为解决滚动轴承在变载荷、大噪声背景下故障诊断困难及所建立智能模型泛化能力不足的问题，基于变分模态分解算法（Variational Mode Decomposition，VMD）及卷积神经网络（Convolutional Neural Network，CNN）技术提出VMD-CNN故障诊断模型。以标准轴承实验数据为研究对象，对11种故障类型和损伤程度不同的轴承状态进行识别。结果表明：不同模态数所对应的VMD-CNN模型诊断性能不同，模态过多导致过分解现象，弱化模型性能；与现有算法相比，基于最优模态数所构建的VMD-CNN模型在变载荷与噪声环境下性能更佳；可视化提取特征“点簇”说明各模态间存在互补机制，此机制可一定程度上提升模型鲁棒性与泛化能力。

Convolutional neural network can be used with variational mode decomposition (VMDCNN) to develop an endtoend fault diagnosis system of rolling bearings for solving the problems that the generalization ability of intelligent fault diagnosis model is insufficient,but it is difficult to realize accurately fault diagnosis under variable load and noise environments.Based on experimental data,VMDCNN is used to identify 11 bearing states with different fault types and damage degrees.The results show that the mode number can affect the performance of the VMDCNN model,and the performance of the model will be weakened by the overdecomposition.Compared with the existing algorithms,the VMDCNN model based on optimized mode number has better performance under variable load and noise environments.The “point cluster” indicates that there is a complementary mechanism between each mode,which can improve the robustness and generalization ability of the model to some extent.

TK47

国家自然科学基金（51976131，51676131）；上海市“科技创新心动计划”地方院校能力建设项目（19060502200）