I. Introduction
Condition monitoring of rotating machinery is vital for smooth processes in industry. Inaccurate manufacturing or extreme operating conditions are the common causes of machinery faults. Apart from financial losses, sudden machinery breakdowns may even lead to catastrophic failures [1]. A variety of machine condition monitoring methods are being used, which mainly include mechanical measurements, electrical measurements, performance and process measurements, tribology and non-destructive testing [2]. Vibration is a typical mechanical measurement parameter which has been used widely for the purpose [3]. Edwards et al [4] presented a broader review on vibration-based fault diagnosis of rotating machinery. Rotor faults are quite common, such as unbalance, misalignment, rub, bent, oil whirl and pedestal looseness [5]. Mehrjou [6] reviewed recent research on identification of rotor faults. Unbalance and misalignment are the most commonly occurred faults, and produce similar kind of frequency patterns that make the diagnosis process very difficult. The rotor balancing procedure involves attachment or removal of certain amount of weight to or from a particular location of the rotor. Such treatment is not appropriate to address misalignment faults. Therefore, accurate identification of these faults is extremely important prior to taking corrective action. Many methods have been presented so far to classify these faults. The methods are based on the analysis of time domain, frequency domain, time-frequency domain, wavelet decompositions, WignerVille distribution, motor current etc. [7]–[14]. Artificial intelligence is also a popular domain for automatic diagnosis of rotor faults. The methods include artificial neural networks [15], Bayesian networks [16], support vector machine (SVM) [5], [17], entropy & optimization methods [18] and fuzzy logic [19].