Critical operating states can be initiated in a power plant shaft lines by a number of various structural mechanical conditions, for example, cracks, bearing faults, coupling damage, imbalance as well as interruptions in the electric grid.

Thus permanent vibration monitoring is required. Many of these causes, as in the case of the cracks, are recognized by existing monitoring systems as deviations from the normal operating state only when it is too late. The reason is the frequency range analysis processes currently integrated in the existing Condition Monitoring Systems. For example, the amplitude and frequency variations within one shaft revolution characteristic of fracture formation are averaged out in fast Fourier-Transformation.

In the area of signal analysis, algorithms have been developed in the recent past for non-linear and non-stationary vibrations that allow computation of the frequency domain information in each measurement step. This process facilitates the detection of breathing cracks, in particular, in the transient operating state. The improved time resolutions of the frequency domain, however, are not sufficient by themselves to detect the crack because no unambiguous criterion for fracture detection yet exists. It is much more a classification problem for an extensive feature vector to be solved, one that separates the cracks from the other structural effects. A department of E.ON Anlagenservice provides services in the field of vibration monitoring on power plant turbines; Fraunhofer ITWM was approached by them to develop a prototypical software tool for the automatic detection of cracks on the basis of a conceptual analysis of the latest methods from the fields of signal analysis and crack classification.

The time signal from the shaft vibration sensors stored by today’s power plant monitoring systems for two different points in time with the same operating state are imported into the toolbox and broken down into single oscillations. Lastly, the frequency domain data is calculated for each sub-signal for every point in time. The classification problem of separating the crack from other structural effects can then be solved using trend data. The analyzed data can be displayed in the form of a frequency spectrum, polar plot, or orbit plot. Relationships between the features can be visualized in radar plots. A follow-on project is planned that will implement a hardware system for the online detection of crack in power plant shaft lines, which is expected to be deployed by E.ON Anlagenservice to power stations.