Crisp-ML – a model for introduction of artificial intelligence
Let’s Talk Science shows how AI models remain reliable even when the environment changes
There are already numerous applications for artificial intelligence (AI) – ranging from predictive maintenance and process monitoring to automated quality checks based on process data. “The relatively high investment costs only pay off if an AI model delivers reliable statements in the long term,” knows Prof. Joachim Metternich, head of the Institute for Production Management, Technology and Machine Tools (PTW) at TU Darmstadt. “For example, we experience time and again that companies are surprised by the amount of data they have to collect and prepare in order to train an AI model for the relevant application scenarios.” Together with his collaborator Nicolas Jourdan, he shows in Let’s Talk Science how a sustainable deployment can succeed and provides valuable advice for practical use.
“We focus on the challenges a company faces when it wants to introduce machine learning models to make its production more efficient and environmentally friendly,” adds Nicolas Jourdan, research associate at PTW. These include, in particular, parameters of production processes and manufacturing machines that change continuously. They arise, for example, from wear and tear and sensor drift and cause the performance of already trained AI models to decline over time. The question of how trained models can be generalized and thus applied to further machines and processes is also considered by the Darmstadt experts in this webinar.
The researchers from Darmstadt demonstrate all these challenges and solution approaches along the freely accessible Crisp ML process model, which they explain using simulated and real data sets.
