基于Bagging-MCNN模型的不均衡样本轴承故障诊断方法

Unbalanced sample bearing fault diagnosis method based on Bagging-MCNN model

针对轴承不均衡样本情景下故障诊断存在的精度与泛用性不高问题,借鉴集成学习获取强监督模型的方法,结合对不均衡样本进行采样处理的类别重组法,提出一种基于Bagging思路的多通道卷积神经网络(Bagging-MCNN)故障诊断模型。首先将原始数据进行标准化处理并划分为训练集与测试集,对训练集进行放回采样构造多个训练子集,同时对测试集进行乱序操作;然后将构造完成的新集合放入多通道卷积神经网络模型进行训练,获得各卷积网络子模型的判别矩阵,融合所有判别矩阵获得最终的诊断结果。在公开轴承数据集上进行试验验证,结合Bagging思路的多通道卷积神经网络故障诊断方法在均衡以及不均衡情景下的诊断精度相较普通卷积神经网络模型,分别提高了1.1%与10.8%,同时提高了模型的收敛速度以及诊断稳定性。

Aiming at the problem of low accuracy and generality of fault diagnosis under unbalanced sample situation of bearing,a fault diagnosis model of the Multi-channel Convolutional Neural Network based on Bagging idea(Bagging-MCNN)was proposed by using the method of ensemble learning to obtain strong supervision model and combining with the class reorganization method of sampling unbalanced samples.Firstly,the original data was standardized and divided into training set and test set.The training set was put back for sampling to construct multiple training subsets,and the test set was disordered;the new set was put into the multi-channel convolution neural network model for training,and the discriminant matrix of each convolution neural network sub model was obtained.The final diagnosis result was obtained by fusing all discriminant matrices.Experiments were carried out on open bearing data sets to verify that the multi-channel convolution neural network fault diagnosis method combined with Bagging idea can improve the diagnosis accuracy of 1.1%and 10.8%respectively compared with the ordinary convolution neural network model in the balanced and unbalanced scenarios,and can improve the convergence speed and diagnosis stability of the model.