AI meets Digital Twin (KIDZ) – Predictive Maintenance based on scarce failure data

A breakdown of machine tool feed drives leads to high costs in operation. Therefore, within the project “KIDZ”, the iwb is investigating the combination of artificial intelligence algorithms and classical machine tool models for data-efficient predictive maintenance in the presence of only little failure observations.

Feed drives have a strong impact on the economic efficiency of machine tools, as their malfunction leads to costly and time-intensive maintenance actions. Furthermore, degradation of feed drives can change the machine dynamics and reduce the machine’s productivity. Hence, the adaptation of a predictive maintenance strategy for machine tool feed drives promises economic benefits and competitive advantages. However, as failure data is scarce, implementing such a predictive maintenance system is difficult. So far, many investigations applied additional sensors on laboratory test benches, which casts doubt on the application of such systems in operation. A predictive maintenance system, which is capable of detecting wear in feed drives of machine tools in operation based on solely numeric control signals is not available yet.

Training artificial intelligence models with simulated data

In this project, a so-called hybrid system for predictive maintenance is under investigation. This system combines a physical digital twin of the machine tool with modern methods of artificial intelligence, in order to estimate the current and future degradation state of a machine tool’s feed drive. The proposed approach has the advantage of requiring only little experimental data, which is costly to gather.

Localizing wear with the help of a physical digital twin

Another advantage of integrating a physical digital twin is the gain in interpretability. Due to the nature of physical models, predictions can be explained and wear cannot only be detected globally, but on a single component level, which is the information needed in a predictive maintenance system. This will enhance the cost-efficiency and acceptance of such a predictive maintenance system. The developed system will be validated with the help of long-term run-to-failure experiments within a machine from the Grenzebach Maschinenbau GmbH.

Acknowledgements

We express our gratitude to the Bavarian Ministry of Economic Affairs, Regional Development, and Energy for the funding of our research within the project “KIDZ” (grant number DIK-2006-0017//DIK0140/01). Furthermore, we would like to thank the VDI|VDE|IT for the exceptional support and the trustful cooperation.

Runtime         Juli 2021 - Januar 2026  
Sponsor         Bayerisches Staatsministerium für Wirtschaft, Landesentwicklung und Energie (StMWi)