Short description of the project
Process engineering processes usually have a large number of process variables which generate an alarm when a previously defined threshold value is exceeded in the event of abnormal process behavior. If a disturbance occurs in a central plant section, this leads to a flood of alarms, because the individual process variables are interdependent and the disturbance propagates through the plant (Fig. 1). Plant operators must then select the most important alarms under time pressure in order to be able to detect the actual cause of the fault. Important decisions may have to be made under an increased stress level in order to avert damage to the plant. The analysis of the occurring flood of alarms costs valuable time. Therefore an alarm management is useful, which helps to detect those process variables which are closest to the cause of the failure. The attention of the plant operator can thus be directed to these variables and the cause of the fault can be efficiently localized.
Project goals
At Fraunhofer IOSB, an intelligent alarm management system has been developed for this purpose, which analyzes plant-wide faults in a data-driven manner and localizes the process variables that are close to the cause of the fault. The approach used here exploits time shifts in the measurement data, which occur in the plant, for example due to process dynamics or in the form of dead times. This enables the fault propagation path of the fault to be reconstructed. Process variables, which are at the beginning of the propagation path are considered as possible causes of the error and are displayed with higher priority to the plant operator.
Project result
The intelligent alarm management was examined on the basis of several errors in an example process. The process is a laboratory setup in which the plant status is recorded over several process variables. Possible sources of error were, for example, a malfunction in the power supply of a pump, the clamping of a valve and the clogging of a line. In all cases it was possible to localize the cause of the error and to reconstruct the error propagation path.