基于DFCNN的滚动轴承定量诊断方法研究

以滚动轴承作为研究对象,设计了深度可分离模块、残差骨干网络、金字塔池化结构和路径聚合结构等特征融合单元,建立了深度特征融合的卷积神经网络（Deep Feature Convolutional Neural Network,DFCNN）,分析了随机梯度下降法对网络参数优化的有效性及数据集传递次数与模型精度的关系,开展了不同样本容量和不同噪声环境下的故障试验。结果表明：提出的DFCNN模型可以有效识别滚动轴承损伤部位以及损伤程度,诊断准确率大于99.5%;该模型对样本容量要求低、抗噪能力出色,当信噪比大于-4时诊断准确率大于98.86%。

A Rolling Bearing Quantitative Diagnosis Method based on DFCNN

Taking rolling bearing as the research object, this paper designed feature fusion units such as depth separable module, backbone network of residual network, pyramid pooling structure,path aggregation structure, and established a deep feature convolutional neural network (DFCNN).The effectiveness of the stochastic gradient descent method for network parameter optimization was analyzed, the relationship between number of transmission of dataset and model accuracy was discussed,and the fault tests under different sample sizes and different noise environments were carried out.The results show that the proposed DFCNN model can effectively identify the damage location and degree of rolling bearing, and the diagnosis accuracy is more than 99.5%; it has low requirements for sample size and excellent anti noise ability.When the signal to noise ratio is greater than -4, the diagnostic accuracy is greater than 98.86%.