Enhanced Unsupervised Noise Cancellation using Angular Resampling for Planetary Bearing Fault Diagnosis

The bearings of planetary gears provide one of the most difficult scenarios for detection and diagnostics of bearing faults, since the fault signals must pass through a tortuous and time-varying path to arrive at external measurement points where they can be detected. In the case of helicopter gearboxes, this is made even more difficult by the fact that strong background masking signals exist over the full acoustic frequency range, in particular from gears which convert an input shaft frequency at gas turbine speed of typically 350 Hz to a rotor output speed of typically 5 Hz. The first step in analysing bearing signals is to remove the contributions from the gears. Classically, in normal gearboxes, this is done by using a simple band pass filter that exploits their different frequency ranges. However, with helicopter gearbox signals, it is usually necessary to first remove the masking signals from gears before the bearing signals can be analysed. In a previous article by Ho, the self-adaptive noise cancellation technique was used to improve envelope analysis results, but this requires the gear signals to be deterministic and phase-locked to shaft speeds. If the shaft speeds vary somewhat, it may be necessary to resample the signals on an angular rather than temporal basis (so-called order tracking), to force the gear signals to be deterministic. This normally requires the use of a shaft phase-locked tachometer signal to perform the angular resampling. The aim of this article is to show how the angular resampling can be performed from the signal itself without the requirement of a tachometer signal, thus combining unsupervised noise cancellation with angular resampling (i.e., effectively eliminating the speed fluctuation) to enhance the quality of the separation. A newly developed separation technique is utilised, which is much more efficient than that used by Ho. After presentation of the separation technique, its application to a particularly difficult diagnostic problem is then detailed.