Study of State Estimation Using LSTM for Semi-active Suspension

Abstract

The study described in this paper aim to further enhance the estimation accuracy of the suspension stroke velocity using the vertical acceleration sensor on the sprung mass. Previous studies have confirmed the effectiveness of including the nonlinear characteristics and delays of semi-active damper in the observer model. However, the observer could not estimate correctly near unsprung resonance point. To solve this problem, we try to use LSTM to estimate the stroke velocity. The effectiveness is verified by carrying out simulations and experiments.