A novel fault diagnostic system for rolling element bearings using deep transfer learning on bispectrum contour maps

Grover, Chhaya; Turk, Neelam

doi:10.1016/j.jestch.2021.08.006

Cited by 21 publications

(12 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Another fault diagnosis method is based on signal conversion into a moment-invariant S-transform and image Hu processing; that is, using the bearing signal and time-frequency spectrum to create two-dimensional images (Guo, JF et al, 2013) [122]. Deep CNN can use transfer learning to train with bispectrum images of fault signals for fault diagnosis (Chhaya Grover et al, 2022) [123].…”

Section: Other Types Of Vibration Image Methodsmentioning

confidence: 99%

A Review on Rolling Bearing Fault Signal Detection Methods Based on Different Sensors

Yan

Yang

et al. 2022

Sensors

View full text Add to dashboard Cite

As a precision mechanical component to reduce friction between components, the rolling bearing is widely used in many fields because of its slight friction loss, strong bearing capacity, high precision, low power consumption, and high mechanical efficiency. This paper reviews several excellent kinds of study and their relevance to the fault detection of rolling bearings. We summarize the fault location, sensor types, bearing fault types, and fault signal analysis of rolling bearings. The fault signal types are divided into one-dimensional and two-dimensional images, which account for 40.14% and 31.69%, respectively, and their classification is clarified and discussed. We counted the proportions of various methods in the references cited in this paper. Among them, the method of one-dimensional signal detection with external sensors accounted for 3.52%, the method of one-dimensional signal detection with internal sensors accounted for 36.62%, and the method of two-dimensional signal detection with external sensors accounted for 19.72%. The method of two-dimensional signal detection with internal sensors accounted for 11.97%. Among these methods, the highest detection rate is 100%, and the lowest detection rate is more than 70%. The similarities between the different methods are compared. The research results summarized in this paper show that with the progress of the times, a variety of new and better research methods have emerged, which have sped up the detection and diagnosis of rolling bearing faults. For example, the technology using artificial intelligence is still developing rapidly, such as artificial neural networks, convolutional neural networks, and machine learning. Although there are still defects, such methods can quickly discover a fault and its cause, enrich the database, and accumulate experience. More and more advanced techniques are applied in this field, and the detection method has better robustness and superiority.

show abstract

Section: Other Types Of Vibration Image Methodsmentioning

confidence: 99%

A Review on Rolling Bearing Fault Signal Detection Methods Based on Different Sensors

Yan

Yang

et al. 2022

Sensors

View full text Add to dashboard Cite

show abstract

“…Wen et al used the VGG-19 architecture in [14] and Inception V3 and TrAdaBoost as feature extractors in [15]. Grover et al [4] utilized bi-spectrum contour maps of the vibration signals in four pretrained networks comprising Alexnet, VGG-19, GoogleNet, and Resnet-50. These architectures are limited by their large number of layers and high computational complexity.…”

Section: Literature Reviewmentioning

confidence: 99%

“…Its high computation cost is infeasible for most applications while the time-frequency space partitioning leads to symmetry loss. In [4], bi-spectrum-based higher-order analysis was utilized to extract distinct signal patterns under inconsistent working conditions. Although the traditional bisector representation permits the phase information to be included and eliminates Gaussian noise, the results are unstable because of the randomly changing phase components of the signals.…”

Section: Introductionmentioning

confidence: 99%

“…To overcome the limitations of DL models, transfer learning (TL) was incorporated into DL models so that knowledge from one problem can be used to solve a related different problem [4,[13][14][15][16][17][18][19][20]. For example, pre-trained ImageNet weights have been used to initialize the parameters in deep networks for classifying industrial faults instead of random initialization.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Hybrid deep transfer learning architecture for industrial fault diagnosis using Hilbert transform and DCNN–LSTM

2022

View full text Add to dashboard Cite

Early-stage fault detection has become an indispensable part of modern industry to prevent potential hazards or sudden hindrances to the production process. With the advent of deep learning (DL) applications in several fields, DL models have been used to classify faults in specific environments. Uniform texture extraction has been performed using transformed-signal processing techniques and deep transfer learning (DTL) architectures in a few studies. Traditional signal processing techniques encounter difficulties in extracting distinct fault features due to the nonlinear and non-stationary nature of the time-series fault data. In this paper, a hybrid DTL architecture comprising a deep convolutional neural network and long short-term memory layers for extracting both temporal and spatial features enhanced by Hilbert transform 2D images is presented. Three standard audio sound fault datasets comprising the malfunctioning industrial machine investigation and inspection dataset, toy anomaly detection in machine operating sounds dataset, and machinery failure prevention technology bearing vibration fault dataset with various loads and noisy environments were utilized in the experimental evaluation. The proposed model with an input size of 32 × 32 achieved an average F1 score of 0.998 on the tested datasets. The implementation of transfer learning using the three benchmark datasets resulted in the highest accuracy of the proposed model and over fivefold reduction in the training epochs. In addition, the proposed model outperformed the state-of-art models in accuracy in various environments.

show abstract

“…Therefore, domain adaptation measures have been proposed to prevent negative transfer or mitigate its adverse impact, including source data filtering using an adversarial approach [22], new architectural design [23][24][25], the use of transitive transfer strategy [26], etc. Although many negative transfer countermeasures have been developed, transferring a pretrained model directly from a distant inter-domain to the target domain still poses problems [27,28]. More specifically, there is a knowledge gap on how to mitigate negative transfer in transfer learning with small samples in fault diagnosis tasks.…”

Section: Introductionmentioning

confidence: 99%

Intra-Domain Transfer Learning for Fault Diagnosis with Small Samples

Zhang

Peng

et al. 2022

Applied Sciences

View full text Add to dashboard Cite

The concept of deep transfer learning has spawned broad research into fault diagnosis with small samples. A considerable covariate shift between the source and target domains, however, could result in negative transfer and lower fault diagnosis task accuracy. To alleviate the adverse impacts of negative transfer, this research proposes an intra-domain transfer learning strategy that makes use of knowledge from a data-abundant source domain that is akin to the target domain. Concretely, a pre-trained model in the source domain is built via a vanilla transfer from an off-the-shelf inter-domain deep neural network. The model is then transferred to the target domain using shallow-layer freezing and finetuning with those small samples. In a case study involving rotating machinery, where we tested the proposed strategy, we saw improved performance in both training efficiency and prediction accuracy. To demystify the learned neural network, we propose a heat map visualization method using a channel-wise average over the final convolutional layer and up-sampling with interpolation. The findings revealed that the most active neurons coincide with the corresponding fault characteristics.

show abstract

A novel fault diagnostic system for rolling element bearings using deep transfer learning on bispectrum contour maps

Cited by 21 publications

References 25 publications

A Review on Rolling Bearing Fault Signal Detection Methods Based on Different Sensors

A Review on Rolling Bearing Fault Signal Detection Methods Based on Different Sensors

Hybrid deep transfer learning architecture for industrial fault diagnosis using Hilbert transform and DCNN–LSTM

Intra-Domain Transfer Learning for Fault Diagnosis with Small Samples

Contact Info

Product

Resources

About