Estimation of Pavement Macrotexture by Principal Component Analysis of Acoustic Measurements

Abstract

A method is developed to estimate the mean texture depth (MTD) value, which to some extent represents pavement quality, using measurements from a microphone mounted underneath a moving vehicle. Such measurements will include tire-generated sound that carries information about the road macrotexture as well as noise generated by the wind and vehicle. The proposed method uses principal component analysis (PCA) to differentiate important information about the road surface from noisy data while the vehicle is moving. The variations in frequency of the noise are assumed to be small compared with the variations in frequency of the signal related to the road-surface condition, which allows the PCA approach to separate noise from signals that carry information about the road-surface condition. The analysis begins with acoustic pressure measurements being made over various road-surface conditions underneath a moving vehicle. Fourier transforms are taken over various time windows and a PCA is performed over the resulting vectors. This yields a set of PC vectors representing the road-surface conditions. The frequency range of concern is from 0 to 2,000 Hz, according to the amplitude of frequency spectra of collected acoustic measurement. The pavement-macrotexture depth (i.e., MTD) is estimated by matching the PC vector set derived from unknown road conditions with one of the vector sets of known road conditions. Successful applications of this method are demonstrated by accurate estimations of the MTD of pavement directly from acoustic measurements. The results indicate that PCA is a powerful approach to eliminate the noise that is not associated with the road surface, and therefore, the PC vectors can be used to accurately match the MTD values. The PCA approach for tire-generated sound might also be used to differentiate subsurface road conditions, a precursor of many defects such as potholes and severe cracking.