Sample Augmentation Using Enhanced Auxiliary Classifier Generative Adversarial Network by Transformer for Railway Freight Train Wheelset Bearing Fault Diagnosis DOI Creative Commons
Jing Zhao, Junfeng Li, Zonghao Yuan

et al.

Entropy, Journal Year: 2024, Volume and Issue: 26(12), P. 1113 - 1113

Published: Dec. 20, 2024

Diagnosing faults in wheelset bearings is critical for train safety. The main challenge that only a limited amount of fault sample data can be obtained during high-speed operations. This scarcity samples impacts the training and accuracy deep learning models bearing diagnosis. Studies show Auxiliary Classifier Generative Adversarial Network (ACGAN) demonstrates promising performance addressing this issue. However, existing ACGAN have drawbacks such as complexity, high computational expenses, mode collapse, vanishing gradients. Aiming to address these issues, paper presents Transformer (TACGAN), which increases diversity, complexity entropy generated samples, maximizes samples. transformer network replaces traditional convolutional neural networks (CNNs), avoiding iterative structures, thereby reducing expenses. Moreover, an independent classifier integrated prevent coupling problem, where discriminator simultaneously identified classified ACGAN. Finally, Wasserstein distance employed loss function mitigate collapse Experimental results using datasets demonstrate effectiveness TACGAN.

Language: Английский

VibrMamba: A lightweight Mamba based fault diagnosis of rotating machinery using vibration signal DOI
Haiming Yi, Danyu Li, Zhenyong Lu

et al.

Measurement, Journal Year: 2025, Volume and Issue: unknown, P. 116881 - 116881

Published: Feb. 1, 2025

Language: Английский

Citations

0
Variable Filtered-Waveform Variational Mode Decomposition and Its Application in Rolling Bearing Fault Feature Extraction DOI Creative Commons

Nuo Li,

Hang Wang

Entropy, Journal Year: 2025, Volume and Issue: 27(3), P. 277 - 277

Published: March 7, 2025

Variational Mode Decomposition (VMD) serves as an effective method for simultaneously decomposing signals into a series of narrowband components. However, its theoretical foundation, the classical Wiener filter, exhibits limited adaptability when applied to broadband signals. This paper proposes novel Variable Filtered-Waveform (VFW-VMD) address critical limitations in VMD, particularly handling and chirp By incorporating fractional-order constraints dynamically adjusting filter waveforms, proposed algorithm effectively mitigates mode mixing over-smoothing issues. The mathematical framework VFW-VMD is formulated, decomposition performance validated through simulations involving both synthetic real-world results demonstrate that superior extracting captures more rolling bearing fault features. work advances signal processing techniques, enhancing capability significantly improving practical diagnostic applications.

Language: Английский

Citations

0
A Novel Multi-Time Scale Heat Load Prediction Model for District Heating System: Hybrid Subtraction Average Based Optimizer (Sabo) and Cnn-Bilstm Model with Attention Mechanism DOI
Xuyang Cui,

Junda Zhu,

Lifu Jia

et al.

Published: Jan. 1, 2024

Accurate and reliable heating load prediction is a prerequisite for the efficient operation of district systems (DHS) basis demand-based heat supply. However, considering high time lag complexity DHS, ability to strengthen bidirectional long short-term memory (BilSTM) model using convolutional neural network (CNN) as well attention mechanism (ATT) DHS has not been effectively demonstrated. A novel multi-time scale (SABO-CNN-ATT-BiLSTM) was proposed, which hybrid subtractive averaging based optimizer (SABO), CNN, ATT, BilSTM. The tested in comparison with BiLSTM model, CNN-BiLSTM CNN-ATT-BiLSTM model. test object 2880-hour dateset real system. results show that SABO-CNN-ATT-BiLSTM better accuracy than other models. an R2 equal 0.954 MAE 0.0241 on set, closer predicted values no significant deviation from data. Also, three models evaluated at different scales (1 hour, 6 hours, 12 24 48 72 hours). Compared models, shows superior performance scales. can adaptively adjust hyperparameters find optimal parameter configuration improve overall more accurate stable scheme field nonlinearity, thermal inertia buildings.

Language: Английский

Citations

0
Sample Augmentation Using Enhanced Auxiliary Classifier Generative Adversarial Network by Transformer for Railway Freight Train Wheelset Bearing Fault Diagnosis DOI Creative Commons
Jing Zhao, Junfeng Li, Zonghao Yuan

et al.

Entropy, Journal Year: 2024, Volume and Issue: 26(12), P. 1113 - 1113

Published: Dec. 20, 2024

Diagnosing faults in wheelset bearings is critical for train safety. The main challenge that only a limited amount of fault sample data can be obtained during high-speed operations. This scarcity samples impacts the training and accuracy deep learning models bearing diagnosis. Studies show Auxiliary Classifier Generative Adversarial Network (ACGAN) demonstrates promising performance addressing this issue. However, existing ACGAN have drawbacks such as complexity, high computational expenses, mode collapse, vanishing gradients. Aiming to address these issues, paper presents Transformer (TACGAN), which increases diversity, complexity entropy generated samples, maximizes samples. transformer network replaces traditional convolutional neural networks (CNNs), avoiding iterative structures, thereby reducing expenses. Moreover, an independent classifier integrated prevent coupling problem, where discriminator simultaneously identified classified ACGAN. Finally, Wasserstein distance employed loss function mitigate collapse Experimental results using datasets demonstrate effectiveness TACGAN.

Language: Английский

Citations

0