基于神经网络的去噪模型在轴承故障诊断中的应用  

Application of Denoising Model Based on Neural Network in Fault Diagnosis for Bearings

作　　者：代鸿 刘新宇 DAI Hong;LIU Xinyu(College of Humanities,Chongqing Metropolitan College of Science and Technology,Chongqing 402160,China;School of Mechanical and Electrical Engineering,Chengdu University of Technology,Chengdu 610059,China)

机构地区：[1]重庆城市科技学院人文学院,重庆402160 [2]成都理工大学机电工程学院,成都610059

出　　处：《轴承》2023年第11期87-94,共8页Bearing

摘　　要：针对轴承微弱故障稀疏振动信号的特征提取,提出了基于模型数据协同链接框架的端到端深度网络稀疏去噪(DNSD)策略。建立了全局可微稀疏模型,引入深度神经网络学习超参数,基于轴承内圈故障机理建立了多模式数据集模拟故障信号,通过DNSD对数据集以去噪自编码器的形式进行训练,重建损失并更新网络和稀疏理论的参数,通过轴承内圈故障的仿真和试验验证了DNSD模型在轴承微弱故障特征提取方面的优越性和鲁棒性。Aimed at feature extraction of sparse vibration signals from bearing weak faults,an end-to-end deep network sparse denoising(DNSD)strategy is proposed based on model data collaborative link framework.A global differentiable sparse model is established,and a deep neural network is introduced to learn hyperparameters.Based on mechanism of bearing inner ring fault,a multi-mode dataset is established to simulate the fault signal.The dataset is trained by DNSD in the form of denoising autoencoder to reconstruct the loss and update the parameters of network and sparse theory.The superiority and robustness of DNSD model in bearing weak fault feature extraction are verified by simulation and experiment of bearing inner ring fault.

关 键 词：滚动轴承 故障诊断 特征提取 深度神经网络 稀疏数据 

分 类 号：TH133.33[机械工程—机械制造及自动化] TP389.1[自动化与计算机技术—计算机系统结构]