Industrial Fault Diagnosis using Hilbert Transform and Texture Features

• Main Authors: Zabin, Mahe, Choi, Ho-Jin, Uddin, Jia, Furhad, Md. Hasan, Ullah, Abu.Barkat
• Format: Conference Proceeding
• Language: English
• Subjects: Computational modeling; Convolution; Convolutional neural networks; Data models; Deep Convolutional Neural Network; Fault diagnosis; Hilbert Transform; Industry 4.0; Training; Transforms
• ISSN: 2375-9356
• DOI: 10.1109/BigComp51126.2021.00031

An automated fault detection is a vital issue in smart industries of Industry 4.0. This paper presents a model of industrial fault diagnosis using deep learning algorithms. In the proposed model, a standard induction motor dataset that consists of six different types of fault is used as an input. Then as a preprocessing method we utilized Hilbert transform to extract the pre-processed signals with absolute values. After that, texture images are generated from the pre-processed signals. The texture pattern of the images is used for training and testing the deep convolutional neural networks. For analyzing the performance of the proposed system, we used the F1-score which is derived from precision and recall. Experimental results demonstrate that the proposed model exhibited average 98.48% F1 score for the dataset, where HC (98.33%), IRF (98.57%), BF (98.41%), and ORF (98.56%), respectively. In addition, the proposed model shows comparatively higher classification accuracy compared to the four state-of-art models by showing the higher F1 score.