基于深度迁移学习的矿井通风机轴承故障诊断

Bearing Fault Diagnosis for Mine Ventilator Based on Deep Transfer Learning

针对实际应用中矿井通风机轴承负样本少导致故障诊断率低的问题,提出一种基于深度迁移学习的矿井通风机轴承故障诊断方法。组合卷积神经网络(CNN)与双向门控循环单元(BiGRU),并采用随机森林(RF)分类器替换CNN的Softmax层,构建CNN-BiGRU-RF诊断模型,提取轴承更深层次故障特征以便于故障识别;利用源域数据对模型训练,确定模型结构参数;最后,引入迁移学习将模型迁移至目标域,使用目标域有标签数据微调模型参数,构建目标域诊断模型进行故障分类。实验结果表明:在矿井通风机轴承负样本稀少情况下,所提方法的故障识别平均准确率在94%以上,与其他方法相比,具有更好的诊断精度和泛化能力。

Aiming at the problem that the fault diagnosis rate is low due to the lack of negative samples of mine ventilator bearings in practical applications,a fault diagnosis method for mine ventilator bearings based on deep transfer learning was proposed.Combi⁃ning convolutional neural network(CNN)with bidirectional gated recurrent unit(BiGRU)and replacing the Softmax layer of CNN with random forest(RF)classifier,a CNN-BiGRU-RF diagnostic model was built to extract deeper bearing fault features for fault identifi⁃cation.The source domain data were used to train the model to determine the model structure parameters.Finally,transfer learning was introduced to migrate the model to the target domain,the labeled data in the target domain were used to fine-tune model parame⁃ters,and a target domain diagnostic model was built for fault classification.The experimental results show that the average fault identi⁃fication accuracy of the proposed method is more than 94%when the negative samples of the mine ventilator bearing are sparse.Com⁃pared with other methods,the proposed method has better diagnostic accuracy and generalization ability.