Use of machine learning techniques for classification of thermographic images

Lozanov, Yavor; Tzvetkova, Svetlana; Petleshkov, Angel

doi:10.1109/bulef51036.2020.9326046

Cited by 8 publications

(1 citation statement)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Finally, the fault classification task was accomplished using the random forest classifier. Lozanov et al 32 developed a framework that utilizes thermal images of induction motors to classify motor states into three categories: cooling fan fault, motor bearing fault, and healthy state. The framework employs Otsu thresholding to segment the moving structures and utilizes three texture features to represent the required information.…”

Section: Related Workmentioning

confidence: 99%

A deep learning approach for electric motor fault diagnosis based on modified InceptionV3

Xu,

Teoh,

Ibrahim

2024

Sci Rep

View full text Add to dashboard Cite

Electric motors are essential equipment widely employed in various sectors. However, factors such as prolonged operation, environmental conditions, and inadequate maintenance make electric motors prone to various failures. In this study, we propose a thermography-based motor fault detection method based on InceptionV3 model. To enhance the detection accuracy, we apply Contrast Limited Adaptive Histogram Equalization (CLAHE) to the input images. Furthermore, we improved the performance of the InceptionV3 by integrating a Squeeze-and-Excitation (SE) channel attention mechanism. The proposed model was tested using a dataset containing 369 thermal images of an electric motor with 11 types of faults. Image augmentation was employed to increase the data size and the evaluation was conducted using fivefold cross validation. Experimental results indicate that the proposed model can achieve accuracy, precision, recall, and F1 score of 98.82%, 98.93%, 98.82%, and 98.87%, respectively. Additionally, by freezing the fully connected layers of the InceptionV3 model for feature extraction and training a Support Vector Machines (SVM) to perform classification, it is able to achieve 100% detection rate across all four evaluation metrics. This research contributes to the field of industrial motor fault diagnosis. By incorporating deep learning techniques based on InceptionV3 and SE channel attention mechanism with a traditional classifier, the proposed method can accurately classify different motor faults.

show abstract

Section: Related Workmentioning

confidence: 99%