German research cluster is looking for pilot customers and partners for a robust condition monitoring system using AI based data fusion of multivariate sensors

Summary German research cluster has bundled competences from non-destructive evaluation with industrial data analytics and artificial intelligence to improve the performance of condition monitoring systems. Performance of classification is enhanced in case various damage or fault states must be distinguished. Potential industrial applications include slowly rotating drives, pumps, or other large industrial equipment. License, research or technical cooperation agreement are envisaged.

Description

Two German research institutes are cooperating in a cluster to develop an improved condition monitoring system. The institutes combine their expertise from non-destructive evaluation (e.g., ultrasound testing) with industrial data analytics and artificial intelligence for sensor data fusion.

A condition monitoring system should provide a high sensitivity to emerging damage for predictable maintenance, robustness, and reliability of defect state classification. The cost-effective system implementation is mandatory for an economic benefit of condition monitoring.

The system should be able to distinguish between different types and locations of damages to trigger maintenance actions. However, the reliable assignment of sensor signal signatures to the relevant damages is made difficult by changing operating conditions or uncertainties in system parameters. In case of slowly rotating drives, the detection of damages in vibration sensor signals is deteriorated by noise, that is always present in a real-world application.

Application of heterogenous sensors, e.g., ultrasonic transducers and low frequency vibration sensors enables an enhanced classification and early damage detection. To this end, the sensor signals are fused by a multi-stage classification based on neural networks. The condition monitoring system delivers damage classification results for both sensor modalities and for the combined approach. The combined classifier allows for higher sensitivity, while the outputs of the separate sensors can be used for validation and as a fallback option in case one of the sensors fails.

The research institutes have a long-term experience in condition monitoring, non-destructive evaluation, and development of machine learning systems for industrial applications, both in funded research and direct contract research. The collaboration resulted in a prototype implementation of the condition monitoring system tested on laboratory scale.

For the next step, pilot users from the industry are sought for a technical cooperation to implement feasibility studies to evaluate the technology in relevant industrial use cases. In case of a successful feasibility stage, the system can also be integrated into existing condition monitoring infrastructure at the user’s site, including a license agreement on the transferred data analytics algorithms.

Also, the system can be enhanced by integration of further sensor modalities in order to explore other industrial applications in a research cooperation.

Advantages & innovations The developed monitoring system has an enhanced sensitivity and classification performance in case various damage or fault states must be distinguished. This enables condition monitoring for use cases where conventional systems might not be able to reliably detect and classify damages. This can especially be relevant when high noise levels deteriorate sensor signals, or in case very slowly moving drives and bearings produce. Furthermore, the sensor signal analysis based on explainable AI methods allows for a higher confidence in the classification, a self-monitoring capability and a robustness in case of single sensor failures. The modular classification approach not only enables a deeper insight into the effects of faults on the classification. It is possible to apply the system as a rapid prototyping development kit for condition monitoring systems, that evaluates several sensor modalities in parallel and gives hints for the most valuable sensor configuration.

Stage of development Prototype available for demonstration

Contact / source: Detail (europa.eu)