Genetic Programming in Civil Engineering

Abstract

[Abstract]: Advances in the field of Artificial Intelligence have a strong influence on the different areas of Civil Engineering. New methods, techniques and algorithms of Artificial Intelligence allow engineers to use these new tools in different ways and on problems of diverse type. In many fields of Civil Engineering, the way to obtain models of physical processes is based on experimentation and obtaining test data. Experts in civil engineering try to extract knowledge from these data in the shape of mathematical equations and expressions. This behaviour is similar to the learning process of a human, and therefore the Artificial Intelligence techniques that emulate this behaviour can be used. One of the great advantages of Genetic Programming is the ability to provide results as mathematical equations. This is a key issue for the civil engineer, because they need to understand and analyse the results of the predictions. This is the main reason why civil engineers do not favour black box tructures like Artificial Neural Network (ANN), which only produce results. Also, the use of mathematical expressions can show the degree of complexity and allows the modification of their behaviour by altering the terms of the equation. In this chapter we show the application of the Genetic Programming (GP) technique in two real-life examples in civil engineering: hydrology and construction. One of the most important processes in Hydrology is the so-named “Rainfall-Runoff transformation process”, meaning the process in which the rain fallen over an area concentrates and runoff-flows over the surface. Using GP we can obtain a mathematical expression that predicts the behaviour of this process. In construction, an important process is the prediction of the ultimate stress in a reinforcing bar at the time of the failure of bond strength. Currently few equations, very different from each other, are capable of performing a prediction of the anchorage length of rebar in structural concrete. These equations are based on the basic straight anchorage length necessary to reach failure of the rebar. Unlike the previous case, we have used modified GP that allows to guide the search process to improve the prediction of the current equations. These examples show how GP can use experimental data in a more efficient and "intelligent" way and produce new tools for modelling and prediction of phenomena, the possibility of creating "virtual laboratories", virtual environments that we can test many phenomena.