Aiming at the problems of the low detection accuracy and difficult identification of the early weak fault signals of rolling bearings, this paper proposes a method for detecting the early weak fault signals of rolling bearings based on a double-coupled Duffing system and VMD. The influence rule of system initial value on the response characteristics of a double-coupled Duffing system is studied, and the basis for its determination is given. The frequency of the built-in power of the system is normalized, and a variance evaluation standard for the output value of the double-coupled Duffing system for weak fault signals detection is established. In order to solve the interference problem of fault monitoring signals, VMD is proposed to pre-process the fault monitoring signals. The weak fault signal detection method proposed in this paper is tested and verified by simulation signals and rolling bearing fault signals. The results show that the method proposed in this paper can detect the weak fault signal with the lowest signal-to-noise ratio reduced by 2.96 dB compared with the traditional Duffing detection system, and it can accurately detect the early weak fault signal of rolling bearings.
Most fault diagnosis models use a single input and have weak generalization performance. In order to obtain more fault information, a fault diagnosis method based on a Multi-channel Residual Attention Network with Efficient Channel Attention (ECA-MRANet) is proposed in this paper. In this method, the original time domain signal is first processed by a multi-domain transform, the result of which is input to the MRANet for feature extraction. Finally, the extracted features are fused by ECA to realize fault identification. The experimental results show that the proposed method can enhance the ability of the network to discriminate key features, and shows good generalization performance under different working conditions and with small-sample transfer between data sets.
17 September 2025 in Measurement Science and Technology
Addressing the scarcity of data reflecting realistic variable-speed operating conditions in current research on aero-engine inter-shaft bearing fault diagnosis, and overcoming limitations in existing methods concerning time-frequency resolution, global cross-channel interaction capability, and robustness under variable speeds, we proposed a High-Channel Triple Attention and LSK Module Fusion Network (HC-TSKNet) characterized by high precision and strong robustness. The method first employs Synchrosqueezing Wavelet Wavelet Packet Transform (SSWPT) to transform the raw vibration signals into high-resolution time-frequency spectrograms. This effectively mitigates the spectral smearing issue prevalent in conventional time-frequency analysis such as Continuous Wavelet Transform (CWT), significantly enhancing the clarity of discriminative fault signatures. The core innovation of HC-TSKNet lies in its synergistic integration of three key techniques: through an increased channel count, the model size is compressed and the per-sample inference time is reduced, while preserving a satisfactory level of diagnostic accuracy; utilizing a Triplet Attention mechanism to capture global cross-channel and spatial interaction dependencies, overcoming the limitations in cross-dimensional interaction inherent in conventional attention mechanisms like CBAM and SENet; and introducing a Large Selective Kernel (LSK) module to adaptively weight local features extracted by deep convolutional kernels of different scales spatially, enabling dynamic focus on localized fault patterns under variable-speed conditions. Validation experiments conducted on an inter-shaft bearing fault dataset from aero-engine dual-rotor simulator test rig demonstrate that HC-TSKNet achieves exceptional diagnostic accuracy of 99.75% under variable-speed conditions, significantly outperforming mainstream methods including Swin-Transformer and recent state-of-the-art models in related domains. It is particularly noteworthy that the model successfully overcomes significant variations in signal characteristics induced by differences in sensor mounting positions, demonstrating robust adaptability to diverse sensor configurations. Ablation studies, hyperparameter experiment, and robustness validation further substantiate the effectiveness of its key modules, the superiority of its structural design, and its substantial potential for engineering applications.
Cross-speed bearing fault diagnosis faces a physical obstacle: bearing characteristic frequencies are proportional to shaft speed, so identical faults shift in the spectrum when source and target speeds differ, and source-trained feature mappings degrade on the target. Under large speed gaps, this drives the inner-race recall of existing domain-adaptation methods toward zero. We propose a Physics-Prior Adaptive Network (PPAN) whose core is Physical Frequency Normalization (PFN), which transforms the vibration spectrum from absolute-Hz coordinates into a BPFI-normalized coordinate system, aligning same-class fault features across speeds, together with Physics-Guided Spectral Attention (PGSA) and conditional domain adaptation performed in the normalized frequency domain. All results are obtained under a strict leakage-free protocol in which target labels are withheld from model selection. On the Paderborn dataset (1500–900 rpm, 40% speed gap), PPAN attains 67.1% target-domain accuracy and raises inner-race recall from near zero (0.0% for source-only transfer) to 88.2%, outperforming eight representative baselines (a source-only lower bound and seven statistical domain-adaptation methods). A controlled ablation isolates the contribution of PFN and shows that extending the model with a time-domain branch and dual-branch fusion degrades inner-race recall, motivating a single-branch, physics-first design.