【摘要】 在强噪声环境下，滚动轴承振动信号的故障特征会被噪声掩盖，从而导致卷积神经网络(convolutional neural network, CNN)的特征提取变得困难，诊断性能会下降。因此，提出一种非局部块宽核卷积神经网络(non-local block wide kernel CNN,NLBWCNN)算法用于滚动轴承的故障诊断。该算法采用了宽卷积核策略，更好地提取强噪声环境下一维振动信号的故障特征，同时采用了非局部块(non-local block, NLB)算法，加强了噪声数据的特征聚集能力，使得基于NLBWCNN的故障诊断模型能够直接对噪声数据进行有效地特征提取和故障分类。最后用帕德博恩大学真实损伤轴承数据集和凯斯西储大学轴承数据集进行了验证，结果显示，和传统的CNN等深度学习方法相比较，该方法对包含有噪声的轴承数据有着更高的故障诊断准确率。更多还原

【Abstract】 Convolutional Neural Networks(CNNs) are widely applied in rolling bearing fault diagnosis. However, in high-noise environments, fault-related features in the vibration signals can be masked by noise, making feature extraction challenging and degrading the diagnostic performance of conventional CNN models. To address this, a Non-Local Block Wide Kernel Convolutional Neural Network(NLBWCNN) is proposed, incorporating wide convolutional kernels and a Non-Local Block(NLB) mechanism. The wide convolutional kernel design enhances the network′s ability to extract meaningful features from one-dimensional vibration signals under strong noise conditions. Simultaneously, the NLB module improves feature aggregation by modelling long-range dependencies, enabling the model to more effectively identify and classify noise-corrupted data. The proposed NLBWCNN algorithm has been validated using real damaged bearing datasets from Paderborn University and Case Western Reserve University. Experimental results demonstrate that the method outperforms traditional CNN-based models in terms of fault diagnosis accuracy when applied to noisy bearing data.更多还原