Neural Networks Based Models for Mechanistic-Empirical Design of Rubblized Concrete Pavements

摘要

Rubblization is an in-place rehabilitation technique that involves breaking the concrete pavement into pieces. This process results in a structurally sound, rut resistant base layer which prevents reflective cracking (by obliterating the existing concrete pavement distresses and joints) that can then be overlaid with Hot-Mix Asphalt (HMA). The design of the structural overlay thickness for rubblized projects is difficult as the resulting structure is neither a true rigid pavement nor a true flexible pavement. The HMA overlay thickness design methodology currently used in the state of Iowa is purely empirical. In the Mechanistic-Empirical (M-E) design approach developed for the analysis and design of rubblized concrete pavements in Iowa, the tensile strain at the bottom of the HMA layer (εt) is used to predict fatigue life using an HMA fatigue design algorithm and the vertical compressive strain on top of the subgrade layer (εc) is used to consider subgrade rutting. In the current study, the use of Artificial Neural Networks (ANN)-based structural models for predicting the critical strains based on FWD deflection data, is successfully demonstrated. The ANN-based structural models were validated by comparing the ANN-based strain predictions with the field-measured strains from an instrumented trial project at highway IA-141 located in Polk County, Iowa.