深度自编码与改进损失函数在极端不均衡故障诊断中的应用

在实际应用中,滚动轴承大多时候都是在正常状态下工作,因此收集到的故障数据较少,这就会产生数据不均衡的问题.这种数据不均衡问题极大地影响着模型的拟合和泛化能力,导致模型产生过拟合情况,而往往忽视对小类别样本的学习.尤其当故障样本数极少时,此问题更突出.针对这个问题,提出一种基于改进交叉熵损失函数的深度自编码器的诊断模型,首先提取振动数据的小波包能量,其次将小波包能量输入到深度自编码器中,最后通过SoftMax分类器得到诊断结果.改进的加权损失函数可以根据各类别样本的数量调整权重系数,样本数量越少,系数越大,使得模型在训练时更专注于数量较少的样本.通过在凯斯西储大学及西安交通大学的轴承数据集上的两个实验表明,加权损失函数可以提高极端不均衡数据的诊断精度.

Application of Deep Auto-encoder and Improved Loss Function in Extremely Unbalanced Fault Diagnosis

The distribution of fault categories is imbalanced in practice, that is, the fault data is often less than normal data. Traditional classifiers tend to the majority of classes in fault diagnosis neglecting the detail in the small fault samples. This problem is especially obvious when the data distribution is extremely imbalanced. A controlled weight coefficient is introduced into the loss function of a deep auto-encoder (DAE). First, the wavelet packet energy of vibration data is extracted, secondly, the wavelet packet energy is input into the DAE, and finally the SoftMax classifier diagnostic result. The weight coefficient is adjusted according to the number of samples for each category. The smaller the number of samples, the larger the coefficient, which makes the model focus more on small sample during training. Two experiments on the bearing datasets of Case Western Reserve University (CWRU) and Xi"an Jiaotong University show that the weighted loss function can improve the diagnostic accuracy of extremely imbalanced dataset.