Convolutional neural network based bearing fault diagnosis of rotating machine using thermal images

Choudhary, Anurag; Mian, Tauheed; Fatima, Shahab

doi:10.1016/j.measurement.2021.109196

Cited by 187 publications

(78 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Classical features have been considered for the analysis [24,25], which have proved sensitive to changes in operating conditions, both in case 1 and in case 2: -Root mean square (RMS): for a vector of n elements…”

Section: Methodsmentioning

confidence: 99%

Use of MEMS sensors for condition monitoring of devices: discussion about the accuracy of features for diagnosis

D’Emilia

Natale

2021

Int. J. Metrol. Qual. Eng.

View full text Add to dashboard Cite

This paper analyses the effect of the variability of metrological characteristics of a set of low-cost Micro Electro-Mechanical Systems (MEMS) for the acceleration measurement, on the calculation of typical features used for condition monitoring (CM) of automatic production lines. The knowledge of the contribution of the variability of metrological characteristics to the final accuracy of features is an aspect of interest when networks of low-cost sensors are used, in particular in case the variability of their characteristics is high. In fact, due to a mass production, the calibration is not carried out sensor by sensor, but the characteristics are determined on a sample basis and assigned to the entire batch. Neglecting the variability between sensors can lead to effects on the results of data analysis, which are not easily predictable. In this paper, the real variability of the sensor's characteristics, experimentally evaluated through the calibration of a set of 25 low-cost MEMS accelerometers, has been taken into account. Digital sensitivity, signal-to-noise ratio and data rate variability of each device have been considered for the analysis. The analysis has been carried out with reference to two different test cases of industrial interest, by modifying the real outputs of high performance piezoelectric accelerometers used for CM, in order to simulate the effect of the metrological characteristics of MEMS sensors. The results show which features, among those typically used for CM, are more affected and which characteristics of MEMS are more influencing the features themselves, with reference to the specific considered applications.

show abstract

Section: Methodsmentioning

confidence: 99%

Use of MEMS sensors for condition monitoring of devices: discussion about the accuracy of features for diagnosis

D’Emilia

Natale

2021

Int. J. Metrol. Qual. Eng.

View full text Add to dashboard Cite

show abstract

“…These failures can also cause serious consequences and should not be ignored. Infrared thermal image can perfectly reflect non-structural fault information and is widely applied in fault diagnosis [ 26 , 27 ]. However, the single infrared thermal image is very sensitive and is easily affected by external factors such as oil temperature [ 28 ].…”

Section: Introductionmentioning

confidence: 99%

Fusion Domain-Adaptation CNN Driven by Images and Vibration Signals for Fault Diagnosis of Gearbox Cross-Working Conditions

Mao

Zhang

Qiao

et al. 2022

Entropy

View full text Add to dashboard Cite

The vibration signal of gearboxes contains abundant fault information, which can be used for condition monitoring. However, vibration signal is ineffective for some non-structural failures. In order to resolve this dilemma, infrared thermal images are introduced to combine with vibration signals via fusion domain-adaptation convolutional neural network (FDACNN), which can diagnose both structural and non-structural failures under various working conditions. First, the measured raw signals are converted into frequency and squared envelope spectrum to characterize the health states of the gearbox. Second, the sequences of the frequency and squared envelope spectrum are arranged into two-dimensional format, which are combined with infrared thermal images to form fusion data. Finally, the adversarial network is introduced to realize the state recognition of structural and non-structural faults in the unlabeled target domain. An experiment of gearbox test rigs was used for effectiveness validation by measuring both vibration and infrared thermal images. The results suggest that the proposed FDACNN method performs best in cross-domain fault diagnosis of gearboxes via multi-source heterogeneous data compared with the other four methods.

show abstract

“…The statistical method measures physical data containing bearing fault information such as heat, pressure, sound, vibration, etc., using various sensors, and analyzing the measured data with numerical analysis to diagnose defects statistically. [4][5][6][7][8] The machine learning methods such as artificial neural networks (ANNs), support vector machines (SVMs), and random forests utilize the measured physical data as training data. 6,[8][9][10][11] Using the statistical method, information on bearing faults can be extracted from heat, pressure, sound, and vibration data.…”

Section: Introductionmentioning

confidence: 99%

“…6,[8][9][10][11] Using the statistical method, information on bearing faults can be extracted from heat, pressure, sound, and vibration data. For example, Seo et al 4 and Choudhary et al 5 diagnosed the condition of bearing wear under dynamic load using an infrared thermal imaging camera. The acoustic analysis method based on the acoustic emission signal uses the root mean square value, amplitude, energy, and counts of acoustic emissions to detect the bearing fault.…”

Section: Introductionmentioning

confidence: 99%

Effect of data preprocessing methods and hyperparameters on accuracy of ball bearing fault detection based on deep learning

Kim

Lee

Jang

et al. 2022

Advances in Mechanical Engineering

View full text Add to dashboard Cite

This paper presents the effect of data preprocessing methods and hyperparameters in deep learning on the accuracy of ball bearing fault detection. In this study, artificial defects in the ball bearing were created to obtain the machine learning data for ball bearing fault detection. Vibration data were acquired by an accelerometer mounted in the bearing housing at three different rotation speeds. The obtained one-dimensional acceleration-based vibration data were changed into five different data forms: one-dimensional fast Fourier transform data, two-dimensional spectrogram image data, etc. One-dimensional numerical data were used as training data in the multi-layer perceptron and two-dimensional image data in the convolutional neural network classifier. After training, the accuracy and effectiveness of the validation test and the training data formats and deep learning models are discussed in this paper. 1D time- and frequency-domain numerical data showed 100% accuracy within the same rotation speed, but the accuracy was down to less than 50% in the mixed rotation speeds. On the other hand, 2D frequency-domain image data presented more than 99% accuracy for the mixed rotation speeds. Among 2D image data, FFT-image data is less sensitive to hyperparameters such as kernel and convolution layer and shows high test accuracy of 99% at least. Consequently, 2D image data format with the convolutional neural network more accurately worked in a complicated situation.

show abstract

Convolutional neural network based bearing fault diagnosis of rotating machine using thermal images

Cited by 187 publications

References 28 publications

Use of MEMS sensors for condition monitoring of devices: discussion about the accuracy of features for diagnosis

Use of MEMS sensors for condition monitoring of devices: discussion about the accuracy of features for diagnosis

Fusion Domain-Adaptation CNN Driven by Images and Vibration Signals for Fault Diagnosis of Gearbox Cross-Working Conditions

Effect of data preprocessing methods and hyperparameters on accuracy of ball bearing fault detection based on deep learning

Contact Info

Product

Resources

About