Condition monitoring and anomaly detection: Machine learning-based bearing and eccentricity faults detection in induction machines

Thumbnail Image

Files

master_Nuha_Nurjahan_2026.pdf (8.45 MB)

URL

Journal Title

Journal ISSN

Volume Title

School of Electrical Engineering | Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.

Authors

Date

2025-12-19

Department

Major/Subject

Electrical Power and Energy Engineering

Mcode

Degree programme

Master's Programme in Automation and Electrical Engineering
Automaation ja sähkötekniikan maisteriohjelma
Magisterprogrammet i automation och elektroteknik

Language

en

Pages

57

Series

Abstract

The efficient condition monitoring of induction motors is very important for industrial systems because it can help to avoid catastrophic failures and improve operational safety. This thesis is concerned with the detection and classification of two main motor faults, namely bearing faults and air-gap eccentricity, using both current and vibration signals. Experimental data were obtained from a 5.5 kW four-pole induction motor under a variety of load conditions. Analysis in the time, frequency, and time-frequency domain was performed in order to identify fault characteristic frequencies such as BPFO and BPFI, with additional attention given to eccentricity features in order to improve diagnostic reliability. A comprehensive signal processing and feature engineering pipeline was developed using MATLAB and Python. This included filtering, feature extraction, feature ranking (using random forest importance) and three classifiers (SVM, XGboost and MLP) were trained and evaluated giving near perfect results in distinguishing between healthy and faulty conditions . In addition to this, pre-trained convolutional neural networks (ResNet-18, ResNet-50 and GoogLeNet) were also used for transfer learning and feature extraction from time frequency representations. The results indicate that the combination of bearing and eccentricity features leads to more reliable classification, while the Convolutional Neural Network-based approach leads to good accuracy and a considerable reduction in computational effort. Hence, the proposed framework provides a modular and reproducible basis for fault diagnosis, thereby linking the gap between experimental validation and realistic industrial application.

Description

Supervisor

Belahcen, Anouar

Thesis advisor

El Bouharrouti, Nada

Keywords

induction motor, fault diagnosis, bearing defect, eccentricity fault, time–frequency analysis, machine learning

Other note

Citation

Permanent link to this item

https://urn.fi/URN:NBN:fi:aalto-202601191443

Collections

[dipl] Sähkötekniikan korkeakoulu / ELEC