Tunneling-induced ground movement and building damage prediction using hybrid artificial neural networks

• Auteur principal: Hajihassani, Mohsen
• Format: Thèse
• Langue: anglais
• Publié: 2013
• Sujets: TA Engineering (General). Civil engineering (General)
• Accès en ligne: http://eprints.utm.my/37932/1/MohsenHajihassaniPFKA2013.pdf

The construction of tunnels in urban areas may cause ground displacement which distort and damage overlying buildings and services. Hence, it is a major concern to estimate tunneling-induced ground movements as well as to assess the building damage. Artificial neural networks (ANN), as flexible non-linear function approximations, have been widely used to analyze tunneling-induced ground movements. However, these methods are still subjected to some limitations that could decrease the accuracy and their applicability. The aim of this research is to develop hybrid particle swarm optimization (PSO) algorithm-based ANN to predict tunneling-induced ground movements and building damage. For that reason, an extensive database consisting of measured settlements from 123 settlement markers, geotechnical parameters, tunneling parameters and properties of 42 damaged buildings were collected from Karaj Urban Railway project in Iran. Based on observed data, the relationship between influential parameters on ground movements and maximum surface settlements were determined. A MATLAB code was prepared to implement hybrid PSO-based ANN models. Finally, an optimized hybrid PSO-based ANN model consisting of eight inputs, one hidden layer with 13 nodes and three outputs was developed to predict three-dimensional ground movements induced by tunneling. In order to assess the ability and accuracy of the proposed model, the predicted ground movements using proposed model were compared with the measured settlements. For a particular point, ground movements were obtained using finite element model by means of ABAQUS and the results were compared with proposed model. In addition, an optimized model consisting of seven inputs, one hidden layer with 21 nodes and one output was developed to predict building damage induced by ground movements due to tunneling. Finally, data from damaged buildings were used to assess the ability of the proposed model to predict the damage. As a conclusion, it can be suggested that the newly proposed PSO-based ANN models are able to predict three-dimensional tunneling-induced ground movements as well as building damage in tunneling projects with high degree of accuracy. These models eliminate the limitations of the current ground movement and building damage predicting methods.