Deep Learning Based Fault Detection from Bearing Vibration Data

Engin Oğuzay; Murat Balta

Makale detayları

Deep Learning Based Fault Detection from Bearing Vibration Data

Yazarlar:

Engin Oğuzay | Murat Balta

Yayıncı:

Giresun University

Ana Başlıklar:

Bilgisayar Yazılımı

Dergi:

Karadeniz Fen Bilimleri Dergisi Vol.14 Issue 3, pp.1159-1175

Belge Tipi:

Makale

Yayın Tarihi:

9/2024

Belge Dili:

Türkçe

Document ID (Org No.):

2001326797

pISSN:

2564-7377

Anahtar kelimeler:

Özet:

 Analysis of bearing vibrations can provide information on the overall health of a machines mechanical components. In this study, deep learning algorithms were integrated in both 1-D and 2-D data spaces to detect defects in motor mechanics commonly utilized in industry, and to increase production efficiency. Ten different classes were studied using the popular and comprehensive Case Western Reserve University CWRU bearing dataset, which includes three types of faults - the outer race, the ball, and the inner race - as well as a healthy class. The bearing vibration signal was handled in four ways: using the original vibration data, extracting features from the original data, applying STFT to the original data, and extracting features from the STFT-applied data. Machine learning approaches such as KNN, SVM, and 1D WDCNN were applied to the 1-D data. Additionally, STFT transformation was applied in the 2-D data space, and performance measurements were made with different statistical metrics using EfficientNetB0, EfficientNetB1, ResNet18, and 2D WDCNN. In the 2-D space, deep learning methods achieved 100 accuracy.

 Analysis of bearing vibrations can provide information on the overall health of a machines mechanical components. In this study, deep learning algorithms were integrated in both 1-D and 2-D data spaces to detect defects in motor mechanics commonly utilized in industry, and to increase production efficiency. Ten different classes were studied using the popular and comprehensive Case Western Reserve University CWRU bearing dataset, which includes three types of faults - the outer race, the ball, and the inner race - as well as a healthy class. The bearing vibration signal was handled in four ways: using the original vibration data, extracting features from the original data, applying STFT to the original data, and extracting features from the STFT-applied data. Machine learning approaches such as KNN, SVM, and 1D WDCNN were applied to the 1-D data. Additionally, STFT transformation was applied in the 2-D data space, and performance measurements were made with different statistical metrics using EfficientNetB0, EfficientNetB1, ResNet18, and 2D WDCNN. In the 2-D space, deep learning methods achieved 100 accuracy.

 Analysis of bearing vibrations can provide information on the overall health of a machines mechanical components. In this study, deep learning algorithms were integrated in both 1-D and 2-D data spaces to detect defects in motor mechanics commonly utilized in industry, and to increase production efficiency. Ten different classes were studied using the popular and comprehensive Case Western Reserve University CWRU bearing dataset, which includes three types of faults - the outer race, the ball, and the inner race - as well as a healthy class. The bearing vibration signal was handled in four ways: using the original vibration data, extracting features from the original data, applying STFT to the original data, and extracting features from the STFT-applied data. Machine learning approaches such as KNN, SVM, and 1D WDCNN were applied to the 1-D data. Additionally, STFT transformation was applied in the 2-D data space, and performance measurements were made with different statistical metrics using EfficientNetB0, EfficientNetB1, ResNet18, and 2D WDCNN. In the 2-D space, deep learning methods achieved 100 accuracy.

 Rulman titreşimlerinin analizi, bir makinenin mekanik bileşenlerinin genel sağlığı hakkında bilgi sağlayabilir. Bu çalışmada, endüstride yaygın olarak kullanılan motor mekaniklerindeki kusurları tespit etmek ve üretim verimliliğini artırmak için derin öğrenme algoritmaları hem 1 boyutlu hem de 2 boyutlu veri uzaylarına entegre edilmiştir. Popüler ve kapsamlı Case Western Reserve Üniversitesi CWRU rulman veri kümesi kullanılarak on farklı sınıf üzerinde çalışılmıştır; bu veri kümesi üç tür hata dış bilezik, bilye ve iç bilezik ve sağlıklı bir sınıf içermektedir. Rulman titreşim sinyali dört şekilde ele alınmıştır: orijinal titreşim verilerinin kullanılması, orijinal verilerden özelliklerin çıkarılması, orijinal verilere STFT uygulanması ve STFT uygulanmış verilerden özelliklerin çıkarılması. KNN, SVM ve 1D WDCNN gibi makine öğrenimi yaklaşımları 1 boyutlu verilere uygulanmıştır. Ayrıca 2 boyutlu veri uzayında STFT dönüşümü uygulanmış ve EfficientNetB0, EfficientNetB1, ResNet18 ve 2D WDCNN kullanılarak farklı istatistiksel metriklerle performans ölçümleri yapılmıştır. 2 boyutlu uzayda derin öğrenme yöntemleri 100 doğruluk elde etmiştir.

Makale detayları

Deep Learning Based Fault Detection from Bearing Vibration Data

Yazarlar:

Engin Oğuzay | Murat Balta

Yayıncı:

Giresun University

Ana Başlıklar:

Bilgisayar Yazılımı

Dergi:

Karadeniz Fen Bilimleri Dergisi Vol.14 Issue 3, pp.1159-1175

Belge Tipi:

Makale

Yayın Tarihi:

9/2024

Belge Dili:

Türkçe

Document ID (Org No.):

2001326797

pISSN:

2564-7377

Motor Bearing Vibration

Deep Learning

Signal Classification

Industrial Fault RecognitionMotor Yatağı Titreşimi

Derin Öğrenme

Sinyal Sınıflandırma

Endüstriyel Arıza Tanıma

Sayı

Makale detayları

Yazarlar: Engin Oğuzay | Murat Balta

Yayıncı: Giresun University

Ana Başlıklar: Bilgisayar Yazılımı

Dergi: Karadeniz Fen Bilimleri Dergisi Vol.14 Issue 3, pp.1159-1175

Belge Tipi: Makale

Yayın Tarihi: 9/2024

Belge Dili: Türkçe

Document ID (Org No.): 2001326797

pISSN: 2564-7377

Sayı

Yazarlar:

Engin Oğuzay | Murat Balta

Yayıncı:

Giresun University

Ana Başlıklar:

Bilgisayar Yazılımı

Dergi:

Karadeniz Fen Bilimleri Dergisi Vol.14 Issue 3, pp.1159-1175

Belge Tipi:

Makale

Yayın Tarihi:

9/2024

Belge Dili:

Türkçe

Document ID (Org No.):

2001326797

pISSN:

2564-7377