基于RCED-DQN的滚动轴承故障诊断方法研究  

Research on Rolling Bearing Fault Diagnosis Based on RCED-DQN

作　　者：李家声 王亭强 周杰 马萍[1] 张宏立[1] 苑茹 LI Jiasheng;WANG Tingqiang;ZHOU Jie;MA Ping;ZHANG Hongli;YUAN Ru(School of Electrical Engineer,Xinjiang University,Urumqi 830017,China;Beijing Jinfeng Kechuang Wind Power Equipment Co.,Ltd.,Beijing 830026,China)

机构地区：[1]新疆大学电气工程学院,乌鲁木齐830017 [2]北京金风科创风电设备有限公司,北京830026

出　　处：《组合机床与自动化加工技术》2025年第2期188-193,199,共7页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金项目(52065064,52267010);新疆维吾尔自治区自然科学基金项目(2022D01C367)。

摘　　要：为提升滚动轴承设备故障诊断中传统卷积神经网络模型的特征提取能力和决策能力,增强诊断模型的准确率和泛化性,提出了基于深度强化学习DQN网络模型的RCED-DQN(residual convolutional encoder decoder-DQN,RCED-DQN)故障诊断框架。框架将一维卷积网络和残差编解码器结合,进行无监督预训练拓宽网络结构,挖掘了网络深层特征,解决了深度强化学习网络难以训练、模型难以收敛的问题;然后,采用预训练后的编码器作为特征提取器,与所设计的特征分类器相连接构建DQN算法的Q网络。通过智能体与环境的交互学习出最佳诊断策略,实现了滚动轴承端到端的故障诊断。实验结果证明,融合残差编解码器的深度Q网络在实验数据集下有效地提取故障特征,提高了诊断决策能力;在不同方法、不同工况下的对比实验结果也验证了所提方法的准确性、有效性和泛化性。In order to improve the feature extraction ability and decision-making ability of the traditional convolutional neural network model in the fault diagnosis of rolling bearing equipment,and to enhance the accuracy and generalization of the diagnostic model,RCED-DQN(residual convolutional encoder decoder-DQN,RCED-DQN) based on the deep reinforcement learning DQN network model is proposed as a fault diagnosis framework.The framework combines a one-dimensional convolutional network and a residual encoder-decoder for unsupervised pre-training to broaden the network structure,which taps into the deep features of the network and solves the problem of difficult training and model convergence of deep reinforcement learning networks;after that,the pre-trained encoder is used as a feature extractor and connected with the designed feature classifier to construct the Q-network of the DQN algorithm.The optimal diagnosis strategy is learned through the interaction between the intelligent body and realizes end-to-end troubleshooting of rolling bearing.The experimental results prove that the deep Q-network with fused residual coder and decoder effectively extracts the fault features and improves the diagnostic decision-making ability under the experimental dataset;the results of the comparative experiments under different methods and working conditions also validate the accuracy,validity and generalization of the proposed method.

关 键 词：深度强化学习 故障诊断 残差编解码器 DQN网络 

分 类 号：TH133.3[机械工程—机械制造及自动化] TG66[金属学及工艺—金属切削加工及机床]