Fully interpretable neural network for locating resonance frequency bands for machine condition monitoring

Wang, Dong; Chen, Yikai; Shen, Changqing; Zhong, Jingjing; Peng, Zhike; Li, Chuan

doi:10.1016/j.ymssp.2021.108673

Cited by 102 publications

(22 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The traditional vibration signal analysis methods, such as short time Fourier transform (STFT) [8], wavelet packet transform (WPT) [9], empirical mode decomposition (EMD) [10] and singular value transform (SVD) [11], can process and analyze non-stationary and nonlinear vibration signals, but they cannot fully extract the rich information of fault features contained in the signal, causing the traditional machine learning diagnosis methods unable to meet the requirements of accurate bearing fault classification [12][13][14][15].…”

Section: Introductionmentioning

confidence: 99%

A Width Multi-Scale Adversarial Domain Adaptation Residual Network with AConvolutional Block Attention Module

Yin

Sun

et al. 2023

Preprint

View full text Add to dashboard Cite

Although the fault diagnosis methods based on deep learning have attracted widespread attention in the academic field in recent years, such methods still face many challenges, including complex and variable working conditions, insufficient ability to extract key features, and large differences in sample data. To address these problems, a width multi-scale adversarial domain adaptation residual network with a convolutional block attention module (WMSRCIDANN) is proposed in this paper, which consists of a feature extraction network, a domain discriminant network, and a label classification network. In the feature extraction network, an improved width multi-scale residual network combined with a convolutional block attention module (WMSRC) is used as the feature extractor to achieve a weighted fusion of multi-depth features.In the domain discriminative network, the fully-connected network is replaced by a four-layer convolutional structure, which can further reduce the difference in feature distribution and improve the cross-domain invariance of deep features. In the label classification network, the classifier uses the extracted domain-invariant features to perform cross-domain fault identification. The experimental results show that WMSRCIDANN is effective in cross-domain bearing fault diagnosis.

show abstract

Section: Introductionmentioning

confidence: 99%

A Width Multi-Scale Adversarial Domain Adaptation Residual Network with AConvolutional Block Attention Module

Yin

Sun

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…However, these methods can only partially explain the physical meaning of structure of neural networks. Wang et al [36] proposed a fully explainable neural network structure that provides an illuminating perspective for the development of this field.…”

Section: Introductionmentioning

confidence: 99%

“…Wang et al . [36] proposed a fully explainable neural network structure that provides an illuminating perspective for the development of this field.…”

Section: Introductionmentioning

confidence: 99%

Interpretable fusion methodology of health indices with an application to industrial turbine cavitation condition monitoring

Fu,

Chen,

Wang

et al. 2023

Phil. Trans. R. Soc. A.

Self Cite

View full text Add to dashboard Cite

Using health indices (HIs) to characterize machine conditions is greatly helpful to prevent machine failures and their subsequent catastrophe. Fusion and interpretation of the main contributions of HIs to machine condition monitoring are still challenging. In this paper, an interpretable fusion methodology of HIs is proposed for machine condition monitoring. The proposed methodology begins with elements of statistical learning for classification, following by an essence of how HIs are fused with their associated linear weights to realize machine condition monitoring. One main contribution of this paper gives a theoretical justification for positive and negative weights of the proposed fusion methodology for understanding their importance for machine condition monitoring and making the proposed methodology physically interpretable. In order to be suitable for two practical situations, in which whether faulty data are available or not, two solutions including an offline solution with healthy and faulty datasets and an online solution with only available healthy datasets are suggested to estimate interpretable weights of the proposed methodology. Finally, industrial turbine cavitation status data collected from our group are used to verify the proposed methodology and show its superiority to two existing popular machine fault diagnosis methods. This article is part of the theme issue 'Physics-informed machine learning and its structural integrity applications (Part 2)'.

show abstract

“…On the other hand, the operation of the equipment generates a huge amount of monitoring data, and how to use these monitoring data to meet the high requirements of equipment safety and stability has likewise become an important issue worth studying. Data-driven fault diagnosis methods can build endto-end fault diagnosis models using massive amounts of monitoring data, without being limited to precise physical models and expert knowledge information [5][6][7][8].…”

Section: Introductionmentioning

confidence: 99%

A Cross Working Condition Multiscale Recursive Feature Fusion Method for Fault Diagnosis of Rolling Bearing in Multiple Working Conditions

Zhang

Zhou

2022

IEEE Access

View full text Add to dashboard Cite

As a key component of electromechanical equipment in the intelligent manufacturing process, rolling bearings play an important role to secure a safe, stable, and efficient operation. Deep learning can be used to guide a data-driven fault diagnosis which requires that all data are independently identically distribution (i.i.d). When the equipment is operated with multiple working conditions, the collected samples violates the assumption of i.i.d, which will inevitably make it difficult to extract accurate feature involved in the data. This paper proposes a deep learninag based fault diagnosis model to recursively fuse the multiscale feature on cross working conditions, such that data without working condition label can also be referred to train a satisfying deep learning model for fault diagnosis of bearing operated in multiple working conditions. In the case when only a small number size of training samples for a separated working condition are available, the proposes fusion mechanism aims to establish a jointly learning mechanism between different working conditions. To verify the effectiveness of the proposed algorithm, experimental validation was performed using the Case Western Reserve University (CWRU) rolling bearing public data set. The experimental results show that the proposed method can make full use of a small amount of labeled data with working conditions and a large amount of labeled data without working conditions. In ten types of fault diagnosis tasks with different fault sizes, the fault diagnosis accuracy reaches more than 94% for 4 working conditions and more than 86% for 8 working conditions.INDEX TERMS Fault diagnosis, feature fusion, multiple working conditions, rolling bearing.

show abstract

Fully interpretable neural network for locating resonance frequency bands for machine condition monitoring

Cited by 102 publications

References 34 publications

A Width Multi-Scale Adversarial Domain Adaptation Residual Network with AConvolutional Block Attention Module

A Width Multi-Scale Adversarial Domain Adaptation Residual Network with AConvolutional Block Attention Module

Interpretable fusion methodology of health indices with an application to industrial turbine cavitation condition monitoring

A Cross Working Condition Multiscale Recursive Feature Fusion Method for Fault Diagnosis of Rolling Bearing in Multiple Working Conditions

Contact Info

Product

Resources

About