Pop, pop, pop. The world’s most complex popcorn making machine makes popcorn at SmartFactoryOWL at the OWL University of Applied Sciences. The automated popcorn popper is just one of the demonstration setups at OWL’s factory floor, where new technologies are tested to produce research results for production, work and business processes.
Among other tasks, the OWL team uses the setups to train and test machine learning algorithms. The algorithms are for instance trained to optimize energy consumption or help prevent alarm floods.
However, the team experienced problems, when collecting data to train the algorithms:
The automated popcorn making machine at SmartFactoryOWL
“You need a lot of data, and that amount of data is often very hard to get from real systems”, says Alexander von Birgelen, M.Sc. and research assistant at OWL. And if the data is collected from a real machine, while it doesn’t work correctly, the trained algorithm will be incorrect.
At SmartFactoryOWL they found that the solution for this problem was to create a Digital Twin of the real machine in Experior and collect the data by running simulations on the Digital Twin.
The OWL team uses the artificially generated data from the simulations to train the algorithms. Afterwards they test the trained algorithms on the model in Experior and then run anomaly detection on the real machine to determine if the algorithms are able to detect anomalous behavior.
Experior is very useful when working with the machine learning approach, because the simulations are very flexible
SmartFactoryOWL’s experience of using Experior was positive:
“Experior is very useful when working with the machine learning approach, because the simulations are very flexible,” Alexander von Birgelen says.
Rudolf Schuster, B.Sc. and research assistant at OWL, also found the modelling process simple and precise:
“I can easily drag the assemblies I need from the catalog, place them where I want, and customize their size and position”, Rudolf Schuster says and adds: