An improved feature extraction technique for diagnosing thermal condition of bearing induction motor based on thermal image analysis /
Norliana Khamisan
- xxi, 254 pages : illustrations (some color) ; 30 cm. + 1 CD ROM
Faculty of Electrical & Electronics Engineering
Thesis (Doctor of Philosophy) -- Universiti Malaysia Pahang – 2022
Includes bibliographical references
This study demonstrates the importance of condition monitoring to detect early failures in bearing machinery automatically, faster and accurately. Bearings is one of the main components which has been used extensively in most machining systems including in electrical and mechanical systems. If it is damaged, it has the potential to produce adverse effects such as thermal stress which can cause damage to motor parts and affect the motor performance until it freezes and burn out. Thus, to prevent this problem occur, early monitoring detection is required as a precautionary measure for bearing components. In this research, a detector device that is infrared thermography (IRT) has been utilized since it is one of the non -destructive and most effective testing techniques to monitor and identify failures on motor bearings. Furthermore, infrared thermal imaging systems and image processing approaches are combined using computing systems to build efficient algorithms to detect motor bearing conditions more effectively. Hence, image processing to solve the problem of feature extraction techniques to detect the thermal state of motor bearing images is the main focus of this study. The Color-based, the original GWT-based feature extraction method has been enhanced and resulted in a new feature extraction method as well as become a novelty in this study. A total of three new feature extraction methods have been developed namely Enhanced Gabor Features (EGF), Enhanced Gabor Wavelet Sharpness Median Histogram (GWSMH) and Enhanced Gabor Wavelet Contrast Limited Adaptive Histogram Equalization (GWCLAHE) that aim to modify the thermal features of the motor bearing image in enhancing the visual of the original image. The ANOVA was also used to measure the strength of features for each developed method. Next, Linear Thresholding (LT) and MultiLayer Artificial Neural Networks (MLANNs) classification methods were also used to classify normal, warning, and abnormal conditions on the thermal image of the bearing. Finally, the experimental results on the thermal image of motor bearings have proven that the newly developed Enhanced Gabor Features (EGF) obtained the highest classification performance compared to other new feature extraction methods. This Enhanced Gabor Features method was found to have produced significant features for detecting normal, warning and abnormal conditions on the thermal image of motor bearings. The accuracy value acquired is 99.47%. This has also proved that the image enhancement techniques have given accurate results to improve the image quality in this study. Findings from this research showed that the Enhanced Gabor Features-based method is capable in detecting the abnormalities features from the thermal group of motor bearing images in this research.
THE0009364(Local)
Faculty of Electrical & Electronics Engineering--Dissertations