23 January 2020 in IEEE Access
Electroencephalogram (EEG) contains important physiological information that can reflect the activity of human brain, making it useful for epileptic seizure detection and epilepsy diagnosis. However visual inspection of large amounts of EEG by human expert is time-consuming, and meanwhile there are often inconsistencies in judgement between physicians. In this paper, we develop a unified framework for early epileptic seizure detection and epilepsy diagnosis, which includes two phases. In the first phase, the signal intensity is first calculated for each data point of the given EEG, enabling the well-known autoregressive moving average (ARMA) model to characterize the dynamic behavior of the EEG time series. The residual error between the predicted value of learned ARMA model and the actually observed value is used as the anomaly score to support a null hypothesis testing for making epileptic seizure decision. The epileptic seizure detection phase can provide a quick detection for anomaly EEG patterns, but the resulting suspicious segment may include epilepsy or other disordering EEG activities thus required to be identified. Therefore, in the second phase, we use pattern recognition technique to classify the suspicious EEG segments. In particular, we propose a new and practical classifier based on a pairwise of one-class SVMs for epilepsy diagnosis. The proposed classifier requires normal and epilepsy data for training, but can recognize normal, epilepsy and even other disorders that would not be trained in the training samples. This point is practical and meaningful in real clinic scenarios as the input EEG may include other brain disordering diseases besides of epilepsy. We conducted experiments on the publicly-available Bern-Barcelona and CHB-MIT EEG database, respectively, to validate the effectiveness of the proposed framework, and our method achieved classification accuracy of 93% and 94% on them. Comprehensive experimental results, outperforming the state-of-...
21 August 2023 in Knowledge-Based Systems
10 May 2022 in IEEE Sensors Journal
Although machine learning methods have demonstrated their effectiveness in fault diagnosis in rotating machinery, there is a major assumption that the training data (source domain) and testing data (target domain) should share the same distribution. However, this assumption is difficult to hold in real scenarios considering the variable working conditions, and it recasts the fault diagnosis problem in a cross-domain manner. Recently, the adversarial domain adaptation methods have become a hot research topic, since they aim to address cross-domain issues and can be well embedded into convolutional neural networks. Most previous studies aimed to achieve the optimal alignment of data in a global view. Unfortunately, they may affect the data which are originally well aligned in the local view between the source domain and the target domain, thus leading to diminished diagnosis performance. In this paper, a pair-wise orthogonal classifier based domain adaptation network is proposed to address this issue. A feature extractor together with a pair-wise orthogonal classifier is designed to learn domain-invariant features from the source domain and the target domain. Then, based on the outputs of the pair-wise classifier, a dynamic weighted domain discriminator is designed to form an adversarial framework with a feature extractor. It considers the sample-level alignment in the domain adaptation process and enables the global alignment without sacrificing the original well-aligned data. Cross-domain experiments via two datasets are carried out to validate the performance of the proposed network. Performance comparisons with state-of-the-art methods are also made. The results have demonstrated the effectiveness and novelty of the proposed network.
30 October 2024 in IEEE Sensors Journal
Features from different domains in vibration signals offer valuable insights for remaining useful life (RUL) prediction of bearings. While fusing these features can improve prediction performance, traditional fusion methods lack effective information exchange across domains, limiting adaptive feature fusion. This limitation can lead to information redundancy and hinder the accurate identification of bearing degradation states. To address these challenges, this study introduces a dual-branch Transformer with gated cross-attention (DTGCA), designed to handle and integrate features from different domains for precise RUL prediction. Specifically, one branch processes one-dimensional time-series features from the time and frequency domains, while the other branch uses a residual convolutional GRU (res-ConvGRU) to handle two-dimensional time-frequency image features. The proposed gated cross-attention (GCA) mechanism enables adaptive information exchange between the branches, effectively fusing their information to provide a clearer representation of bearing degradation states. The proposed method is validated on two real run-to-failure datasets. Comprehensive ablation experiments confirm the method’s underlying rationality, while detailed comparative experiments with other approaches clearly demonstrate its superiority.
3 February 2022 in Ionics
25 June 2024 in IEEE Sensors Journal
The emerging graph convolutional network (GCN) provides a way to extract additional relationship information and has demonstrated superior performance in fault diagnosis. However, the GCN-based methods have some limitations when applied to cross-domain fault diagnostics. First, the graph topology constructed according to certain criteria remains fixed while it may contain domain-specific information, which could hamper the extraction of domain-invariant features. Second, the non-Euclidean structure of the graph hinders its augmentation through mixing and interpolation, bringing its application challenges in scenarios with small samples. To address these limitations, this paper proposes an Adaptive Topology-Aware Siamese Network (ATA-SN) for cross-domain fault diagnosis with small samples. Specifically, the multi-channel data is constructed as a complete graph as input. The adaptive topology-aware module updates and augments the graph simultaneously through biased uncertainty edge perturbation in a non-parametric manner. In combination with a contrastive loss, augmented data with clear class boundaries can be obtained. A statistical metrics-based style loss is used for domain-invariant feature learning. A pair of updated graphs is then fed into the siamese framework, which utilizes the semantic consistency of the paired graphs to recalibrate the contribution of the augmented data to the gradient descent, thus ensuring the effectiveness of the graph augmentation and domain adaptation. Comparative studies and cross-domain fault diagnosis experiments indicate that the superiority and effectiveness of the proposed method even in scenarios with extremely limited training data. The code of ATA-SN is released at https://github.com/CQU-ZixuChen/ATA-SN.