域不对称因子加权的滚动轴承故障深度局部迁移诊断方法,首先利用深度残差网络提取源滚动轴承与目标滚动轴承监测数据中的深度迁移故障特征;其次利用深度迁移故障特征训练领域混淆网络,并计算域不对称因子;再次计算深度残差网络适配层故障特征的多核最大均值差异,并利用域不对称因子加权抑制源滚动轴承中无效故障特征的贡献度;最后利用加权的多核最大均值差异构建目标函数,训练深度残差网络;训练所得的由领域混淆网络和深度残差网络构成的局部迁移诊断模型能够有效克服域不对称因素对迁移诊断的不利影响,实现了运用源滚动轴承局部诊断知识识别目标滚动轴承的不平衡健康状态,显著提高了迁移诊断模型的诊断精度。
The method of deep local migration diagnosis of rolling bearing faults with domain asymmetry factor weighting, firstly, extracts the deep transfer fault features in the source rolling bearing and the target rolling bearing monitoring data using the deep residual network; secondly, trains the domain confusion network using the deep transfer fault features and calculates the domain asymmetry factor; again, calculates the multicore maximum mean difference of the fault features in the adaptation layer of the deep residual network and uses the domain asymmetry Finally, the weighted multicore maximum mean difference is used to construct the objective function and train the depth residual network; the resulting local transfer diagnosis model composed of domain confusion network and depth residual network can effectively overcome the negative influence of domain asymmetry factor on transfer diagnosis, and realize the local diagnosis knowledge of source rolling bearing to identify the target The local transfer diagnosis model is trained to overcome the negative influence of domain asymmetry on the transfer diagnosis, and to identify the unbalanced health state of the target rolling bearing using the local diagnosis knowledge of the source rolling bearing.