A Novel Symmetric Stacked Autoencoder for Adversarial Domain Adaptation Under Variable Speed

Abstract

At present, most of the fault diagnosis methods with extensive research and good diagnostic effect are based on the premise that the sample distribution is consistent. However, in reality, the sample distribution of rotating machinery is inconsistent due to variable working conditions, and most of the fault diagnosis algorithms have poor diagnostic effects or even invalid. To dispose the above problems, a novel symmetric stacked autoencoder (NSSAE) for adversarial domain adaptation is proposed. Firstly, the symmetric stacked autoencoder network with shared weights is used as the feature extractor to extract features which can better express the original signal. Secondly, adding domain discriminator that constituting adversarial with feature extractor to enhance the ability of feature extractor to extract domain invariant features, thus confusing the domain discriminator and making it unable to correctly distinguish the features of the two domains. Finally, to assist the adversarial training, the maximum mean discrepancy (MMD) is added to the last layer of the feature extractor to align the features of the two domains in the high-dimensional space. The experimental results show that, under the condition of variable speed, the NSSAE model can extract domain invariant features to achieve the transfer between domains, and the transfer diagnosis accuracy is high and the stability is strong.

Autoencoder-Enhanced Clustering: A Dimensionality Reduction Approach to Financial Time Series

“…There are multiple forms of AE [12], such as those that are constructed using restricted Boltzmann machine [13] and deep learning structures [14] which change according to the complexity and arrangement of the data. Examples included fully connected deep AEs, convolutional neural networks AEs and recurrent neural network (RNN) AEs, which can be applied to different tasks and fields such as feature extraction for econometrics [15], fault diagnosis [16], fraud detection FIGURE 1. The structure of an AE [17], genetics [18], image processing [19], language translation [20], remote sensing [21], robotics [22],and security [23] among others.…”

The Surprising Effectiveness of Deep Orthogonal Procrustes Alignment in Unsupervised Domain Adaptation

“…Successful state-of-the art methods use powerful feature extractors such as convolutional neural networks (CNNs), and aim to jointly minimize source error along with domain divergence error. Adversarial learning [22] has been the workhorse in these solutions, and can be implemented with different additional regularizers [11], [23]- [27]. Additionally some recent works have considered the use of pseudo-labeling and self-training [28]- [31], on top of these regularization strategies to further boost the UDA performance.…”

Adversarial Deep Transfer Learning in Fault Diagnosis: Progress, Challenges, and Future Prospects

“…Li et al [ 72 ] used two feature extractors and classifiers, trained using MMD and domain adversarial training, respectively, and used ensemble learning to obtain the final results. Li et al [ 73 , 74 ] aligned the target domain features with the source domain features by adding MMD in the feature extraction stage. Zhou et al [ 75 ] and Wan et al [ 76 ] used MK-MMD and domain discriminators to adjust the edge and conditional distributions.…”