet node elevation, handling sensor dropouts, and extending the approach to networks with elevated storage. These advancements aim to push the boundaries of innovation in smart water management, making water systems more resilient and better equipped to handle water scarcity and environmental sustainability. By addressing these areas, Iterative Learning Control (ILC) can be further optimized for broader and more complex water distribution networks.

Newsletter Subscribe

Subscribe to get the latest articles and insights directly in your inbox.