Aggregating network-level pavement performance data based on Gaussian Mixture Models

Network-level pavement performance data collected from many pavement segments often exhibit heterogeneity. Summarizing these data with their mean and standard deviation is far from enough to characterize network-level performance data. In order to account for performance heterogeneity, this study proposes Gaussian Mixture Model (GMM) to aggregate pavement performance data for network-level pavement management. The Expectation-Maximization (EM) algorithm is adopted for parameter estimation. Pavement performance data are finally aggregated using mixture weights, means, and standard deviations of the fitted GMMs. The proposed method is tested on the rutting data of Freeway G30 in Jiangsu, China. The results show that GMMs can capture the characteristics of heterogeneous performance data and provide an accurate summary for network-level pavement management. The aggregated performance data of different years can be easily compared using parameters or mixture density plot of corresponding GMMs. Additionally, this method can classify pavement segments into multiple groups according to pavement condition.