基于深度迁移学习的滚动轴承剩余使用寿命预测

Remaining useful life prediction of rolling bearings based on deep transfer learning

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.