Imbalanced Fault Diagnosis of Rolling Bearing Using Data Synthesis Based on Multi-Resolution Fusion Generative Adversarial Networks

Abstract

Fault diagnosis of industrial bearings plays an invaluable role in the health monitoring of rotating machinery. In practice, there is far more normal data than faulty data, so the data usually exhibit a highly skewed class distribution. Algorithms developed using unbalanced datasets will suffer from severe model bias, reducing the accuracy and stability of the classification algorithm. To address these issues, a novel Multi-resolution Fusion Generative Adversarial Network (MFGAN) is proposed for the imbalanced fault diagnosis of rolling bearings via data augmentation. In the data-generation process, the improved feature transfer-based generator receives normal data as input to better learn the fault features, mapping the normal data into fault data space instead of random data space. A multi-scale ensemble discriminator architecture is designed to replace original single discriminator structure in the discriminative process, and multi-scale features are learned via ensemble discriminators. Finally, the proposed framework is validated on the public bearing dataset from Case Western Reserve University (CWRU), and experimental results show the superiority of our method.

Improved Conditional Domain Adversarial Networks for Intelligent Transfer Fault Diagnosis

“…The baseline CNN parameters are consistent with Table 1, and there is still room to improve its FD performance in the presence of class imbalance. In contrast to CNN and ICDAN-F, WGANML [48] and MFGAN [49] focus on mitigating the effects of class imbalance via generative training, balancing the number of fault samples before proceeding to FD. The average accuracy results suggest that using a generative approach to the class imbalance problem may lead to more promising results.…”

“…Whether to Generate Fault Samples Average Accuracies (%) 1 CNN (Baseline) No 98.95 WGANML [48] Yes 99.89 MFGAN [49] Yes 100 ICDAN-F (Ours)…”

Generation of high-dimensional vibration signal and its application in fault diagnosis

“…Generative adversarial network (GAN) has emerged as the most popular oversampling method, and various variants have been introduced and applied to fault diagnosis, including the Wasserstein GAN (WGAN) [32,33], Wasserstein GAN with a gradient penalty (WGAN-GP) [34,35], conditional GAN (CGAN) [36,37], and auxiliary classifier GAN [38,39]. To enhance the quality of generated data, researchers have focused on optimizing the structure and loss function of GANs.…”

“…Separate classification networks have been incorporated to improve the feature extraction capabilities of GANs [40]. Additionally, long short-term memory networks [41], variational autoencoders [42], residual networks [39], and capsule networks [43] have been introduced to enhance the generative performance of the generators. Common training techniques such as attention mechanisms [44], normalization [45], and dropout [46] are also utilized.…”