Did you know that Mariborski vodovod and the University of Maribor are cooperating within the AI REGIO project, in which they are monitoring the water consumption of the main water supplier of the Vzhodna Slovenija region with an experiment?
The experiment is intended to detect possible water leakage from the pipe network with a modern sensor network system developed by Mariborski vodovod. Because the data obtained by the flow sensor is written to the database without proper reasoning, artificial intelligence will be used for this purpose.
The experiment aims at three main goals:
- automatically estimate wet zones in the water pipe network using remote sensing data;
- compare the results with the water consumption of the network;
- build up AI based reasoning to detect water leakage using satellite remote sensing data and the ground data (source: AI REGIO Newsletter).
For the purpose of monitoring water leakage, the advantages of water flow meters installed over the entire water supply network and the concept of estimating soil moisture using remote sensing data were used. Nevertheless, challenges have emerged in implementing this approach, namely the facts that water consumption is measured by flow meters; that water leakage can be detected using ratio between income and outcome; that the micro location cannot be accurately identified due to the length of the water pipe network; and that some parts of the water pipe network are 50 years old (source: AI REGIO Newsletter).
The advantages of the experiment are:
- impact on the preservation of natural resources,
- increased water production and
- optimized management.
Since up to 15-20% of water in the pipe network was lost during the use of water flow meters in the past, the experiment decided to:
- use remote sensing technique to estimate soil moisture;
- continuously monitor using radar imagining satellite;
- use soil moisture estimation techniques for the precise definition of micro-locations (source: AI REGIO Newsletter).
In order to successfully implement the experiment, a two-part solution was used, namely, the construction of deep neural networks:
- Gathering ground measurement using remote sensing data via radar satellite (this task was performed manually – at least 200 ground points were selected. Data using previous acquisitions, current acquisitions and theoretical models has been used to train the deep neural network.
- Use of online flow meters by integrating theoretical models and past experience of human operators to predict water leakage (source: AI REGIO Newsletter).
The first prototypes of the mentioned deep neural networks are currently being tested in the infrastructures of the Mariborski vodovod. The final results of the Experiment will be published by the end of June 2022.