ABSTRACT
Fault detection and elimination in industrial machineries can help prevent loss of life and financial assets. Fault diagnosis is a delicate matter and needs the assistance of experts and professionals, therefore it costs money. In this study the authors try to mechanize the process of machinery fault diagnosis without the help of experts. By the method developed in this article an ordinary rotating machinery operator can diagnose the fault and do the required action.
In this study four common faults in rotating machineries namely: 1) Mass Unbalance 2) Angular Misalignment 3) Bearing Faults and 4) Mechanical Looseness have been considered. Each of these defects has been created separately on a test rig comprising of an electrical motor coupled to a rotor assembly. A Vibrotest 60 vibration spectrum analyzer has been used to collect velocity spectrum of the vibration on the bearings. Eleven characteristic features have been chosen to distinguish different faults.
Based on the acquired data an Artificial Neural Network Multi Layer Perceptron has been designed to recognize each one of the aforementioned defects. After training the Neural Network, it was checked by new data gathered by new experiments and the results showed that the designed network can predict the faults with more than 70% reliability, and it can be a good assistance to an ordinary machine operator to guess the problem and hence make a good decision.
Keywords: Fault diagnosis, Neural network, Rotating machineries, Vibration.
