I was thinking of trying to look for the high frequency signals from the fretting of the cracks - AE (acoustic emissions?) as it used to be called. Problem is that the noise generated by the bearings and so on would drown the signal from the impeller, unless the accelerometer was on the impeller itself. Typical frequency was 50 kH, Bruel and Kjaer issued some papers on this.
So far as I know it is not a common, robust, technique. Cheers
Using the equipment you are talking about, all you can do is closely monitor the 1X and 2X amplitude and phase, and hope that as the crack progresses that there is sufficient influence on the shaft stiffness, or balance state to cause a change in these parameters over time such that you have a heads up. Also carfully compare the start up and shutdown data from each time you start the machine to see if the transient data is changing.
Having said all that I will warn you that I have seen impeller and turbine disk cracks, that never caused a significant change in vibration characteristics until the machine failed. Not pretty.
If you suspect impeller cracks, nothing beats a direct inspection of the rotor.
Nice try, but won't work for typical dangerous cracks. By the time the crack is long enough to measurably affect the resonant frequency by changing the stiffness the blade will be a very short time from destruction.
Impeller cracks are detectable in certain circumstances. A technique which has been successful is 'Peak Vue'; a very haigh frequency signal enveloping system (similar to ESP) but which typically finds 'impact' damage frequencies (often emitted by cracks).
You can find more info by typing 'Peak Vue' at your search box.
