Prediction of track irregularities using NARX neural network

Abstract

The paper proposes an approach to predict track irregularities based on accelerations of vehicle body using neural network. Firstly, a simulation vehicle model is constructed in Adams software to collect accelerations data. Secondly, two types of NARX neural networks are listed, and the series-parallel NARX neural network is selected as the inverse model to predict track irregularities. The proposed approach is applied to the estimation of the left vertical irregularity, the left lateral irregularity and level irregularity, and the results show the validity of the proposed method.