Single-domain generalization (SDG) in fault diagnosis aims to train a model that generalizes to unseen conditions using data from a single operating condition. Despite its promise, SDG faces significant challenges. Recent methods focus on generating new domain features to expand data distribution and extract domain-invariant features, but they encounter three issues: 1) separating domain expansion from domain-invariant feature learning limits their synergy and generalization; 2) domain expansion often uses random or inconsistent enhancement directions, introducing noise or generating unreliable features; and 3) domain-invariant feature extraction methods emphasize global or class-level alignment, neglecting the optimization of their relationships. Therefore, an SDG fault diagnosis framework based on the adversarial interaction between discrepancy and consistency is proposed, where adversarial training is repeatedly applied to dynamically balance domain discrepancy and consistency to avoid single-direction bias. Specifically, a learnable statistical feature perturbation module for domain discrepancy perception is proposed, which purposefully guides the model to learn statistical perturbations that effectively characterize domain discrepancies, thereby increasing the reliability of domain extension. In addition, covariance alignment and semantic alignment strategies are proposed to maintain global and category consistency between original and perturbed features. Covariance alignment is employed to mitigate stylistic variations introduced by perturbed features while preserving mission-critical diagnostic information, and explicit semantic constraints are utilized to further promote consistency in the model’s prediction results. Finally, extensive experiments are carried out on five datasets to verify the effectiveness of the proposed method.
Federated learning (FL) enables the collaborative training of intelligent diagnostic models across multiple clients while preserving data privacy. However, the complex and dynamic working conditions often cause federated diagnostic models to suffer from performance degradation when applied to unseen clients, exposing deficiencies in model generalization. Domain generalization fault diagnosis methods aim to enhance model adaptability under unknown working conditions, but most existing algorithms rely on shared data, making them unsuitable for direct application in federated learning environments. To address this challenge, we propose a federated domain generalization fault diagnosis method based on Cross-Client Style Integration and Dual Alignment representation (CCSIDA). This method significantly enhances model generalization in FL settings while ensuring data privacy. Specifically, CCSIDA integrates three key strategies: cross-client style data enhancement, domain-sensitive feature suppression, and predictive alignment. For cross-client style integration, we design a stochastic mixed-style sample generation module to overcome the limitations of conventional domain generalization data enhancement methods in FL. This module generates diverse samples by constructing a shared style repository, enabling inter-client style information transfer. For cross-client dual alignment representation, we introduce a domain-sensitive feature suppression strategy to reduce domain discrepancies between original and enhanced samples. Additionally, Jensen-Shannon divergence is employed to ensure prediction consistency between these samples. Finally, extensive experiments on five benchmark datasets validate the effectiveness of the proposed method.
Insulated Gate Bipolar Transistor (IGBT) has multiple degradation mechanisms, it is challenge to accurately integrating multiple signals to capture the device’s degradation patterns and health state. Therefore, comprehensively characterizing the health state of IGBT and predicting its remaining useful life using multiple signals poses a significant challenge. To address this challenge, a remaining useful life prediction method for IGBT based on deep fusion of nonlinear features from multi-source data is proposed. Firstly, the time-domain multi-features of IGBT degradation data are constructed and key features are selectively selected, then dimensionality reduction is performed and these features are fused into health indicators to characterize the health level. Secondly, the health of IGBT is effectively evaluated by unsupervised clustering without data labeling. Thirdly, end-condition monitoring is refined to enable the identification of near-failure state. Finally, deep learning is utilized to provide accurate and reliable prediction of the remaining useful life of IGBT devices (R2 are all greater than 0.98, the MAE are all less than 2.3, and the RMSE are all less than 5.5.). The results demonstrate that the method effectively integrates multi-source information, characterizes the health state of the device, and can more accurately and reliably predict the remaining useful life of IGBT. The proposed method can enhance the scientific basis for the health management of new energy systems such as wind power and photovoltaic systems.
18 January 2024 in Measurement Science and Technology
The monitoring data (current and rotational speed) of the subway sliding plug door transmission system changed positively and negatively with the acceleration and weight of the door. How to perceive the changes is a challenging problem in the health state assessment of transmission system. To address this problem, an enhanced perception health state assessment method was proposed for the transmission systems. In the method, firstly, the equivalent resistance force is calculated by monitoring the current and rotational speed data according to mechanical dynamic knowledge. Secondly, the sensitive features of normal and abnormal states are screened out from the enhanced dataset constructed by current, rotational speed data, and equivalent resistance force data. Finally, the health state of the transmission system is assessed using an integrated learning algorithm. The effectiveness of the method is verified by benchmark experimental data, and the results indicate that the method has a higher accuracy with four classifiers and a broader suitable range with varying door acceleration and weight.
14 March 2019 in IEEE Transactions on Components, Packaging and Manufacturing Technology