摘要
针对轴承剩余使用寿命(RUL)预测模型训练样本少导致预测精度低的问题,提出一种基于深度迁移学习的滚动轴承剩余使用寿命预测方法。首先利用深度信念网络(DBN)和自组织映射神经网络(SOM)直接对原始振动信号构建轴承健康因子(HI),然后以长短时记忆网络(LSTM)模型为基础,通过共享隐含层的迁移方法训练RUL预测模型,最后利用LSTM-DT进行RUL预测。实验证明,构建HI能够精确反映轴承的健康状态,LSTM-DT算法有效提高RUL预测精度。
To address the problem of low prediction accuracy due to insufficient training samples of bearing remaining useful life(RUL)prediction model,a rolling bearing RUL prediction method based on deep transfer learning is proposed.First,the original vibration signal is used by the deep belief network(DBN)and the self-organizing mapping neural network(SOM)to construct the bearing health factor(HI).Then,prediction model is trained based on the LSTM model by the shared hidden layer transfer method.Finally,LSTM-DT is used to predict RUL value.The experiment results show that the constructed HI can accurately reflect the health state of the bearing,LSTM-DT algorithm can effectively improve the accuracy of RUL prediction.
作者
汪立雄
王志刚
徐增丙
林辉
WANG Lixiong;WANG Zhigang;XU Zengbing;LIN Hui(Key Laboratory of Metallurgical Equipment and Control Technology of Ministry of Education,Wuhan University of Science and Technology,Wuhan 430081,CHN;College of Machinery and Automation,Wuhan University of Science and Technology,Wuhan 430081,CHN;The 708 Research Institute of China Shipbuilding Industry Corporation,Shanghai 200011,CHN)
出处
《制造技术与机床》
北大核心
2020年第12期130-134,137,共6页
Manufacturing Technology & Machine Tool
基金
国家自然科学基金项目(51775391)
装备预研基金项目(6142223180312)。
关键词
剩余使用寿命预测
深度信念网络
自组织映射神经网络
轴承健康因子
长短时记忆网络
共享隐含层迁移
remaining useful life prediction
deep belief network
self-organizing feature map neural network
bearing health indicator
long-short term memory network
shared hidden layer transfer
