Planetary gearbox spectral modeling based on the hybrid method of dynamics and LSTM

Abstract

It is the prerequisite for mechanical diagnosis to clarify the mapping relations between vibrations and damage conditions based on the dynamic model analyses and the systematic experiments. However, there is a certain deviation from the actual measurements in the existing planetary dynamic modeling methods which utilizes linear models or just considering simple non-linear factors. To solve the problem, the hybrid spectral modeling method based on the fault dynamics analysis and the long short-term memory (LSTM) network is proposed in this paper. Firstly, the response of several parts in planetary sets are simulated by the classical dynamic method. Then, as an exploratory attempt, an LSTM network is used to rectify the predicted spectral deviation caused by the over-simplification and parameter errors of the dynamic models. From the inputs of simulated spectral sequences of the several components, the LSTM successfully predicts the spectral sequence of the observation. Finally, by the overlapping weighted average, the power spectral density of the planetary system is estimated through the Welch method. The effectiveness and superiority of the proposed method are verified via the experimental data of a planetary gearbox test rig.