Enhanced unsupervised noise cancellation using angular resampling for planetary bearing fault diagnosis

Abstract

In this paper, some techniques for bearing diagnosis are reviewed (unsupervised angular resampling and noise cancellation, envelope analysis) and applied in combination for the first time to solve a particularly difficult diagnostic problem. Unsupervised noise cancellation exploits the periodicity of gear signals. Since the vibrations from gears are periodic in the angular domain, we propose an enhanced method that uses an unsupervised order-tracking algorithm to perform noise cancellation in the angular domain rather than in the time domain. This method is then applied to bearing fault diagnosis of a planetary bearing in a helicopter gearbox. Due to random speed fluctuation, unsupervised noise cancellation initially did not separate the gear and bearing signals. However, the enhanced noise cancellation, which includes a pre-treatment to suppress speed fluctuation based on phase demodulation of gearmesh frequencies, without the need for a tachometer signal, provides better results. Finally the denoised signal was studied using the envelope analysis technique, and the bearing fault frequency was then detected. Without proper noise cancellation this was not readily detectable in the spectrum noise.