轴承是传动系统重要的支撑部件,也是整个系统的薄弱环节,卷积神经网络(convolutional neural network,CNN)强大的特征提取和学习能力为轴承运行状态模式识别提供了可能性.针对CNN处理多分类模式识别过程中,由超参数问题引起的准确率低、收敛速度慢等问题,提出了一种基于哈里斯鹰优化(Harris hawks optimization,HHO)算法优化的CNN分类模型.首先,对不同故障类型和故障程度轴承故障数据集进行划分,初始化CNN模型参数;然后,使用HHO算法对CNN模型的超参数空间进行优化,计算适应度值并获取全连接层的单元数量和迭代次数;最后,利用优化后的CNN模型对轴承数据集进行模式识别.通过不同故障类型和故障程度轴承实验数据验证,表明HHO-CNN模型可以使得全连接层的单元数量和迭代次数迅速收敛,及时准确调整CNN的网络参数,提升分类器的性能,提高了故障模式识别准确性,增强了模型的稳定性.

Bearing is an important supporting component of drive system, and it is also the weak link of the whole system. The powerful feature extraction and learning ability of convolutional neural network(CNN)provides the possibility for the pattern recognition of the idler bearing. Aiming at the problems of low accuracy and slow rate of convergence caused by the super parameter problem in using CNN to process multi classification pattern recognition, a CNN classification model optimized based on HHO algorithm was proposed. Firstly, the bearing fault dataset is partitioned with different fault types and degrees, and initializes the CNN model parameters; Then, the HHO algorithm is used to optimize the CNN model, calculate fitness values, and obtain the number of units and iteration times of the fully connected layer; Finally,- DOI:10.3969/j.issn.1000-1565.2023.06.002基于HHO-CNN的轴承故障诊断方法研究刘玉鑫¹,武文博²,张雄²,万书亭²(1.国家电投内蒙古能源有限公司 通辽霍林河坑口发电有限责任公司,内蒙古 通辽 028000;2.华北电力大学 机械工程系,河北 保定 071003)摘 要:轴承是传动系统重要的支撑部件,也是整个系统的薄弱环节,卷积神经网络(convolutional neural network,CNN)强大的特征提取和学习能力为轴承运行状态模式识别提供了可能性.针对CNN处理多分类模式识别过程中,由超参数问题引起的准确率低、收敛速度慢等问题,提出了一种基于哈里斯鹰优化(Harris hawks optimization,HHO)算法优化的CNN分类模型.首先,对不同故障类型和故障程度轴承故障数据集进行划分,初始化CNN模型参数;然后,使用HHO算法对CNN模型的超参数空间进行优化,计算适应度值并获取全连接层的单元数量和迭代次数;最后,利用优化后的CNN模型对轴承数据集进行模式识别.通过不同故障类型和故障程度轴承实验数据验证,表明HHO-CNN模型可以使得全连接层的单元数量和迭代次数迅速收敛,及时准确调整CNN的网络参数,提升分类器的性能,提高了故障模式识别准确性,增强了模型的稳定性.关键词:HHO;CNN;轴承;故障诊断中图分类号:TH133.3;TP181 文献标志码:A 文章编号:1000-1565(2023)06-0571-13Fault detection method of bearings based on HHO-CNNLIU Yuxin¹, WU Wenbo², ZHANG Xiong², WAN Shuting²(1.Tongliao Huolinhe Kengkou Power Generation Co. Ltd., State Power Investment Inner Mongolia Energy Co. Ltd.,Tongliao 028000, China;2.Department of Mechanical Engineering, North China Electric Power University, Baoding 071003, China)Abstract: Bearing is an important supporting component of drive system, and it is also the weak link of the whole system. The powerful feature extraction and learning ability of convolutional neural network(CNN)provides the possibility for the pattern recognition of the idler bearing. Aiming at the problems of low accuracy and slow rate of convergence caused by the super parameter problem in using CNN to process multi classification pattern recognition, a CNN classification model optimized based on HHO algorithm was proposed. Firstly, the bearing fault dataset is partitioned with different fault types and degrees, and initializes the CNN model parameters; Then, the HHO algorithm is used to optimize the CNN model, calculate fitness values, and obtain the number of units and iteration times of the fully connected layer; Finally,- 收稿日期:2023-06-13 基金项目:国家自然科学基金资助项目(52105098);河北省自然科学基金资助项目(E2021502038) 第一作者:刘玉鑫(1979—),男,内蒙古通辽人,国家电投内蒙古能源有限公司高级工程师,主要从事电力系统科技创新与管理研究.E-mail:zxlvbzy@163.com 通信作者:张雄(1990—),男,河北保定人,华北电力大学副教授,博士,主要从事旋转机械设备状态检测与故障诊断研究. E-mail:hdjxzx@ncepu.edu.cn第6期刘玉鑫等:基于HHO-CNN的轴承故障诊断方法研究the optimized CNN model is used for pattern recognition on the bearing dataset. Through experimental data of bearings, it is shown that the HHO-CNN model can quickly converge the number of units and iteration times in the fully connected layer, adjust the network parameters of the CNN in a timely and accurate manner, improve the accuracy of fault pattern recognition, and enhance the stability of the model.