| As a primary component of a Pavement Management System (PMS), prediction models are crucial for one or more of the following analyses: maintenance planning, budgeting, life-cycle analysis, multi-year optimization of maintenance works program and authentication of design alternatives. The main focus of the dissertation is to develop pavement deterioration models. Two techniques are used for this purpose: regression and a relatively new technique--namely, artificial neural network (ANN). The study begins with a review of relevant literature with the aim of identifying the commonly employed explanatory variables and the forms of models. A brief introduction to artificial neural networks, and genetic algorithm, employed in the study, is included.;Five pavement families are identified for the model development: original flexible, overlaid flexible, composite, jointed concrete, and continuously reinforced concrete pavements. Models for each family are developed for predicting distresses, roughness, and a composite condition index (Pavement Condition Rating). These distresses/performance measures form the inputs for the maintenance strategy selection of the PMS, under development by the Mississippi Department of Transportation (MDOT). The database employed is divided into 'in-sample' data constituting a major portion (75-90%), with 'out-of-sample' data comprising the remaining part. Totally 27 models are developed, with the in-sample data: seven for original flexible, six for overlaid flexible, six for composite, and four each for jointed concrete and continuously reinforced concrete pavements. The models are subsequently verified with the independent 'out-of-sample' data.;Among the many models attempted, power and exponential forms are dominantly used as they satisfy many boundary conditions yielding physically meaningful relationships. With identical in-sample data, genetic algorithm-based artificial neural networks are trained to yield predictions. In addition to "summary statistics" such as coefficient of multiple determination (R... |
