Thesis DESARROLLO DE UN MODELO DE PREDICCIÓN DEL ESCALONAMIENTO EN PAVIMENTOS RÍGIDOS
Date
2017
Authors
NAVARRO COFRÉ, JORGE IGNACIO
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
En el contexto de continuar avanzando en la implementación de la Guía Empírico Mecanicista de Diseño de Pavimentos AASHTO 2008 (GEMDP) en Chile, este estudio se enfoca en el modelo de predicción del escalonamiento de pavimentos rígidos. Específicamente, esta memoria de título se orienta en la parte mecanicista del modelo, la cual busca calcular las respuestas del pavimento (deflexiones), desarrollando un método de redes neuronales artificiales para disminuir los tiempos de cálculo en comparación con métodos tradicionales como el de elementos finitos.Los datos necesarios para el entrenamiento de las redes neuronales artificiales se obtienen realizando una caracterización y análisis de más de 27.000 estructuras de pavimentos rígidos con distintas configuraciones de carga posicionadas en la esquina de la losa, determinando así la respuesta crítica del pavimento con respecto al deterioro de escalonamiento mediante el software ISLAB2000. A partir de los resultados obtenidos con el entrenamiento de la red neuronal artificial, se desarrolla una herramienta de cálculo para la predicción de las deflexiones de esquina de la losa cargada y no cargada utilizando el programa computacional MATLAB. Con los resultados arrojados por el programa se puede calcular el escalonamiento mediante el modelo de escalonamiento de la GEMDP calibrado en el proyecto Innova Chile – Corfo 11BPC-10220 para las condiciones chilenas.Las seis redes neuronales programadas en este estudio presentan una buena capacidad de predicción de las deflexiones de esquina de la losa cargada y no cargada exhibiendo errores bajos si se compara con los resultados calculados mediante elementos finitos por lo que se asegura un adecuado comportamiento del modelo para utilizarlo en la predicción del escalonamiento mediante los métodos recomendados por la GEMDP.
In order to continue in the implementation of the Mechanistic-Empirical Pavement Design Guide AASHTO 2008 (MEPDG) in Chile, this study focuses in the faulting model of Jointed Plain Concrete Pavement. Specifically, this thesis emphasizes in the mechanistic part of the model, which calculate the pavement responses such as deflections, developing an artificial neural network model to reduce the computation time compared to traditional methods like finite elements.The necessary data to train the artificial neural networks are obtained with the analysis of more than 27.000 JPCP structures with different axle load configurations positioned in the corner of the slab. ISLAB2000 is used to do the analyses and calculate the critical response of the pavement. Based on the results obtained from the artificial neural network training a computational tool is developed in MATLAB for the prediction of the corner deflection of the loaded and unloaded slab. With the results of the computational tool it is possible to predict mean faulting using the MEPDG faulting model calibrated for the Chilean conditions in the Innova Chile – Corfo 11BPC-10220 project.The six artificial neural networks developed in this thesis present a good prediction capability of the corner deflections of the loaded and unloaded slab showing low errors compared to the results of the finite element program. Therefore, the neural network model guarantee a good performance to use it in the prediction of mean faulting with the recommended methods of the MEPDG.
In order to continue in the implementation of the Mechanistic-Empirical Pavement Design Guide AASHTO 2008 (MEPDG) in Chile, this study focuses in the faulting model of Jointed Plain Concrete Pavement. Specifically, this thesis emphasizes in the mechanistic part of the model, which calculate the pavement responses such as deflections, developing an artificial neural network model to reduce the computation time compared to traditional methods like finite elements.The necessary data to train the artificial neural networks are obtained with the analysis of more than 27.000 JPCP structures with different axle load configurations positioned in the corner of the slab. ISLAB2000 is used to do the analyses and calculate the critical response of the pavement. Based on the results obtained from the artificial neural network training a computational tool is developed in MATLAB for the prediction of the corner deflection of the loaded and unloaded slab. With the results of the computational tool it is possible to predict mean faulting using the MEPDG faulting model calibrated for the Chilean conditions in the Innova Chile – Corfo 11BPC-10220 project.The six artificial neural networks developed in this thesis present a good prediction capability of the corner deflections of the loaded and unloaded slab showing low errors compared to the results of the finite element program. Therefore, the neural network model guarantee a good performance to use it in the prediction of mean faulting with the recommended methods of the MEPDG.
Description
Catalogado desde la version PDF de la tesis.
Keywords
AASHTO 2008 , DISENO DE PAVIMENTOS , ESCALONAMIENTO DE PAVIMENTOS , GEMDP