Abstract:
In large machines, the possible failures are numerous and complex events. Furthermore, monitoring a multitude of measurement points leads to an accumulation of a large amount of data, making the task of tracking the equipment even more difficult. Often, there might even be an emerging problem that can influence several indicators simultaneously without exceeding the alert level of any one of them in particular. For end-users, the proposed monitoring system acts like a black box with an alert that comes on when the equipment’s behaviour changes, much like the “check engine” of a car. The behavior of the equipment is tracked by a continuously updated statistical model and it is thus possible to compare future measurements with model predictions. Both the model and measurement uncertainties are taken into account by the procedure, and the confidence level can be determined by the user and is therefore a fixed quantity. In addition, the approach combines all the individual exceedances of each indicator in a single global deviation. When an alert is raised, the individual exceedances can be used in order to determine the source of the alert. Two different examples are presented where failures occurred on hydraulic turbine-generator units: the loss of a hydraulic turbine runner cone and a breakage of a thrust bearing.
