Validating road profile reconstruction methodology using ANN simulation on experimental data

Abstract

This paper reports the performance of an ANN-based road profile reconstruction methodology on measured data. The methodology was previously verified on numerical data and it was shown that road profiles and their associated defects could be reconstructed to within a 20% error at a minimum correlation value of 94%. The data used in the present paper were measured on a Land Rover Defender 110 using an eDAC-lite measurement system. The measurements were carried out under different test conditions, namely, different road surface profiles, different vehicle suspension settings, and different vehicle speeds. The neural network was trained with data extracted from 20 m length of a typical test run for each road profile. The results confirm the findings of the numerical study with the methodology achieving a maximum error of about 25% and correlation of above 90%. The methodology performs relatively much better in reconstructing bumps than the Belgian pave.