Acoustic Signal Based Condition Monitoring of Gearbox using Wavelets and Decision Tree Classifier

Objectives: Most machineries employ gears for efficient power transmission. Even minor faults with the gear box can lead to severe losses both in terms of energy and money. The vibration and acoustic signals from the gear box, which usually are said to be as an unwanted by-product of the operation, can be used for the condition monitoring and fault diagnosis of the gearbox. This study proposes the usage of machine learning algorithm for condition monitoring of a helical gearbox by using the sound signals produced by the gearbox. Methods/Analysis: The acoustic signals were captured using microphone from a gearbox with artificially created fault conditions. An exhaustive study using different discrete wavelet transformations for feature extraction from the acoustic signals was carried out and subsequently J48 Decision Tree algorithm was employed for selection and classification of the extracted features. Findings: The time domain acoustic signals were converted into frequency time domain data using different discrete wavelet transforms. Of all the wavelet ransforms, the Daubechies 5 Discrete Wavelet Transform was found to be the best suited for the current scenario. The methodology yielded a satisfactory classification accuracy of 97.6% when classified using J48 algorithm. Novelty/ Improvements: The classification accuracy yielded through this methodology is higher than what was obtained by similar experiments with different methodologies till date. The results and their analysis is discussed in the study. The whole methodology when put in a real time sytem will have the capability to monitor the condition and diagnose the faults in the gearbox quickly and effectively. The performance of this methodology may be further improved by using different classifier algorithms.