Domain adaptation (DA) methods are widely used in rolling bearing remaining useful life (RUL) prediction to align source and target domains. However, DA relies on previously collected target datasets, which are not always available in practical applications. To address this problem, an uncertainty-weighted domain generalization (UWDG) method is proposed for RUL prediction of rolling bearings under unseen conditions. Firstly, a knowledge distillation framework is established to capture internally-invariant features, which helps keep the features from varying with the existence of other domains. Secondly, correlation alignment is applied to acquire mutually-invariant feature distributions across domains, while regularization is employed to increase the disparity between the internally-invariant and mutually-invariant features. Then, multiple domain-generalization-task losses are weighted by homoscedastic uncertainty, ensuring consistency among separate task outputs. Finally, the effectiveness and superiority of UWDG have been validated through comparative experiments using the experimental dataset.
25 October 2023 in IEEE Transactions on Instrumentation and Measurement
In recent years, deep learning-based fault diagnosis methods have witnessed significant advancements and successful applications in engineering practice. However, the increasing complexity of network structures demands higher computational resources in terms of floating point operations per second (FLOPs) and parameters, which poses challenges when deploying diagnostic models in edge computing scenarios with limited run resources. To address this issue, this study proposes a novel lightweight network, namely cheap ghost network (CGhostNet), incorporating fine-grained feature knowledge distillation (FFKD). FFKD-CGhostNet leverages CGhostNet, a lightweight architecture, and transfers diagnostic knowledge from ResNet, a complex yet high-performing network, through FFKD. Extensive experiments are conducted on two test benches to demonstrate that FFKD-CGhostNet achieves comparable diagnostic performance to ResNet while significantly reducing parameter count by nearly 88 times and computational requirements by almost 14 times. These findings highlight the effectiveness of FFKD-CGhostNet in achieving superior diagnostic performance in resource-constrained edge computing scenarios.
19 September 2023 in IEEE/ASME Transactions on Mechatronics
The success of fault diagnosis based on deep learning benefits from a large amount of labeled fault samples. However, the scarcity of labeled fault samples in fault diagnosis of wind turbine gearbox (WTG) makes it difficult to train a satisfactory diagnostic model. To address this issue, this article proposed a semisupervised weighted centroid prototype network (SSWCPN) for WTG fault diagnosis. Specifically, SSWCPN is a few-shot semisupervised learning framework, which alleviates the matter of overfitting caused by the lack of supervision information. First, to capture abundant semisupervised information to guide network training, a sample selection model based on the evolution trend of posterior probability is proposed, which could efficiently cherry-pick out the unlabeled samples of high confidence to refine prototypes. Second, a new prototype updating strategy based on a weighted centroid prototype is designed, which controls the prototype drifting issue caused by incorrect pseudolabels and the introduction of new data distribution. Finally, experiments performed on test-bench data and successful application on WTG data show that the proposed SSWCPN-based WTG fault diagnosis achieves the best fault diagnosis performance among the comparison methods.
Semi-supervised learning (SSL) is effective to address the scarcity of label information in the fault diagnosis of rotating machinery. However, existing SSL methods generally assume that the labeled and unlabeled fault samples share a consistent label space, which is difficult to guarantee in practical applications and limits the applicability of SSL. To address this issue, this study presents a method denoted as open-set semi-supervised learning (O3SL) for intelligent fault diagnosis. First, an orthogonal projection network (OPN) is proposed to classify fault samples based on feature projection. Moreover, an open-set recognition module based on feature projection residual is further presented to discover potential unknown fault types in unlabeled fault samples. Finally, a pseudo label correction updating module is developed to further dynamically generate corrected pseudo labels, and thus to improve the performance of OPN. Extensive experiments on two gearbox fault datasets have validated the effectiveness of the proposed OPN-based O3SL method.
Extracting the most sensitive and effective characteristics from original high dimensional feature space is significant for accurately identifying the condition of machinery. This paper presents a locality sensitive discriminant analysis (LSDA) based feature dimensionality reduction approach. Different from locally preserving projection (LPP) that only preserves the local structure of data manifold, LSDA can also employ the discriminant information at each local area when conducting the projection. In the proposed fault diagnosis approach, various multi-domain features are calculated aimed at exploring the machinery condition information roundly from different perspectives. LSDA is applied to discover the representative information contained in the high dimensional feature set. Then a lower informative feature space can be constructed with LSDA. Finally, the effective KNN classifier is used to recognize the condition of machinery with the most representative eigenvectors. A bearing fault experiment is conducted to verify the proposed method. The comparison study with PCA, LPP and KLPP indicates that LSDA has advantages in dimensionality reduction and the proposed fault diagnosis approach is effective.
9 June 2023 in IEEE Transactions on Instrumentation and Measurement
The vibration signal of the wind turbine gearbox (WTG) contains weak fault information due to its complex structure and noise interference, which hampers the performance of WTG fault diagnosis. To address this challenge, a novel deep learning approach is proposed, called adaptive gated attention network with weighted metric enhancement (AGANet). The AGANet comprises an adaptive gated attention mechanism that enhances the capability of fault feature extraction by selecting and weighing relevant feature channels from multiple perspectives. Additionally, an adaptive weighted metric framework is proposed, which can further strengthen the identifiability of fault features. By integrating these two innovations, the proposed AGANet can effectively extract distinguishable fault features from vibration signals and achieves outstanding performance in WTG fault diagnosis. The proposed AGANet is validated through extensive experiments using test-bench data as well as WTG data, and the results demonstrate the superior performance of AGANet-based WTG fault diagnosis.