Survival Analysis of Fatigue Crack Initiation in Thin Asphalt Surfaces Based on In-Service Pavement Data

This paper details a survival analysis of fatigue crack initiation on thin asphalt surfaces using in-service pavement data from a toll road network in southern Africa collected over a 15 to 20 year period. Assessment of time to crack initiation creates a potential data censoring issue that traditional deterministic methods struggle to incorporate, resulting in bias toward pavements that crack earlier. Survival analysis, which models time to an event, in this case crack initiation, is capable of incorporating censored data in the analysis. The Cox proportional hazard model, a semiparametric statistical method, was used to analyze key trends in the data and estimate the survival function. It was found that the base layer index (BLI) determined from falling-weight deflectometer data was a key predictor for fatigue crack initiation. Traffic loading was found to not be significant on its own, reflecting the fact that the pavements were designed to meet the estimated design traffic, but was significant when used in conjunction with BLI. The Kaplan-Meier estimator model, a nonparametric statistical method, was then used to estimate the survival probability curve. For an unstratified data set, the surface age at which there was a 50% probability of observing cracks was approximately 13 years. For segments with a BLI above 90 mu m in the stratified data set, the median survival time was 11 years. The median survival time was not reached for the data set with BLI below 90 mu m; however, it is clear that significantly lower probabilities of crack initiation were observed, with approximately 85% probability of not being cracked at 8 years of age, compared with 65% for the subset with BLI above 90 mu m.