Automated fault detection, diagnosis and prediction through robust learning algorithms
Initial Situation and Problem Description
Asset health and asset performance are essential for the efficient and trouble-free operation of technical systems such as machines, industrial plants, electrical devices, motor vehicles, and many more. Modern information systems continuously monitor technical systems to analyze and predict the state of health and performance.
Scientific foundation of these information systems are models and algorithms for detecting system faults (fault detection), determination of possible causes of faults (fault diagnosis), and the prediction of system failures (fault prediction). Current research assumes that exact mathematical models of technical systems are available (model-based or knowledge-based diagnosis). Manual modelling of systems is however time-consuming in practice and thus causes high costs. Many projects are therefore often not realized. Manual modelling is often not possible due to sensors that are necessary for the models.
Goals of the Research Project
The overall project goal is the design and analysis of robust learning algorithms for automated fault detection, fault diagnosis, and fault prediction. Robust learning works without complex system models. Robust algorithms learn the necessary model from available data. Because of their robustness, they can learn from data of poor quality such as data that contains itself faults. With the help of robust learning algorithms, we want to reduce the time necessary for manual system modelling. This automatization should enable new types of applications.
Software applications of learning-based fault detection, diagnosis & prediction are information systems for monitoring and analyzing technical systems. Foundations for these information systems are technologies of the “Internet of Things”. These are for example cheap and accurate sensors (example hardware) or geo-information systems (example software) such as their efficient interaction.
The basis for a generally applicable software platform for the analysis of the condition of technical systems is created. This software platform will be simple and generally applicable. Possible application examples are:
- Fault message prioritization: Prioritizing fault messages according to their relevance. This enables the monitoring of a large amount of technical systems.
- Fault diagnosis in real time: If a downtime occurs suddenly, FDD algorithms identify possible causes instantaneous to guide maintenance work.
- Fault prediction (or predictive maintenance): Predicting downtimes, e.g. for supporting the planning of maintenance.
Duration: 01.01.2018 - 31.12.2020
Budget: ca. 490,000 € total costs and ca. 407,000 € funding contribution
- Software Competence Center Hagenberg GmbH (coordinator)
- ISW Industriesoftware GmbH
Funding Partner: Innovatives Oberösterreich 2020, Call "Digitalisierung"