Fault diagnosis of rotating machinery components

Abstract

Fault diagnosis of an electro-mechanical system (EMS) has been a real challenge for the engineers. For increasing operational reliability and machine availability, it becomes essential to monitor the operation of an electro-mechanical system for the detection and diagnosis of faults. Keeping the above facts in mind, this research is aimed at investigating the relationship between vibration and motor current signature analysis (MCSA) for faults such as - bent and unbalanced rotors, ball bearings and gears with seeded faults in the mechanical load components coupled with an induction machine. A mathematical model has been developed to understand the effects of loading on the induction motors. Along with this, a test rig with associated instrumentation for simulation of faults like - bent rotors, bearings and gears with seeded faults has been developed in the lab for performing experiments. In the first study on shafts, combined unbalance and bowed shaft faults have been investigated using Fourier transform and Hilbert-Huang transform. Machine learning techniques like artificial neural networks (ANNs) and support vector machines (SVMs) have been applied for the fault detection of shaft faults in the second part of this study on shafts. A unique case of detection of faulty bearing installed in a load machine coupled to an induction motor. Time domain and frequency domain spectrum of the stator current signals acquired and analyzed for the detection of an outer race fault of the ball bearing installed in the load machine and is a step towards building a diagnosis model of typical bearing faults occurring in mechanical load systems using MCSA. It has been found that the energy levels at fault scale increased with the severity of the bearing fault. A signal processing algorithm involving auto regressive (AR) based linear prediction filter, minimum entropy deconvolution (MED) and spectral kurtosis (SK) has been used for separating impulses from the faulty current signal and enhancing them by effectively deconvoluting the impulses. Time domain current signals of the healthy and faulty bearing (outer race fault) with various defect sizes have been acquired and spectrally analyzed for the bearing fault frequencies. In present study, we have also investigated the bevel gear faults. The response of gear signatures (healthy and faulty) have been acquired using vibration and current sensors. Advance signal processing technique such as continuous wavelet transform has been used to detect and compare the bevel gear faults.