At MC-monitoring, innovation is a core part of our development strategy. As industrial requirements evolve, so do the technologies we explore to enhance the monitoring and protection of critical rotating equipment.
As part of our efforts to explore how machine learning (ML) and artificial intelligence can complement our proven solutions, one of our recent projects focused on unsupervised learning — a technique that identifies anomalies based solely on healthy operating data.
In collaboration with the Haute école d’ingénierie et d’architecture Fribourg (HES-SO), our team explored the use of autoencoders — a type of neural network — trained on vibration data from healthy roller bearings.
The goal? To detect early deviations in machine behavior without requiring fault-labeled data, which is often hard to obtain in real operating conditions. This approach reinforces one of the great promises of ML in industrial maintenance: being able to act early, even with limited failure history.
Our research showed that this method is highly effective at identifying abnormal patterns in vibration signals, even under varied loads and noise levels. With robust detection performance, autoencoders confirmed their potential as an intelligent layer in future monitoring strategies.
We also explored advanced techniques to extract more information from these anomalies — including the possibility of distinguishing between different types of bearing faults. While this part remains exploratory, the insights gained will guide our next R&D steps.
This project is part of our broader strategy to enhance the intelligence of our systems, enabling:
By building our internal expertise in ML, we are preparing the ground for future enhancements that are data-driven, scalable, and user-focused.
Machine learning is expected to play a growing role in enhancing diagnostic capabilities, improving anomaly detection, and supporting maintenance strategies across rotating equipment in energy and industrial applications.
To request the full white paper on unsupervised learning for anomaly detection in rotating machinery, please contact our team.