top of page

Article in trade magazine "Factory" - April 2021

Predictive maintenance with AI and ultrasound


Predictive maintenance can be complex and expensive. Does that exclude SMEs? A Viennese company is making predictive maintenance accessible to smaller companies. And is relying on the combination of AI and ultrasound.



When Markus Loinig visits potential new customers, he is often met with scepticism. After all, in addition to a tablet and a few ultrasonic sensors, he also has the promise in his luggage to provide information within a few minutes about the condition of a machine or system that he has never seen before.


"At first, no one believes that this can be solved in a short time without specific data," says Loinig, "and that is probably also because there have already been too many empty promises in the area of predictive maintenance."


Loinig, who studied mechanical engineering and has worked for Daimler and as a management consultant, founded Senzoro in 2019. The company uses a combination of ultrasonic sensor technology and artificial intelligence to get to the heart of machine and plant failures.


Users can create their own AI models


And this can be done quite quickly - especially in the case of roller bearings. Using magnets, powerful broadband ultrasonic sensors are attached to the machine to be examined, which record the specific structure-borne sound of the machine component. The values are determined within a few seconds and transferred to the tablet, whose AI extracts the necessary features and displays the results in the form of a health score from 0 to 100. Then it goes directly to the next system, fixed sensors are not installed.


This is made possible on the basis of great experience: the AI works with the information from more than 20,000 measurements on hundreds of roller bearings. "In the meantime, we have achieved a depth of knowledge that allows us to say precisely after the measurement whether the roller bearing will last another three months or not," says Markus Loinig. Above all, the "BeepMeep®" system can determine both the progress of damage and its location based on the frequencies emitted by the roller bearing. For example, damage in the outer ring "sounds" different from damage in the inner ring.


And the learning can certainly continue. BeepMeep® is designed so that users can record their specific data and create their own AI models based on it. The highlight: BeepMeep® automatically selects the best model and shows both its accuracy and the weights of the underlying features.


"Our AI can also work standalone," says Loinig, "but marrying it to specific data naturally makes the results even more accurate." Knowing whether an element is approaching or moving away from the failure signature also enables AI-based smart trending: using BeepMeep®, users can thus also track progressions.


Knowing whether an element is approaching or moving away from the failure signature enables AI-based smart trending.

A lot speaks in favour of ultrasound


For the measurements, Senzoro uses the most powerful ultrasonic sensors currently available on the market. They allow frequencies of up to 700 kHz to be detected, and this structure-borne sound lies in a frequency range that is neither audible nor perceptible to humans. In Markus Loinig's eyes, ultrasound is a tool of choice for predictive maintenance that will gain in importance.


On the one hand, because the technology works non-invasively: Ultrasound can be used to measure machines and plants, motors, pumps and other technology during operation and without installing permanent sensors.


On the other hand, according to Loinig, the combination of ultrasound and AI is future-proof: measurements are hardly influenced by neighbouring sources and therefore provide very pure data. This means that data from different sources can be combined to apply trained AI models across plants on this basis. Ultrasound also has a high information density and the measurement time at a plant is only a few seconds.


Stationary applications to follow


Senzoro has the most experience in measuring roller bearings. So far, errors have only occurred in one direction, says Markus Loinig: it has happened that components lasted longer than predicted, "but it has never happened that something failed earlier".


Another area of application is gearboxes. Their higher complexity currently still requires three measurements, "but we are working on getting by with only one measurement here as well". BeepMeep® can also record data from other mechanical components such as motors, CNC spindles, pumps or generators.


Stationary applications are currently still in the research phase at Senzoro, but Markus Loinig sees exciting use cases coming here as well. The same applies to the detection of energy losses, for example through compressed air or gas leaks - the financial pressure in the industry is not yet too great here, but could increase significantly with the Green Deal, says the Senzoro founder.


"This has to become much easier"


The fact that BeepMeep® does not require an internet connection and thus no uploads to the cloud points to the target group addressed by Senzoro. Those who understand predictive maintenance in the sense of Industry 4.0 - i.e. who want to make continuously acquired data available to other applications in real time - are not served by it.


The impatience towards the big plans that all too often do not make it into practice is probably also driving Markus Loinig. "Predictive maintenance has to become much simpler, otherwise it will not work," says the company founder. "That is the reason why we also stay away from the large corporations with their long decision-making processes. With SMEs, I immediately have the decision-makers at the table and can show them immediately what our performance is."


Which usually only takes a few minutes.


"Predictive maintenance has to become much easier, otherwise it won't work." Markus Loinig, Senzoro


Translated into English by Senzoro GmbH

0 comments

Comments

Rated 0 out of 5 stars.
No ratings yet

Add a rating
bottom of page