CLASSIFICATION OF BEARING FAULTS USING EXTREME LEARNING MACHINE ALGORITHMS

Roller element bearing fault diagnosis is crucial for industry to maintain machine in
good condition so that there is no delay of work due to machine breaks down. This
project implements the bearing fault diagnosis that classifies the bearing data into four
classes which are healthy bearing, inner race defect bearing, outer race defect bearing,
and roller element defect bearing. Most of existing bearing fault diagnosis are done
using Back Propagation (BP) algorithm which take a long time to train the neural
network resulting in inefficiency of training the Single Hidden Layer Feedforward
Neural Network (SLFN). Therefore, this project introduces three learning algorithms
which are Extreme Learning Machine (ELM), Finite Impulse Response Extreme
Learning Machine (FIR-ELM) and Discrete Fourier Transform Extreme Learning
Machine (DFT-ELM) to improve the bearing fault diagnosis accuracy and shorten the
time used to train and test the neural network.