基于卷积神经网络的轴承故障诊断研究

Study on bearing fault diagnosis based on convolution neural network

针对复杂工况下的采煤机摇臂轴承故障诊断,以经典AlexNet为基础,为适应一维时域信号,采用滑窗法,以滑窗长度150 ms,移动步长120 ms构建样本,建立一种由池化层和多级交替卷积层组成的轴承故障诊断模型1CNN,可完成原始输入信号特征的自适应提取,并通过全连接层分类识别轴承故障。为验证1CNN模型故障诊断率,利用机械综合故障试验台对MG650/1750-WD采煤机摇臂轴承6008-2Z内圈、外圈、滚子不同形态及大小的故障测试试验,得到振动信号的时域频谱。通过1CNN模型直接读取振动信号时域数据的故障识别试验,结果表明:1CNN模型算法识别准确率达到99.6%,是一种有效的采煤机截割系统轴承故障诊断技术。

In view of the complicated working conditions of the fault diagnosis of shearer’s rocker bearing,this paper adopted the sliding window method to construct the sample with the window length of 150 ms and the step length of 120 ms.Based on classic AlexNet,the bearing fault diagnosis model 1CNN was established to adapt to one-dimensional time-domain signals.Composed of a multilevel alternating convolutional layer and a pooling layer,this model could complete the adaptive extraction of the original input signal features and realize fault classification and recognition through the fully-connected layer.To verify the fault diagnosis rate of the model,experiments were carried out to test the fault with different forms and sizes of the inner ring,outer ring,and the roller bearing of the MG650/1750-WD shearer’s rocker bearing 6008-2Z to obtain the time-domain spectrum of vibration signals,which were then read by the model ICNN directly.The results indicate that with a high recognition accuracy of 99.6%,the proposed 1CNN is an effective bearing fault diagnosis technology for cutting system of shearer.