Use this URL to cite or link to this record in EThOS:
Title: The development of new artificial intelligence based hybrid techniques combining Bees Algorithm, Data Mining and Genetic Algorithm for detection, classification and prediction of faults in induction motors
Author: Al-Musawi, Ammar
ISNI:       0000 0004 7973 0208
Awarding Body: Cardiff University
Current Institution: Cardiff University
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
This thesis focuses on applying Artificial Intelligence (AI) methods for detecting, classifying and predicting the faults in induction motors in order to prevent any failures happening during their operation due to loading conditions. It is very important to monitor and detect any faults in the motor during its operation in order to alert the operators so that potential problems could be avoided. In this study, a new AI algorithm has been developed and applied to detect, classify and predict the induction motor faults at an early stage. This is based on a hybrid approach using the Bees Algorithm (BA) and Data Mining called Bee for Mining (B4M), which overcomes the drawbacks of current AI methods in achieving higher classification accuracy with reduced rule set generated from the training data. The proposed B4M algorithm has been implemented, tested and validated using the University of California at Irvine (UCI) dataset, and was compared with other well-known classifiers. Later, the proposed B4M algorithm was applied in dealing with two most common faults, firstly, that of rotor (one rotor bar, four rotor bars and eight rotor bars), and secondly, bearing defects (inner race, outer race and ball bearing defects). In this research, three condition monitoring techniques involving thermal imaging, current and vibration signal processing have been used to monitor these faults. Further, features such as image metrics and Discrete Wavelet Transform (DWT) coefficients were extracted from the thermal images, and DWT coefficients from the current and vibration signals. Later, five-feature selection methods were applied in order to select the best features for defect classification. Finally, an improvement to the proposed B4M was made by producing a new hybrid classification algorithm by combining Genetic Algorithm (GA) with B4M referred to as GA-B4M where the GA was used for feature selection. The new algorithms were successfully implemented on MATLAB and its performance was tested on real data and compared with other algorithms using the WEKA software. The results obtained for the thermal image monitoring data showed 98.97% classification accuracy with a reduced rule set containing 10 rules for B4M while a 100% accuracy with a larger rule set of 63 and 72 rules were achieved by Decision Table and OneR classifiers respectively. For the current monitoring data, the classification accuracy fell to 79.62% with only 8 rules for B4M, while 79.20% with 837 rules was achieved by Random Tree. Similarly, for the vibration monitoring data the B4M achieved 80.05% with 7 rules in comparison with Naïve Bayes tree at 79.25% with 31 rules. Furthermore, the results achieved by the proposed hybrid approach GAB4M on thermal imaging dataset also showed an overall improvement on the classification accuracy reaching 99.85% with 7 rules. Similarly, on the current and vibration dataset the GA-B4M obtained 79.98% with 16 rules and 98.74% with 7 rules respectively. This study has shown that the new proposed classification algorithms B4M and GA-B4M are able to detect, classify and predict the induction motor faults more reliably.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available