【摘要】 针对大型机械装备环境噪声复杂,深度学习网络层数过深导致的巨大计算开销以及故障诊断人工特征提取的复杂性,提出改进残差结构的轻量级SCARN模型。SCARN模型使用蓝图可分离卷积代替常规卷积层,减少大量参数,设计轻量级空间通道注意力结构,加强特征表达能力,改进深度残差收缩模块,提高模型复杂噪声场景的鲁棒性。通过增加不同幅值的高斯白噪声模拟轴承信号复杂环境场景。实验结果表明,该模型4种评价指标均优于对比算法,具有良好的抗噪性能。更多还原

【Abstract】 Aiming at the complex environmental noise of large-scale mechanical equipment,the huge computational overhead due to the deep layer of deep learning network and the complexity of fault diagnosis artificial feature extraction,lightweight SCARN model with improved residual structure was proposed.The conventional convolution layer was replaced by blueprint separable convolution to reduce a large number of parameters.A lightweight spatial channel attention structure was designed to enhance the ability of feature expression.The depth residual shrinkage module was improved to improve the robustness of the model in complex noise scenes.The complex environment scene of bearing signal was simulated by adding Gaussian white noise of different amplitude.Experimental results show that four evaluation indicators of the model are better than that of the comparison algorithm,and it has good anti-noise performance.更多还原