As train loads and travel speeds have increased over time, railway axle bearings have become critical elements which require more efficient non-destructive inspection and fault diagnostics methods. This paper presents a novel and adaptive procedure based on ensemble empirical mode decomposition (EEMD) and Hilbert marginal spectrum for multi-fault diagnostics of axle bearings. EEMD overcomes the limitations that often hypothesize about data and computational efforts that restrict the application of signal processing techniques. The outputs of this adaptive approach are the intrinsic mode functions that are treated with the Hilbert transform in order to obtain the Hilbert instantaneous frequency spectrum and marginal spectrum. Anyhow, not all the IMFs obtained by the decomposition should be considered into Hilbert marginal spectrum. The IMFs’ confidence index arithmetic proposed in this paper is fully autonomous, overcoming the major limit of selection by user with experience, and allows the development of on-line tools. The effectiveness of the improvement is proven by the successful diagnosis of an axle bearing with a single fault or multiple composite faults, e.g., outer ring fault, cage fault and pin roller fault.

High-resistance connection (HRC) is one of the common stator faults in motors. The HRC fault detection method for brushless DC motors (BLDCMs) has not been reported yet. This study proposes a two-step strategy for the detection, localization, and evaluation of HRC fault in a BLDCM. In the first step, the zero sequence voltage component (ZSVC) of the motor is acquired and analyzed, and the HRC fault can be detected from the amplitude of the fundamental frequency component of the ZSVC spectrum. In the second step, once the HRC fault has been detected, three indicators are calculated from the ZSVC and three-phase currents to localize the fault phase. The resistance value of the HRC is then calculated. Theoretical, simulated, and experimental analyses have been performed, and the results of them are consistent with one another, thereby validating the effectiveness of the proposed strategy. Both the HRC fault detection algorithm and the motor control algorithm are implemented on a single microcontroller unit in real time, which makes the proposed strategy efficient and flexible in industrial applications.

Broad Learning System (BLS) has shown remarkable results in various machine learning tasks, but the fragility of BLS based on the minimum mean square error criterion under the label noise environment in Internet of Things(IoT) limits its application. To solve these problems, a dual-model collaborative label adaptive correction BLS based on the collaboration of KRPBLS and SKBLS, namely SC-KSBLS is proposed to significantly improve the robustness of BLS under label noise environment. Firstly, a new broad learning system based on the Kernel risk-sensitive mean p-power(KRP) criterion, namely KRPBLS is developed to overcome the sensitivity by optimizing the output weight matrix and replacing the minimum mean square error criterion. Then, the label feature space is constructed, and the popular regularization strategy is designed to construct the SKBLS model with the collaborative training of KRP criterion and regularization to improve the learning ability of BLS. Finally, the interaction mechanism between KRPBLS and SKBLS is designed to effectively identify and correct label noise. The robustness theory of the SC-KSBLS has been analyzed and proven. The experimental results on nine datasets in IoT show that the SC-KSBLS significantly improves the robustness and performance, which provide an innovative solution under label noise environment.

In recent years, substantial progress has been made in cross-working condition fault diagnosis through transfer learning. However, a significant trade-off between diagnostic accuracy and generalization capability remains unresolved, especially in practical industrial scenarios where data from unseen domains is unavailable. To address this challenge, this paper proposes a novel domain generalization (DG) method that effectively integrates both discriminative and generalizable features for rolling bearing fault diagnosis under varying working conditions. Firstly, Fast Fourier Transform (FFT) is employed to extract phase-frequency representations, which are fused with time-domain signals to form a comprehensive input to the diagnostic network. Then a dual-branch architecture based on Instance-Batch Normalization (IBN) and MixStyle is constructed, the IBN captures discriminative features to enhance classification accuracy, and the MixStyle learns generalizable features to improve cross-domain adaptability. Furthermore, a Dynamic Weighting Invariant Risk Minimization (DWA-IRM) strategy is designed to assign higher importance to source domains with poor performance, thereby enforcing robust invariance constraints. The extensive experiments show that the proposed method achieves an average accuracy improvement of approximately 5.3% to 15.47% over state-of-the-art DG approaches, which demonstrate its effectiveness. This work offers a practical and scalable solution for intelligent fault diagnosis in highly variable industrial environments.