LRegional are on the rise. In Baden-Württemberg alone, the State Health Office registered around 100 Legionella infections last summer. Seven people died from it. Constant sampling and controls are life-saving in the truest sense of the word.
The Swabian inventors of Legio Tools GmbH from Walddorfhäslach have developed an analysis system with artificial intelligence for drinking water control. Legionella, like bacteria in general, but also other contaminants in drinking water, are recognized by this system with pattern recognition. The system can be installed directly in the house connection.
“In many buildings there is no maintenance of the water pipes at all,” says Rainer Kaifel, Managing Director of Legio Tools, summarizing the problem. “This is where sediments and bacteria are transported and are deposited.” This creates a biofilm that is sometimes highly infectious. “Sediments that are otherwise harmless in water also increase the bacterial load in the house in this way,” explains the graduate engineer and shows a piece of pipeline that he has removed from a house supply system. The layer of sediment that has deposited here is a good centimeter thick. Bacterial infections spread from here in a house pipe system, which then also affect people. “Some of these are serious illnesses, and some people die from such a bacterial infection,” reports Kaifel.
Detected: The Legio Tools software in action
:
Image: Peter Welchering
Such illnesses and deaths could be avoided by constant quality control of the water. For this purpose, a small connection for sampling is installed on the house supply line or the line that is to be monitored. Water then enters the analysis system at regular intervals via corresponding regulating valves. This consists of a microscope with an image processor and other connected optical sensors. “The sample is scanned with object recognition software,” says Kaifel. A neural network evaluates the scanned images and, thanks to the corresponding training data, can precisely identify whether it is legionella, other bacteria, microplastics or sediment contamination.
In the initial training phase, the neural networks learned from 100,000 images what exactly which bacteria or sediment looks like. Microbiologists made the assignment manually in advance. This initial training is not enough. The scientists regularly check the pattern recognition of the neural networks. The software learns new images with their assignments, and the scientists evaluate the previous hit rate. “It’s a continuous optimization process,” says Christine Anderko from Legio Tools, describing the process.
Separated: contamination and rust in the water pipe
:
Image: Imago
She presented the system, which is now ready for the market, at various clinics and in larger companies. “There is great interest, because even in large corporations, water is the only product that is not constantly monitored and therefore a potential risk,” says Anderko. It is important to the users that other parameters are evaluated in addition to the images from light microscopy. This includes the outside and water temperature, the pH value and the conductivity of the water. Kaifel: “This data and the optical recording then go through a forecast system.”
The forecast system also works with AI software and not only calculates the future water quality, but also warns of damage that may soon occur. These are assessed using recorded damage patterns and a probability calculation.
Take the digital future into your own hands.
Learn more
“If we can assign certain particles, we can immediately see which part of a pipeline will break next,” says Kaifel, summarizing the complex interplay between simulation and probability calculation. The line network operator can use the knowledge and carry out minor repairs in good time before a pipeline, for example, starts to leak.
Some of the analyzes take place on site, i.e. directly at the house connection or on the line to be monitored. The greater part of the analyzes and, above all, the calculation of the forecasts, however, take place in the cloud. Because these complex calculations require considerably more computing power.
Only small systems are used on site. Ultimately, this has cost reasons. Because the more computing power is installed on site, the more expensive the monitoring system. The data transfer is secured with the usual transport encryption. In particularly sensitive areas, such as in hospitals, however, this is not enough. The data is then sent to the cloud via a specially secured virtual private network.
.