基于Yu范数深度迁移度量学习的夹送辊剩余寿命预测

Remaining life prediction of pinch rolls based on Yu parametric deep migration metric learning

针对夹送辊历史数据少和相关寿命预测方法匮乏的问题,提出基于Yu范数深度迁移度量学习的夹送辊剩余寿命预测方法。首先使用Yu范数深度度量学习(DMN-Yu)对振动信号提取深层特征,并以主成分分析法(PCA)和自组织映射神经网络(SOM)相结合对特征进行约简,构建一维健康因子(HI);再结合长短时记忆网络(LSTM)模型,通过迁移策略利用共享隐含层的方法对目标夹送辊进行预测分析。实验验证,经过深度迁移学习的LSTM模型预测效果更好,对夹送辊设备的健康状态评估及剩余使用寿命预测具有一定的指导意义。

To address the problems of little historical data of pinch rolls and the lack of relevant life prediction methods,a method for predicting the remaining life of pinch rolls based on Yu parametric deep migration metric learning was proposed.Firstly,Yu parametric depth metric learning(DMN-Yu)was used to extract deep features from vibration signals,and the features were combined with principal component analysis(PCA)and self-organizing mapping neural network(SOM)to construct a one-dimensional health factor(HI)by reduction.Then,the target pinch rolls were predicted and analyzed by a migration strategy using a shared hidden layer approach in combination with a long and short term memory network(LSTM)model.The experiment verified that the LSTM model with deep migration learning had a better prediction effect,which was a guideline for the health state assessment and remaining service life prediction of pinch rolls equipment.