Output-only road roughness identification from vehicle axle accelerations through a universal smoothing method

TL;DR

This paper introduces a universal smoothing method for accurately identifying road roughness profiles from vehicle axle accelerations, demonstrating improved stability and robustness over existing techniques in real-world tests.

Contribution

It develops a novel output-only estimation approach using a universal smoothing method with regularisation, enhancing robustness and accuracy in road roughness identification.

Findings

US method outperforms Dual Kalman filter in accuracy

Stable reconstructions across different speeds and noise levels

Method remains numerically well conditioned under high noise

Abstract

This paper presents an output-only method to identify road roughness profiles from axle accelerations of a moving vehicle. A two degree of freedom half-car model is discretised with a zero-order hold and a backward-difference approximation of the roughness rate, which introduces both the current and previous roughness inputs into the observation equation. This modification enables joint input state estimation with limited measurements using a Universal Smoothing (US) method, which belongs to the family of Minimum-Variance Unbiased (MVU) estimators. To improve numerical robustness under high process noise, stemming from modelling errors such as neglected bridge vehicle interaction, the system inversion is regularised by truncated singular value decomposition. The method is validated on a full-scale bridge with a commercial SUV at two different speeds. Compared to the Dual Kalman filter and an MVU-based smoother, the proposed US achieves stable, accurate reconstructions across different scenarios and remains numerically well conditioned when noise increases. Practical aspects of tuning, window length selection, and computational cost are also discussed.