Clever maintenance: New way to detect faults in machines at an early stage

Many conventional Artificial Intelligence (AI) methods rely on large amounts of data in order to work reliably. In practice, this is often not feasible - data is difficult to access, expensive or not available at all. Radtke is therefore pursuing a different approach: Expertise from the field is integrated directly into AI models so that they can manage with significantly less information - and still function robustly and precisely.

"We combine machine-specific expertise with the latest Artificial Intelligence methods," explains Radtke. This makes the system particularly adaptable - even with previously unknown machines or unusual error patterns.

The approach was trialled with freely available data sets, including ball bearing damage, a rock drilling machine and even a rocket engine. The results have already been presented at scientific conferences. They offer companies new perspectives for data-poor but intelligent condition monitoring - efficient, flexible and practical.

The work was supervised by Professor Jürgen Bock from THI and Institut AIMotion Bavaria, who is also the long-term contact person for the research topic. He emphasises:
"What we already know no longer needs to be extensively trained using machine learning methods."

The combination of empirical knowledge and AI opens up new avenues for intelligent and resource-saving solutions in the industrial environment.