No abstract

This paper introduces a new technique for the vibration condition monitoring of a set of spur gears. This technique, the Kolmogorov± Smirnov (KS) test, is based on a statistical comparison of two vibration signatures, which tests the`null hypotheses that the cumulative density function (CDF) of a target distribution is statistically similar to the CDF of a reference distribution' . In practice, the KS test is a time-domain signal processing technique that compares two signals and returns the likelihood that the two signals are statistically similar (i.e. have the same probability distribution function). Consequently, by comparing a given vibration signature with a number of template signatures for known gear conditions, it is possible to state which is the most likely condition of the gear under analysis. It must be emphasized that this is not a moment technique as it uses the whole CDF instead of sections of the CDF.In this work, the KS test is applied to the speci® c problem of direct spur gear condition monitoring. It is shown that this test not only successfully identi® es the condition of the gear under analysis (brand new, normal, faulty and worn out), but also gives an indication of the advancement of the crack. Furthermore, this technique identi® es cracks that are not identi® ed by popular methods based on the statistical moment and/or time± frequency (TF) analysis and the vibration signature. This shows that, despite its simplicity, the KS test is an extremely powerful method that eVectively classi® es diVerent vibration signatures, allowing for its safe use as another condition monitoring technique.

No abstract

No abstract

This paper presents a brief review of the signal processing techniques being used in industrial condition monitoring and diagnostics. Four main types of monitoring methods are discussed; Neural Networks, Wavelets, Time-Frequency methods, and Non-linear series. References to existing applications are also included throughout the paper. This aims to introduce the most common techniques which are being used today, namely the multilayer perceptron network, orthogonal and non-orthogonal wavelets, spectrogram, Wigner and Choi-Williams distributions, and Volterra series.