基于KOA-CNN-BiLSTM-AM的风电齿轮箱故障诊断研究  

Fault Diagnosis Research of Wind Turbine Gearbox Based on KOA-CNN-BiLSTM-AM

作　　者：龙霞飞 何志成 周凌[1] 刘伟强 梁凯 LONG Xiafei;HE Zhicheng;ZHOU Ling;LIU Weiqiang;LIANG Kai(College of Electrical and Information Engineering,Hunan University of Technology,Zhuzhou Hunan 412007,China)

机构地区：[1]湖南工业大学电气与信息工程学院,湖南株洲412007

出　　处：《机床与液压》2025年第4期214-220,共7页Machine Tool & Hydraulics

基　　金：湖南省教育厅科学研究项目(22B0590,23B0537)。

摘　　要：为了保障大型风电机组的可靠运行并实现风电机组的早期故障识别,提出一种融合开普勒优化算法(KOA)、卷积神经网络(CNN)、双向长短时记忆网络(BiLSTM)和注意力机制(AM)的深度神经网络早期故障诊断混合模型。对齿轮箱原始振动数据进行预处理,利用CNN+BiLSTM方法建立输入参数和输出参数之间的逻辑关系,并融合AM+KOA方法,自动、有效地提取深层次故障特征信息,改善传统深度神经网络容易陷入局部最优、收敛速度慢、提取特征能力不足而导致诊断效果不佳等问题,实现风电齿轮箱的智能故障诊断。通过对华中科技大学齿轮箱传动系统动力学实验平台的故障数据进行分析,验证了KOA-CNN-BiLSTM-AM方法的有效性和实用性。与现有多种方法进行对比,结果表明:所提方法能够更有效地提取故障特征信息,并具有更高的故障类型识别率。In order to guarantee the reliable operation of large wind turbines and realize the early fault identification of wind turbines,a hybrid model of deep neural network early fault diagnosis was proposed by integrating Kepler optimization algorithm(KOA),convolutional neural network(CNN),bidirectional long short-term memory(BiLSTM)and attention mechanism(AM).After pre-processing the original vibration data of the gearbox,the CNN+BiLSTM method was used to establish the logical relationship between the input parameters and output parameters,and the AM+KOA method was integrated to automatically and effectively extract the deep fault feature information,which could improve the problems that the traditional deep neural network is prone to fall into the local optimum,slow convergence speed and insufficient ability to extract the features that leads to the poor diagnostic results,and to realize the intelligent fault diagnosis of the wind turbine gearbox.By analyzing the fault data of the gearbox driveline dynamics experimental platform at Huazhong University of Science and Technology,the effectiveness and practicality of the KOA-CNN-BiLSTM-AM method was verified.Comparison and analysis with existing methods show that the method can extract fault feature information more effectively and it has a higher fault type recognition rate.

关 键 词：风电齿轮箱 故障诊断 深度学习 注意力机制 开普勒优化算法 

分 类 号：TH165.3[机械工程—机械制造及自动化]