Feature extraction using Discrete Wavelet Transform for fault classification of planetary gearbox – A comparative study

Abstract

Monitoring the condition of unreachable gears of epicyclic gearbox in real-time increases the asset reliability by anticipating the failures through preventive maintenance. Machine vibrations exhibit fruitful information about the failure at different operating conditions when measured at a convenient location through signal processing techniques. This study demonstrates the potency of discrete wavelet analysis for fault diagnosis of the planetary gearbox using Artificial Neural Network and Support Vector Machine. The mean squared energy of the detail coefficients at eight decomposed levels for different wavelet families is considered a feature. The J48 algorithm selects the prominent features which form an input to the classifiers. Afterward, their relative ability in classifying faults is compared in detail. Further, the classification performance of commonly used machine learning algorithms in the literature is also compared. It is found that the mean square energy of detailed coefficients of Discrete Wavelet Transform exhibit excellent fault diagnosing characteristics with various classifiers and is highly recommended.