陈闯, 李先锋, 史建涛, 岳冬冬. 基于柔性残差神经网络的滚动轴承智能故障诊断方法[J]. 工程科学学报. DOI: 10.13374/j.issn2095-9389.2024.06.24.006
引用本文: 陈闯, 李先锋, 史建涛, 岳冬冬. 基于柔性残差神经网络的滚动轴承智能故障诊断方法[J]. 工程科学学报. DOI: 10.13374/j.issn2095-9389.2024.06.24.006
Intelligent fault diagnosis method for rolling bearings based on flexible residual neural network[J]. Chinese Journal of Engineering. DOI: 10.13374/j.issn2095-9389.2024.06.24.006
Citation: Intelligent fault diagnosis method for rolling bearings based on flexible residual neural network[J]. Chinese Journal of Engineering. DOI: 10.13374/j.issn2095-9389.2024.06.24.006

基于柔性残差神经网络的滚动轴承智能故障诊断方法

Intelligent fault diagnosis method for rolling bearings based on flexible residual neural network

  • 摘要: 滚动轴承作为旋转机械的重要组成部分,其正常运行直接影响机器的使用寿命和运行状态。为了提高滚动轴承故障诊断的准确性,本文提出一种基于动态减法平均优化器(DSABO)和平行注意力模块(PAM)的柔性残差神经网络(FResNet),用于滚动轴承故障诊断。具体而言,首先设计一种基于卷积神经网络的柔性残差模块来构建FResNet。该模块允许在DSABO迭代时更改卷积层数、卷积核数和跳跃连接数,从而增强网络故障特征提取能力并减少网络退化。其次,设计具有卷积层的PAM来融合通道注意力和空间注意力输出权重,通过与滚动轴承运行数据结合,实现数据特征增强。于是,DSABO、PAM和FResNet的集成形成了一个有效的滚动轴承故障诊断模型,命名为DSABO-PAM-FResNet。最后,利用美国凯斯西储大学滚动轴承故障数据集验证所提DSABO-PAM-FResNet模型的可行性和有效性。实验结果显示,在信噪比为-6 dB环境下所提模型对滚动轴承故障诊断的准确率为97.18 %,证明所提模型具有较好的抗噪能力;在1 hp、2 hp和3 hp的不同负载条件下,所提模型对滚动轴承故障诊断的平均准确率为98.2 %,证明所提模型具有良好的变工况诊断适应能力。与其他智能故障诊断方法的对比结果表明,所提DSABO-PAM-FResNet模型的诊断精度更高,为滚动轴承故障诊断提供了一种新的有效智能方法。

     

    Abstract: As an important component of rotating machinery, the normal operation of rolling bearings directly affects the service life and operating status of the machine. Aimed at improving the accuracy of fault diagnosis for rolling bearings, a flexible residual neural network (FResNet) based on dynamic subtraction average based optimizer (DSABO) and parallel attention module (PAM) is proposed for rolling bearing fault diagnosis to address this problem. Specifically, a flexible residual module based on convolutional neural networks is first designed to construct FResNet. This module allows for changing the numbers of convolutional layers, convolutional kernels, and skip connections during DSABO iterations, thereby enhancing fault feature extraction capabilities and reducing network degradation. Secondly, a PAM with convolutional layers is designed to integrate the output weights of channel attention and spatial attention, and achieve data feature enhancement by combining it with rolling bearing operation data. Thus, the integration of DSABO, PAM, and FResNet forms an effective rolling bearing fault diagnosis model, named DSABO-PAM-FResNet. Finally, the feasibility and effectiveness of the proposed DSABO-PAM-FResNet model are validated using the rolling bearing fault dataset from Case Western Reserve University in the United States. The experimental results show that the accuracy of the proposed model in diagnosing rolling bearing faults in an environment with a signal-to-noise ratio of -6 dB is 97.18 %, proving that the proposed model has good noise resistance ability. Under different load conditions of 1 hp, 2 hp, and 3 hp, the average accuracy of the proposed model in diagnosing rolling bearing faults is 98.2 %, proving that the model has good adaptability to variable working condition diagnosis. The comparison results with other intelligent fault diagnosis methods show that the proposed DSABO-PAM-FResNet model has higher diagnostic accuracy, providing a new and effective intelligent method for rolling bearing fault diagnosis.

     

/

返回文章
返回