Mechanical state prediction based on LSTM neural netwok

Effective mechanical state prediction systems are critical to modern manufacturing systems and industries. As a method of deep learning algorithm, Recurrent neural network, (RNN) has been playing an increasingly important role in the field of time series prediction. In order to solve the problem of hard training and gradient extinction of RNN model, a long short-term memory network (LSTM) algorithm is proposed and applied to the prediction of mechanical state (PMS). On the basis of the motor bearing data, the simulation is carried out. Aiming at the non-stationary of bearing data, the empirical mode decomposition (EMD) is used to decompose the bearing data into stationary singles, and the intrinsic mode function (IMF) energy entropy is calculated as the feature of mechanical state. In comparison with support vector regression machine (SVRM), LSTM achieves a better result in the single-step prediction of mechanical state. It shows that LSTM is superior in machine state prediction and monitoring.