AN EXPERIMENTAL COMPARISON OF RANDOM FOREST AND SUPPORT VECTOR MACHINE FOR BEARING FAULT DIAGNOSIS: A MICRO ELECTRO MECHANICAL SENSOR APPROACH

A sudden bearing failure leads to fatal accidents and results in loss of human life and increases the down time of the machine. Neural Network and Support Vector Machines (SVM) are widely used in rotating machinery fault diagnosis, while Random Forest (RF) based on the ensemble learning method, is relatively unknown in this field. Currently, use of Micro Electro Mechanical Sensors (MEMS) for machinery fault diagnosis is receiving more attention as it is low cost, compact and portable. In this paper, vibration signals are collected from the Rolling Element Bearing (REB) using the MEMS sensor. Statistical features have been extracted from the wavelet packet coefficients of the vibration signals and used as input feature for the classification purpose. A framework for the comparison of RF and SVM is presented to identify the best classifier for bearing fault diagnosis. RF emerged as the best classifier based on the classification accuracy especially with a small training set leading to a promising tool for bearing fault diagnosis.