Evaluating Cracking Deterioration of Preventive Maintenance–Treated Pavements Using Machine Learning

Abstract

Preventive maintenance treatments are extensively used in pavements due to their many benefits, including lowering life-cycle costs. Although the effectiveness of preventive treatments has been evaluated, many highway projects that have implemented preventive maintenance treatments have failed to achieve the expected maintenance effectiveness. Transitioning from empirical decision-making to data-dependent solutions is identified as an urgent need. Accurately modeling and predicting the deterioration of a treated pavement can be used to accurately predict the future condition of a pavement and the effectiveness of a treatment. In this study, long-term pavement performance (LTPP) data were used to develop relationships between pavement deterioration and climate, traffic, and pavement conditions for pavements treated with overlay, slurry seal, crack seal, and chip seal. The variation trends in rates of deterioration (RD) and increases in rates of deterioration (IRD) after receiving treatment were analyzed using measured LTPP data. Machine-learning concepts using an artificial neural network (ANN) were used to model the complicated relationship between output variables RD and IRD and input climate, traffic, and pavement condition variables. The trained ANN for each treatment was used to make predictions for RD and IRD up to 13 years after receiving treatment under different extreme climate, traffic, and pavement conditions. The results enabled the determination of which treatments and which extreme climate, traffic, and pavement condition conditions predicted high RD and IRD values.