Real-time Fault Detection for Advanced Maintenance of Sustainable Technical Systems

Most fault detection systems (FDS) have proved their efficiency in the detection of anomalies and disruptions in technical systems. However, the detection of these anomalies and disruptions is time consuming and not applicable to real-time applications. Moreover many technical systems like e.g. machines and plants require a real-time decision support for emergency shutdowns in order to enable human machine interactions and avoid cost-intensive machine breakdowns. Thereby, anomalies and system changes have to be detected in a fast and secure way. In this context, conventional FDS are not designed for real-time control of sensor networks. Such networks continuously generate a large volume of data and are vulnerable for sensor failures. Therefore, it is a challenging task to derive appropriate information and analyze them in real time. In this paper we propose an event-driven fault detection systems (ED-FDS) using complex event processing (CEP) approach, which provides a robust detection mechanism for severe machine failures in real-time. The idea of ED-FDS is based on discretization of continuous sensor signals by applying an event scheme for sensor data changes. The event scheme includes several event rules, which are generated by applying of data mining methods. Thereby, relevant features are extracted and uncommon behavior of the system is described by event rules. One emphasis of the paper is the combination of different basic events to complex events in a way that different warning levels can be processed. The evaluation of the proposed system is performed on a test-bed demonstrator, which allows the tuning of different disruptions.