Webinar Artificial Intelligence in Water Technology

Have you ever considered that the ever-larger amounts of data can be transformed into new insights about our surrounding world? On our upcoming webinar on the 11th of February, 15.30 – 17.15, we will address Artificial Intelligence in Water Technology.
The management of the water infrastructures carrying high quality water in large quantities has already large benefits from novel and automated ways to collect, process, and transform data into information. Predicting upcoming malfunctions is one of them.
During this webinar we will learn about some more examples and go deeper in this fascinating world. Our invited speakers are Peter Baltus from the Technical University of Eindhoven, Mark Roest from VORtech and Caspar Geelen MSc, PhD researcher at Wetsus, more information below.
You can register for this interesting webinar at
You will receive a link to the livestream of the webinar in the week before the event.

Speakers and topics Artificial Intelligence: Peter Baltus, TUE, Eindhoven
Inspecting pipelines using swarms of evolving sensor spheres
Inspection of water distribution pipelines shares many properties with other applications that require exploration of inaccessible environments, such as sewer systems, industrial mixers, nuclear reactor vessels, space or the inside of the human (or animal) body:
for all these applications, the size of the sensors is limited and, as a consequence, only a limited amount of energy will be available. Also, communication will be difficult and GPS and similar systems for determining the location of a sensor measurement is usually unavailable. In the Phoenix project, this class of problems was addressed through technology consisting of a swarm of evolving sensor spheres that optimize their performance for a specific application over a number of generations. This technology is currently being developed further by Antea and TU/e. In this presentation, after a short introduction to the technology, an overview of pipe inspection applications and remaining challenges will be discussed. Mark R.T. Roest, VORtech, Delft
A small selection of techniques for smart network management
As a company, we specialize in software development for models and simulation in an operational settings. These days, the systems that we typically work on would be called digital twins. They typically consist of collected data, analysis algorithms and models to run what-if scenarios. These models can be traditional numerical models, models that are created from data using machine learning or a combination of numerical models and real-time data. In many cases, the data is limited or noisy and combining observational data with numerical models brings significant benefits. We will discuss several examples of such a combination. In addition, we will discuss work that we have done on detecting faulty sensors for traffic monitoring, which would also be applicable for water distribution systems. Caspar Geelen, MSc,  WUR, Wageningen & Wetsus, Leeuwarden
Leakage detection and optimal sensor placement in Drinking Water Networks
In order to detect leakages in the drinking water network as fast as possible, a water company relies on flow and pressure sensors in
the distribution net. However, sensors are costly and not every company can afford to saturate their network with sensors. Even for a limited number of sensors, robust leakage detection is possible by using real-time machine learning algorithms.

Since water demand changes with weather, holidays, pandemics and various other factors, often data sets from these factors are needed to correctly predict water demand and thus correctly detect leakages. However, the detection of leakages is also possible by only looking at the relations between different sensors in the network. This alternative leakage detection strategy, called exogenous nowcasting, will be illustrated.

In addition, the accuracy of leakage detection increases with the placement of additional sensors, but where should these sensors be placed? Traditionally, sensors are placed based on hydraulic model simulations or observability analysis of a specific sensor configurations. These techniques have some severe limitations. An alternative method of optimal sensor placement solely relying on network asset properties and locations, as well as estimates for flows and demands will be outlined. Thus, optimal sensor placement can be used to improve the accuracy of leakage detection, which in turn helps to find new optimal sensor locations.