Date of Award


Publication Type

Master Thesis

Degree Name



Electrical and Computer Engineering


Saif, Mehrdad




Induction motors are frequently used in many automated systems as a major driving force, and thus, their reliable performances are of predominant concerns. Induction motors are subject to different types of faults and an early detection of faults can reduce maintenance costs and prevent unscheduled downtime. Motor faults are generally related to three components: the stator, the rotor and/or the bearings. This study focuses on the fault diagnosis of the bearings, which is the major reason for failures in induction motors. Data-driven fault diagnosis systems usually include a classification model which is supported by an efficient pre-processing unit. Various classifiers, which aim to diagnose multiple bearing defects (i.e., ball, inner race and outer race defects of different diameters), require well-processed data. The pre-processing tasks plays a vital role for extracting informative features from the vibration signal, reducing the dimensionality of the features and selecting the best features from the feature pool. Once the vibration signal is perfectly analyzed and a proper feature subset is created, then fault classifiers can be trained. However, classification task can be difficult if the training dataset is not balanced. Induction motors usually operate under healthy condition (than faulty situation), thus the monitored vibration samples relate to the normal state of the system expected to be more than the samples of the faulty state. Here, in this work, this challenge is also considered so that the classification model needs to deal with class imbalance problem.