A neural network compression method based on knowledge-distillation and parameter quantization for the bearing fault diagnosis

Ji, Mengyu; Peng, Gaoliang; Li, Sijue; Cao, Feng; Chen, Zhao; Li, Zhixiong; Du, Haiping

doi:10.1016/j.asoc.2022.109331

Cited by 35 publications

(13 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The slight difference is only in accuracy, precision and recall due to different ways of working. Specifically, the compression methods examined are Naive, DoReFa, BNN, LSQ, Observer [50], and PTQ [24]. We find that the efficiency of the quantizers does not differ too much, ranging from 92.47% to 96.95% with a margin of only 4.5%.…”

Section: Model Quantization and Classification Resultsmentioning

confidence: 99%

“…In addition, one of the new trends in bearing fault detection and diagnosis is optimizing the machine learning model (model capacity, computational resources, inference latency) so that it can be deployed on constrained hardware [17]. Regarding this trend, there are several popular approaches such as neural network architecture search [18][19][20][21] or model compression using pruning [17,[21][22][23], quantization [23][24][25][26], and knowledge distillation [24,[27][28][29]. Each approach will have different advantages and disadvantages, so how to use them should be carefully considered.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Efficient bearing fault diagnosis with neural network search and parameter quantization based on vibration and temperature

et al. 2023

View full text Add to dashboard Cite

In this work, we propose a deep-learning method to diagnose bearing faults of electric motors based on vibration and bearing housing temperature. Our methods can accurately diagnose faults related to bearing cracking and lubricant shortages. The proposed method is effective in terms of computational complexity and model capacity thanks to the advantages of neural architecture search (NAS) and parameter quantization in the model establishment. The experimental results found that the information on bearing temperature improved the diagnostic accuracy for the bearing fault diagnosis task. The proposed method has explored the most optimal model in terms of computational resources and model capacity with a pre-defined accuracy target. The searched model has a relatively high diagnostic accuracy of 98.7% and a size of about 27.3 kB. After quantization, the obtained model maintained 96.9% accuracy and reduced 4 times in size. All experiments are executed elaborately on our custom bearing fault dataset.

show abstract

Section: Model Quantization and Classification Resultsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Efficient bearing fault diagnosis with neural network search and parameter quantization based on vibration and temperature

et al. 2023

View full text Add to dashboard Cite

show abstract

“…MSCF are used to enhance the fault signal, then the multisensor data are fused by MCNN for feature extraction and fault pattern classification. Ji et al [14] proposed a CNN compression method based on knowledge-distillation (K-D) and parameter quantization to facilitate rolling bearing fault diagnosis on an embedded platform. However, the current application of CNN in the field of fault diagnosis has the following problems: (1) the existing CNN has large scale and large parameters, which is difficult to be mounted on a small embedded platform; (2) Current model compression methods, such as network pruning, quantization and knowledge distillation, will inevitably reduce the performance of network recognition.…”

Section: Introductionmentioning

confidence: 99%

Rotating machinery fault diagnosis using dimension expansion and AntisymNet lightweight convolutional neural network

Luo

Peng

Dong

et al. 2023

Meas. Sci. Technol.

View full text Add to dashboard Cite

Deep learning-based methods have made remarkable progress in the field of fault diagnosis for rotating machinery. However, convolutional neural networks (CNNs) are not suitable for industrial applications due to their large model size and high computational complexity. To address this limitation, this paper proposes the Antisym module and constructs AntisymNet, which is combined with dimension expansion algorithms for fault diagnosis of rotating machinery. To begin with, the original vibration signal of the rolling machinery is subjected to time-frequency transformations using the Discrete Fourier Transform (DFT) and Discrete Wavelet Transform (DWT). Subsequently, each transformed time-frequency signal is expanded in dimensions, resulting in two-dimensional matrix single channel images. These single channel images are then fused into RGB images to enhance the sample features. Finally, the proposed AntisymNet is utilized for recognizing and classifying the expanded signals. To evaluate the performance of AntisymNet, the MiniImageNet image dataset is employed as a benchmark, and a comparison is made with other state-of-the-art lightweight convolutional neural networks (L-CNNs). Additionally, the effectiveness of the proposed fault diagnosis model is validated using the CWRU bearing dataset, Ottawa bearing dataset, and the hob dataset. The model achieves an impressive accuracy rate of 99.70% in the CWRU dataset, 99.26% in the Ottawa dataset, and an error rate of only 0.66% in the hob dataset. These results demonstrate the strong performance of the proposed fault diagnosis model.

show abstract

“…Shen and Guo 37 developed a modified probabilistic K-D method for the bearing fault diagnosis. Ji et al 38 combined the K-D method and the parameter quantization to compress the bearing fault diagnosis networks. In these works, the K-D process was proven to be able to improve the classification accuracy of the lightweight student networks.…”

Section: Introductionmentioning

confidence: 99%

Iterative-AMC: a novel model compression and structure optimization method in mechanical system safety monitoring

Peng

et al. 2023

Structural Health Monitoring

Self Cite

View full text Add to dashboard Cite

With the rapid development of artificial intelligence, various fault diagnosis methods based on the deep neural networks have made great advances in mechanical system safety monitoring. To get the high accuracy for the fault diagnosis, researchers tend to adopt the deep network layers and amount of neurons or kernels in each layer. This results in a large redundancy and the structure uncertainty of the fault diagnosis networks. Moreover, it is hard to deploy these networks on the embedded platforms because of the large scales of the network parameters. This brings huge challenges to the practical application of the intelligent diagnosis algorithms. To solve the above problems, an iterative automatic machine compression method, named Iterative-AMC, is proposed in this paper. The proposed method aims to automatically compress and optimize the structure of the large-scale neural networks. Experiments are carried out based on two test benches. With the proposed Iterative-AMC method, the problems of the parameter redundancy and the structure uncertainty can be addressed. The scale of the original network can be greatly compressed, and the compressed fault diagnosis network is successfully deployed on a small-scale FPGA chip.

show abstract

A neural network compression method based on knowledge-distillation and parameter quantization for the bearing fault diagnosis

Cited by 35 publications

References 44 publications

Efficient bearing fault diagnosis with neural network search and parameter quantization based on vibration and temperature

Efficient bearing fault diagnosis with neural network search and parameter quantization based on vibration and temperature

Rotating machinery fault diagnosis using dimension expansion and AntisymNet lightweight convolutional neural network

Iterative-AMC: a novel model compression and structure optimization method in mechanical system safety monitoring

Contact Info

Product

Resources

About