Relating Vehicle Skid Friction To Pedestrian Slip Resistance on Pavements with Machine Learning Integration and Master Curve Development for Distress Prediction

Abstract

This dissertation addresses one of the main concerns in pavement engineering in regard to safety. Agencies need to relate vehicular skid resistance with pedestrian slip resistance on the pavement surfaces located at critical locations such as crosswalks. Thus, there was a critical need to examine and relate such parameters on typical marking materials by: conducting an extensive assessment at both lab and field conditions under alternative environmental and surface conditions; developing prediction models linking such parameters; and identifying friction specification revisions for acceptance, as well as routine condition evaluation recommendations. To achieve these objectives, the research was organized into the following steps: lab and field experimentation; an extensive statistical analysis assessment between various parameters and testing conditions; an attempt to relate friction evaluation measurements to “ground-truth” conditions by incorporating image-based microtexture analysis; and development of machine learning modeling.

The first study phase explored the friction assessment of widely used pavement marking materials in Maryland under various surface conditions, including dry (D), wet (W), and icy (I) surfaces. The British Pendulum Tester (BPT) was used to assess both laboratory and field performance in in-service conditions. Material-dependent variability and consistent trends between field and laboratory data were obtained, paving the way for the use of laboratory testing as an alternative to field inspection, which traditionally requires costly traffic control and is associated with significant safety concerns.

The second phase addresses the repeatability and reproducibility of BPT measurements across different devices and operators. Results confirmed that BPT provides consistent measurements for both tire skid and pedestrian slip resistance. Microscopic imaging and Fast Fourier Transform (FFT) analysis demonstrated how microtexture and macrotexture are affected by surface wear and material composition, and how these significantly influence friction levels.

The third phase considered the use of the Aggregate Image Measurement System (AIMS) to analyze pavement marking surface texture characteristics. Using asphalt cores with and without markings, surface condition texture parameters, such as root mean square (RMS), mean profile depth (MPD), and mean texture depth (MTD), were quantified across four scan directions. Directional sensitivity, regression among indices, quartile classification, and texture distribution analyses were performed. Results confirmed that AIMS-derived texture metrics align with BPN friction values and distinguish material performance.

The next phase considered the use of supervised machine learning to predict British Pendulum Numbers (BPN) based on a dataset of 1,092 laboratory and field data. Comparison of five ML models was examined, including Random Forest, XGBoost, Support Vector Regression, Gaussian Process Regression, and Multilayer Perceptron. XGBoost yielded the best predictive performance with a coefficient of determination (R²) of 0.951. Analysis of feature importance identified surface condition, material type, and test environment as the dominant predictors.

Since highway agencies have a multiyear set of data related to distress, the final portion of the study was focused beyond friction, in an effort to develop distress prediction models using the master Curve approach. Since the primary distresses governing pavement performance were identified to be fatigue cracking and permanent deformation (i.e., rutting), as documented by many Balanced Mix Design (BMD) studies currently under development in the US, data from the Long-Term Pavement Performance (LTPP) database were used with the “master curve” approach in developing such prediction models. Such modeling provided highly precise predictions and thus provided the means to identify future patterns of deterioration with reduced need for time-consuming and costly field surveys.

Finally, it is worth mentioning that the importance and value of relating vehicular skid friction to pedestrian slip resistance, and important findings, contributions, and value of this study in this area, addressing systematically and linking such critical parameters for safety, were recognized nationally with the AASHTO 2025 Sweet 16 High-Value Research Award.