Running Condition Identification of High-speed Shaft Based on Shaft-end-data Driven LSTM-CNN

Aiming at the problem that it is difficult to accurately real-time monitor and identify the running condition of high-speed shafts with complex structures, a composite neural network shaft running condition identification method based on the shaft-end-data driven is proposed. Firstly, a composite neural network model（LSTM-CNN） based on Long short-term memory（LSTM） and Convolutional neural networks（CNN） is proposed. A dual-disk shaft dynamics simulation model is then established. The Newmark-βmethod is used to numerically solve the shaft system for acquiring the dynamic response characteristics of the key fixed nodes of the shaft system; at the same time, the dynamic response characteristics of the key rotating nodes are obtained based on the finite element simulation. Two types of data are input into the LSTM-CNN model for running condition identification, and its accuracy and efficiency are compared and analyzed. Finally, a high-speed shaft experimental platform is designed and established, and the shaft-end data and fixed-end data are respectively used to train and test the LSTM-CNN model. The performance of different models for the running condition identification of the high-speed shaft is compared. Simulation and experimental verification analysis results show that the shaft-end-data driven LSTM-CNN have better running condition identification accuracy and efficiency than that based on the fixed-end-data driven.