ClassBD: A New Method for Enhanced Bearing Fault Diagnosis in Noisy Environments

Table of Links

Abstract and 1. Introduction

2. Preliminaries and 2.1. Blind deconvolution

2.2. Quadratic neural networks

3. Methodology

3.1. Time domain quadratic convolutional filter

3.2. Superiority of cyclic features extraction by QCNN

3.3. Frequency domain linear filter with envelope spectrum objective function

3.4. Integral optimization with uncertainty-aware weighing scheme

4. Computational experiments

4.1. Experimental configurations

4.2. Case study 1: PU dataset

4.3. Case study 2: JNU dataset

4.4. Case study 3: HIT dataset

5. Computational experiments

5.1. Comparison of BD methods

5.2. Classification results on various noise conditions

5.3. Employing ClassBD to deep learning classifiers

5.4. Employing ClassBD to machine learning classifiers

5.5. Feature extraction ability of quadratic and conventional networks

5.6. Comparison of ClassBD filters

6. Conclusions

Appendix and References

6. Conclusions

In this study, we have introduced a novel approach termed as ClassBD for bearing fault diagnosis under heavy noisy conditions. ClassBD is composed of cascaded time and frequency neural BD filters, succeeded by a deep learning

classifier. Specifically, the time BD filter incorporates quadratic convolutional neural networks (QCNN), and we have mathematically proved its superior capability in extracting periodic impulse features in the time domain. The frequency BD filter includes a fully-connected linear filter, supplementing the frequency domain filter subsequent to the time filter. Furthermore, a deep learning classifier is directly integrated to empower classification capabilities. We have devised a physics-informed loss function composed of kurtosis, 𝑙2∕𝑙4 norm, and cross-entropy loss to facilitate the joint learning. This unified framework transforms traditional unsupervised BD into supervised learning, providing interpretability due to its retention of conventional BD operations. Finally, comprehensive experiments conducted on three public and private datasets reveal that ClassBD outperforms other state-of-the-art methods. ClassBD represents the first BD method that can be directly applied to classification and it exhibits a good noise resistance, portability, and interoperability. Therefore, ClassBD holds a great potential for further generalization on other difficult tasks such as cross-domain and small sample issues in future research.