Analysis and design of concrete pavement systems using artificial neural networks

The main objectives of this study was to employ artificial neural networks (ANNs) in the analysis of concrete pavement systems and develop ANN-based design tools that will enable pavement engineers to incorporate the state-of-the-art finite element solutions into routine practical design. To achieve these objectives ANN models were trained using the results from the ILLI-SLAB finite element program as pavement analysis engines. The study mainly focused on the development and performance of comprehensive ANN models based on ILLI-SLAB solutions for the analysis of jointed concrete slabs under different aircraft gear and climatic loadings.; Four different ANN models were developed as part of this research for the analysis of pavement systems under dual, dual-tridem and general multi-wheel loading cases. Under the standard dual-wheel and six-wheel dual-tridem type Boeing 777 gear loadings, the ANN models predicted maximum stresses and deflections with average errors of less than 0.5% when compared to those computed by the ILLI-SLAB model. The use of the ANN models also resulted in both a drastic reduction in computation time (ANN models are 50,000 to 300,000 times faster than the finite element models for analyzing the standard gear loadings) and a simplification of the complicated finite element program input and output requirements. The ANN model developed for the analysis of climatic loadings in addition to the standard gear configurations provides solutions for the following three loading cases: (1) aircraft gear loading only, (2) climatic loading (curl) only, and most importantly, (3) simultaneous aircraft gear and climatic loading with average absolute errors of less than 1.4% when compared to those computed by the ILLI-SLAB model. The comprehensive ANN model that analyzes the general multi-wheel loadings can be used to analyze the effects of any possible aircraft gear configuration in designing concrete pavement systems. Since the critical pavement responses are predicted almost instantly by the use of trained ANN models, studying several ‘what if’ scenarios are easily affordable before making any final design decisions, for example, based on concrete fatigue life predictions.; As part of the ANN-based design toolbox concept, a computer program was developed to easily use the developed ANN models. This program reads the desired pavement input parameters and plots the predicted pavement responses for the specific ANN model selected for analyzing the pavement systems. The user does not need to run any complicated finite element analysis to obtain the pavement responses.; The findings of this study proved that ANN models could be used to capture the complex multi-dimensional mapping of a large-scale finite element analysis in its connection weights and node biases. Artificial neural networks can perform such complex mappings in real time. The implementation of mechanistic based pavement design concepts can be easily done with the use of similar ANN-based concepts developed in this research. The methodology followed in this research can be applied to map other available complex programs in all fields of engineering with the help of ANNs.