Remaining Useful Life Prediction of Rolling Bearings Based on Heterogeneous Knowledge Distillation Network
Aiming at the problems of low prediction accuracy and limited resources of edge equipment in rolling bearing life prediction,a heterogeneous knowledge distillation network is proposed to predict the re-maining useful life of rolling bearings.The network uses a teacher student knowledge distillation architec-ture.Firstly,the network introduces the self-attention mechanism and the long-term short-term memory net-work to construct a teacher model with high prediction accuracy.Secondly,on the basis of convolutional neural network,a variational auto-encoder is introduced to construct a student model with strong feature ex-traction ability,less parameter quantity and lower complexity.Then,a composite loss function is designed to train the absorption ability of the student model to the knowledge of the teacher model and the adaptability to the training data.Finally,the life prediction experiment is carried out on the XJTU-SY bearing dataset.The results show that compared with other prediction methods,the proposed method can effectively reduce the number and complexity of model parameters and have higher prediction accuracy.
rolling bearingremaining useful lifeknowledge distillationself-attention mechanismvaria-tional auto-encoder
NETL
NSTL
万方数据
