【摘要】 针对传统智能故障诊断方法在强噪声和复杂工况环境下难以提取可靠特征、诊断准确率低及特征可解释性差的问题，该文提出一种面向航空发电机、风力发电机等大型旋转设备内部轴承的自适应谱图小波卷积神经网络故障诊断方法。该方法考虑多传感器间的交互作用，振动信号转换成图结构数据，将融合猎豹优化算法的自适应谱图小波与重加权小波系数策略嵌入图卷积层，构建自适应谱图小波卷积神经网络。自适应谱图小波卷积神经网络根据振动信号特性动态调整谱图小波的设计参数和分解层数，实现多尺度特征的高效提取，并将不同尺度特征赋予不同权重，增强信号去噪和弱特征提取能力。该方法实现在高噪声环境下端到端的轴承原位故障诊断，同时为图卷积操作提供高效且可解释的框架。实验结果表明，所提方法在高噪声和复杂工况下均表现出良好的诊断性能及鲁棒性。更多还原

【Abstract】 Traditional intelligent fault diagnosis methods face challenges in extracting reliable features, achieving high diagnostic accuracy, and providing feature interpretability under strong noise and complex operating conditions. To address these issues, this paper proposes an adaptive spectral graph wavelet convolutional neural network(ASGWCN) fault diagnosis method for the bearings in large rotating equipment,such as aircraft engines and wind turbines. The method considers the interactions between multiple sensors and converts vibration signals into graph-structured data. It integrates the adaptive spectral graph wavelet and re-weighted wavelet coefficient strategies,optimized by the Cheetah optimization algorithm,into the graph convolution layers,thereby constructing the ASGWCN. This approach dynamically adjusts the design parameters and decomposition levels of the spectral graph wavelet based on the characteristics of the vibration signal,enabling efficient multi-scale feature extraction. The different scale features are assigned different weights,enhancing the signal denoising and weak feature extraction capabilities. The method achieves end-to-end bearing fault diagnosis in situ under high noise environments,while also providing improved interpretability for the graph convolution process. The experimental results show that the proposed method exhibits excellent diagnostic performance and robustness under both high noise and complex working conditions.更多还原