The bearing state signal collected by the vibration sensor contains a large amount of environmental noise in actual processes, which leads to a reduction in the accuracy of the convolutional network in identifying bearing faults. To solve this problem, a one-dimensional convolutional neural network with a multiscale kernel (MSK-1DCNN) is proposed for the classification information enhancement of the input. A two-layer multiscale convolution structure (MSK) is used at the front of the network. MSK has five convolutional kernels with different sizes, and those kernels are used to extract features with varying resolutions in the original signal. In the multiscale convolution structure, the ELU activation function is used instead of the ReLU function to improve the antinoise ability of MSK-1DCNN, also by adding pepper noise to the training set data to destroy the input data and forcing the network to learn more representative features to improve the robustness of the network. Experimental results illustrate that the improved methods proposed in this paper effectively enhance the diagnostic performance of MSK-1DCNN under intense noise, and the diagnostic accuracy is higher than that of other comparison algorithms.
Traditional graph neural network-based gearbox fault diagnosis methods often exhibit limited adaptability to variations in signal characteristics caused by measurement disturbances and changing operating conditions. This limitation primarily stems from their reliance on predefined static graph structures and fixed adjacency relationships, which constrain the effective modeling of dynamic inter-signal dependencies and may impair diagnostic performance under varying conditions. To address this issue, this paper proposes a multi-channel adaptive dynamic graph neural network (MCADG-Net) for robust fault diagnosis based on dynamic graph learning. First, a dynamic graph learning module is introduced to adaptively construct a global adjacency matrix through a multi-head attention mechanism with regularization, enabling flexible graph topology optimization and improved robustness to measurement perturbations. Second, a multi-scale adaptive Gaussian perception layer is designed to adjust dimension-wise feature weights via parameterized Gaussian functions, enhancing sensitivity to local feature variations. Finally, a multi-level aggregation network is developed to integrate the complementary advantages of GCN-based local smoothing and graph attention network-based global attention weighting, achieving multi-scale feature fusion and improved generalization across fluctuating working environments. Experimental results on two benchmark gearbox datasets demonstrate that the proposed MCADG-Net consistently outperforms representative dynamic GNN models and exhibits improved robustness and predictive reliability under diverse operational states. These findings indicate that MCADG-Net provides an effective and adaptive solution for intelligent measurement-based mechanical fault diagnosis.