FAULT DIAGNOSIS OF ROTATING MACHINERY USING ARTIFICIAL NEURAL NETWORK

Abstract

Machine fault problems are the broad sources of high maintenance cost and unwanted down time across the industries. The prime objective of machinery and plant equipment is good operating condition. Condition based maintenance (CBM) consists of continuously evaluating the condition of a monitored machine and there by successfully identifying faults before the catastrophic breakdown occurs. There are various components in rotating machinery whose condition monitoring is essential for the machinery to perform effectively. Defects in the bearings and cracks in the rotor have been identified as factors limiting the safe and reliable operation of turbo-machines. Transverse cracks are the most common and most serious cracks as they reduce the cross-section and thereby weaken the rotor (as stress concentration is significant around the crack). In this work, the influence of transverse cracks on vibration response in a rotor is analyzed. Experimental tests have been conducted on the healthy rotor with the bearings having defects such as outer race, inner race, and ball defects at speed ranging from 1000 to 5000 RPM on MFS (Machinery Fault Simulator) and data has been collected .Similarly the tests were conducted on the rotor with depth of crack as 10% and 20% of diameter of the rotor along with the defects on the bearings. Cracks have been introduced in the rotor at the center of the shaft using Electrical Discharge Machining (EDM). Piezoelectric accelerometers were used for picking up the vibration signals from various stations on the test rig. Change in the response examined for small and deep slots that can be considered as open cracks. The dynamic response of a rotor with various depth of crack is evaluated by using Fast Fourier Transform (FFT).The denoising of the signal has been carried out so as to extract the features out of the frequency response in order to train the ANN (Artificial Neural Network). Statistical methods have been used in order to extract features from the FFT and to reduce the dimensionality of original vibration features. The result shows that artificial neural network can be used for automated diagnosis of transverse crack and severity in terms of depth of crack.