An unsupervised bearing fault diagnosis based on deep subdomain adaptation under noise and variable load condition

As one of the mainstream methods for transfer learning, Correlation Alignment (CORAL) has been widely applied in the field of fault diagnosis and has achieved certain achievements. However, CORAL ignores the differences between domain expectations in the matching process, which makes it difficult to accurately measure the discrepancies between domains. To compensate for the shortcomings of the CORAL method, this paper proposes a new feature correlation matching (FCM) method, and further uses it as an objective function to propose a deep feature correlation matching network (DFCMN). The FCM method focuses on both first-order feature correlation and second-order feature correlation of the source and target domains, which can measure the discrepancies between different domains more comprehensively and accurately. With the powerful feature mapping capability of neural network, DFCMN can improve the feature similarity in different domain centers while reducing the discrepancies of feature distribution between different domains, so as to obtain more reliable shared features and improve the cross-work-conditions diagnosis accuracy. The effectiveness of the proposed method was verified under multiple transfer tasks using the public rolling bearing dataset.

Section: Introductionmentioning

confidence: 99%

A novel transfer learning fault diagnosis method for rolling bearing based on feature correlation matching

Wang

Ning

2022

“…Thus, 1D-CNN is suitable for fault diagnosis based on the vibration signal. Meanwhile, in some new fault diagnosis areas, where there is a lack of samples [23], variable conditions [24] and minimal or no supervision [25], etc, 1D-CNN has also been more widely used.…”

Section: Introductionmentioning

confidence: 99%

Grouping sparse filtering: a novel down-sampling approach toward rotating machine intelligent diagnosis in 1D-convolutional neural networks

Wang¹,

Song²,

Wang³

et al. 2022

Convolutional neural networks (CNNs) has weight sharing and feature learning ability, which can efficiently and effectively complete the health monitoring of industrial equipment. However, the pooling operation in a typical CNN can cause a loss of valuable impulse features during data down-sampling. We propose grouping sparse filtering (GSF) to overcome this problem. Instead of using a pooling operation, the GSF splits the channels of feature obtained after convolution into equal-length groups. A feature selector with a feature aggregation function based on the channel importance factors and a lasso constraint is used to filter the groups to perform down-sampling. The GSF method preserves the impulse features because of the block sparsity of the vibration signal. Theoretical analysis demonstrates that the GSF has a similar computational complexity as using a pooling layer in a CNN for the same number of layers. Two experimental studies were conducted using data from a laboratory test and industrial environments. The experimental results show that the 1D-CNN with GSF provides better performance for retaining the impulse features of the rotating machinery signals and higher fault identification accuracy than a CNN with a pooling layer.

“…Based on the pseudo test output of the nearest-neighbor classifier, Tong et al [14] refined the diagnosis model by continuously reducing the MMD between domains, so as to learn representation of the transferable fault features. Correspondingly, moment matching is also popular in deep-learning based DA diagnosis [21,22]. In [23], a domain adaptive convolutional neural network (DACNN) model is designed, which extracts the transferable fault features by minimizing the squared MMD between the feature representations of the source and target domains.…”

Section: Introductionmentioning

confidence: 99%

A game theory enhanced domain adaptation network for mechanical fault diagnosis

Sun¹,

Liu²,

Liu³

et al. 2022

Transfer learning (TL) technology have been successfully applied to address the domain adaptation (DA) problem in machinery fault diagnosis. However, partial DA problem is more suitable for industrial applications, where the target data only covers a subset of the source classes, which makes it difficult to know where to transfer the target data. To overcome this problem, a novel game theory enhanced domain adaptation network (GT-DAN) is proposed in this paper. Based on different metrics, including Maximum mean discrepancy (MMD), Jensen-Shannon (JS) divergence and Wasserstein distance, three attention matrices are constructed to describe the distribution discrepancies between the source domain and the target domain, and the optimal coordination between these attention matrices is achieved by the combination weighting based on game theory to generate the optimal probability weights, which can guide to filter out the irrelevant source examples in domain adaptation. Two experiments show that the proposed GT-DAN is superior to the existing methods in the partial DA diagnosis performance.