| dc.description.abstract |
Bearings are essential component of rotating machines and are often prone to failure. Early detection of bearing faults thus becomes important for predictive maintenance strategies. Conventionally, vibration measurement is considered to be the most reliable and widely used indicator of fault signatures, which are to be extracted from the raw signal. Traditional signal processing techniques, like envelope spectrum, are employed for extraction of such features. However, selection of optimal band and center frequency remains the main objective of research in the field. Use of spectral kurtosis (kurtogram) is now a standard method for this selection. However, a benchmark study on Case Western Reserve University dataset shows several non-diagnosable cases using kurtogram method. The purpose of this study is to quantify diagnosability in the form of an index and use it as a selection criterion for getting optimal band and center frequency. The proposed method is validated using non-diagnosable cases of the benchmark study, and the results are compared with that of conventional Hilbert transform method and autogram method. |
en_US |