摘要
针对多种工况和故障共存引起的滚动轴承故障,采用传统的基于卷积神经网络的故障诊断模型进行诊断时,存在提取特征不丰富、容易丢失故障敏感信息、计算复杂和准确性低的问题,为此,提出了一种二维卷积神经网络(2D-CNN)与门控循环单元(GRU)相结合的滚动轴承多故障诊断模型(2D-CNN-GRU),并采用XJTU-SY和QPZZ-II两个公开轴承数据集,对其有效性进行了验证。首先,采用2D-CNN作为空间特征提取器,获取了信号的多种局部和全局特征,并将GRU层作为信号时序信息特征提取器;然后,对模型的特征提取过程进行了可视化处理;最后,将所选择的有效信号输入2D-CNN-GRU模型中,完成了数据分类,进而完成了轴承故障诊断。研究结果表明:利用XJTU-SY实验数据和QPZZ-II实验数据,验证了该方法在多种工作条件下对多种轴承故障共存的情况具有优秀的分类效果,准确率达到了95%以上。与传统轴承故障诊断方法相比,2D-CNN-GRU模型具有更高的准确性和一定的实际应用价值。
Aiming at the rolling bearing faults caused by the coexistence of various working conditions and faults,the traditional fault diagnosis model based on convolutional neural network has some problems,such as insufficient feature extraction,easy loss of fault sensitive information,complicated calculation and low accuracy.Therefore,a rolling bearing multi-fault diagnosis model(2D-CNN-GRU)combining 2D convolutional neural network(2D-CNN)and GRU was proposed,and XJTU was adopted.Firstly,2D-CNN was employed as a spatial feature extractor to obtain various local and global features of signals.This approach effectively captures the spatial characteristics of the input data,it was crucial for identifying different types of bearing faults.Meanwhile,the GRU layer was used as a temporal feature extractor to capture the time-series information of the signals,allowing the model to account for the sequential nature of vibration data.Then,the feature extraction process of the model was visualized to understand the model s internal mechanisms and to validate the effectiveness of the feature extraction methods.Finally,the selected effective signals were input into the 2D-CNN-GRU model to complete data classification and achieve bearing fault diagnosis.The research results indicate that the proposed method has excellent classification performance in the coexistence of multiple bearing faults under various working conditions,with an accuracy rate of over 95%,verified by the XJTU-SY experimental data and the QPZZ-II experimental data.Compared with traditional bearing fault diagnosis methods,the 2D-CNN-GRU model demonstrates higher accuracy and certain practical application value.The integration of spatial and temporal feature extraction provides a more comprehensive understanding of fault characteristics,leading to improved diagnostic performance.This study highlights the potential of combining deep learning techniques to address complex fault diagnosis tasks and offers a valuable reference for future research in this field.
作者
张雄
渠伟瀅
王文强
董乐聪
万书亭
ZHANG Xiong;QU Weiying;WANG Wenqiang;DONG Lecong;WAN Shuting(Hebei Key Laboratory of Electric Machinery Health Maintenance&Failure Prevention,Baoding 071003,China;Department of Mechanical Engineering,North China Electric Power University,Baoding 071003,China)
出处
《机电工程》
北大核心
2025年第10期1931-1939,共9页
Journal of Mechanical & Electrical Engineering
基金
国家自然科学基金资助项目(52105098)
中央高校基本科研业务费项目(2025MS137)
河北省自然科学基金资助项目(E2024502052,E2021502038)。
关键词
滚动轴承
多故障诊断
二维卷积神经网络
门控循环单元
特征提取
数据预处理
rolling bearing
multi-fault diagnosis
two-dimensional convolutional neural network(2D-CNN)
gate recurrent unit(GRU)
feature extraction
data preprocessing
