<abstract xmlns="http://www.w3.org/1999/xhtml">

<p>Rotating machines are widely used in today’s world. As these machines perform the biggest tasks in industries, faults are naturally observed on their components. For most rotating machines such as wind turbine, bearing is one of critical components. To reduce failure rate and increase working life of rotating machinery it is important to detect and diagnose early faults in this most vulnerable part. In the recent past, technologies based on computational intelligence, including machine learning (ML) and deep learning (DL), have been efficiently used for detection and diagnosis of bearing faults. However, DL algorithms are being increasingly favoured day by day because of their advantages of automatically extracting features from training data. Despite this, in DL, adding neural layers reduces the training accuracy and the vanishing gradient problem arises. DL algorithms based on convolutional neural networks (CNN) such as DenseNet have proved to be quite efficient in solving this kind of problem. In this paper, a transfer learning consisting of fine-tuning DenseNet-121 top layers is proposed to make this classifier more robust and efficient. Then, a new intelligent model inspired by DenseNet-121 is designed and used for detecting and diagnosing bearing faults. Continuous wavelet transform is applied to enhance the dataset. Experimental results obtained from analyses employing the Case Western Reserve University (CWRU) bearing dataset show that the proposed model has higher diagnostic performance, with 98% average accuracy and less complexity.</p>
</abstract>

Rotating machines are widely used in today’s world. As these machines perform the biggest tasks in industries, faults are naturally observed on their components. For most rotating machines such as wind turbine, bearing is one of critical components. To reduce failure rate and increase working life of rotating machinery it is important to detect and diagnose early faults in this most vulnerable part. In the recent past, technologies based on computational intelligence, including machine learning (ML) and deep learning (DL), have been efficiently used for detection and diagnosis of bearing faults. However, DL algorithms are being increasingly favoured day by day because of their advantages of automatically extracting features from training data. Despite this, in DL, adding neural layers reduces the training accuracy and the vanishing gradient problem arises. DL algorithms based on convolutional neural networks (CNN) such as DenseNet have proved to be quite efficient in solving this kind of problem. In this paper, a transfer learning consisting of fine-tuning DenseNet-121 top layers is proposed to make this classifier more robust and efficient. Then, a new intelligent model inspired by DenseNet-121 is designed and used for detecting and diagnosing bearing faults. Continuous wavelet transform is applied to enhance the dataset. Experimental results obtained from analyses employing the Case Western Reserve University (CWRU) bearing dataset show that the proposed model has higher diagnostic performance, with 98% average accuracy and less complexity.

Bearing Fault Detection and Diagnosis Based on Densely Connected Convolutional Networks

Faculty of Engineering Science, Department of Information and Communication Technology

Acta Mechanica et Automatica

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Rotating machines are widely used in today’s world. As these machines perform the biggest tasks in industries, faults are naturally observed on their...

Bearing Fault Detection and Diagnosis Based on Densely Connected Convolutional Networks

Online veröffentlicht: 24. März 2022

Seitenbereich: 130 - 135

Eingereicht: 18. Dez. 2021

Akzeptiert: 12. Feb. 2022

DOI: https://doi.org/10.2478/ama-2022-0017

Schlüsselwörter
bearing, deep learning, machine learning, transfer learning, fault detection and diagnosis, CWRU dataset

© 2022 Julius Niyongabo et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Bearing Fault Detection and Diagnosis Based on Densely Connected Convolutional Networks

Online veröffentlicht: 24. März 2022

Seitenbereich: 130 - 135

Eingereicht: 18. Dez. 2021

Akzeptiert: 12. Feb. 2022

DOI: https://doi.org/10.2478/ama-2022-0017

Schlüsselwörterbearing, deep learning, machine learning, transfer learning, fault detection and diagnosis, CWRU dataset

© 2022 Julius Niyongabo et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Schlüsselwörter
bearing, deep learning, machine learning, transfer learning, fault detection and diagnosis, CWRU dataset