摘要
针对滚动轴承振动数据耦合程度高,信号特征提取和识别模型建立困难的问题,提出了一种基于深度学习理论的状态监测方法。提取振动信号的时域、频域和时频域特征构成特征向量;通过稀疏自编码非监督学习网络对输入向量进行特征学习,并将单层网络叠加构成深度神经网络;最后采用少量有标签数据对整个深度神经网络进行微调训练,建立轴承状态监测模型。试验结果表明,提出的方法对于轴承状态识别准确率达到90.86%,且性能退化阶段识别率最高,能满足视情维修的工程需求。
The vibration signal of rolling element bearing occurs with a high degree of coupling,which means that the features and recognition model are difficult to build. For solving these problems,we proposed a novel bearing condition-monitoring model based on deep learning. Time domain,frequency domain and time-frequency domain features are extracted. Then these feature vectors are entered into an unsupervised auto-encoder to learn the high-level features. At the same time,the middle layers of the auto-encoder network are stacked into a multilayered network. Finally,a small number of labeled training samples are used to fine-tune the deep learning network. The bearing condition recognition experiment shows that the proposed method achieves state-of-the-art results,and its high accuracy in terms of the performance degradation condition is very helpful when it comes to condition-based maintenance.
出处
《振动与冲击》
EI
CSCD
北大核心
2016年第12期167-171,195,共6页
Journal of Vibration and Shock
基金
国家自然科学基金(NSF51275426)
关键词
深度学习
非监督学习
滚动轴承
视情维修
deep learning
unsupervised learning
antifriction bearing
condition-based maintenance
