Please use this identifier to cite or link to this item:
Authors: Ramírez Villanueva, Fredy Gabriel
Cardozo López, Sergio Gabriel
Keywords: Hormigón armado
Aprendizaje automático
Redes neuronales
Vector soporte
Issue Date: Oct-2019
Publisher: Postgrado en Ingeniería Civil FIUNA
Citation: Técnicas de aprendizaje automático para análisis a flexión en vigas rectangulares de hormigón armado/ Fredy Gabriel Ramírez Villanueva. – San Lorenzo: PPGIC / UNA, 2019.
Abstract: Four machine learning techniques were used, namely, Linear Regression (LR),Support Vector Machine (SVM), Artificial Neural Networks (ANN) and Extre me Learning Machine (ELM) for the applicability analysis of these techniques to the study of rectangular reinforced concrete beams subjected to bending mo ments. Two algorithms were previously developed, the first one in order to obtain the ultimate bending moment of a reinforced concrete beam with rectangular cross-section, and the second one, to obtain beam deflections using the moment curvature relationship, both by limit state method following the guidelines of the Eurocode No. 2. Using the aforementioned algorithms, a database for beam sectional analysis was generated, which considers the characteristic strength of steel fyk, and concrete fck, base b and height h of the cross section, the area of longitudinal tensile reinforcement Ast, the compression reinforcement Asc, as well as the ultimate beinding moment Mu. This database was subjected to the Prin cipal Components Analysis or PCA, in which the decoupling between variables that define the geometry of elements of the beam cross section and the variables that define the resistance of materials was determined, arriving at a graph that allows the verification of reinforced concrete beams cross sections. Subsequently, machine learning techniques, mentioned above, were used for predictions of ul timate moment bendings of given sections. The same procedure was performed for the analysis of deflections, generating a database with values of fyk, fck, b, h, length L, Ast, Asc, uniformly distributed load q over the entire length of the beam and deflection f, the latter calculated using the moment - curvature re lationship assuming a simply supported beam. After the studies, the use of LR was rejected due to the unreliable predictions in the present case, finding that the other techniques predict the expected results quite well, especially the ANN method. The use of ELM is highlighted by its mathematical simplicity, efficiency in predictions and low computational cost. Finally ANN was used to predict the characteristic strength of concrete from the beam’s deflection. It is concluded that machine learning techniques are powerful tools for beam analysis and could be used in structures in general.
Description: Esta tesis fue juzgada adecuada para la obtención del Grado de MÁSTER EN CIENCIAS DE LA INGENIERÍA CIVIL, y aprobada en su forma final por los profesores orientadores y por el Programa de Postgrado en Ingeniería Civil de la Universidad Nacional de Asunción, en 2019
Appears in Collections:Trabajos de Investigación - Docentes de Ing. Civil

Files in This Item:
File Description SizeFormat 
Tesis Maestría - Fredy Ramírez_2019_compressed.pdfTesis Fredy Ramírez FIUNA 20191,61 MBAdobe PDFView/Open

Licencia de Creative Commons
Todos los trabajos están bajo una licencia de Creative Commons Reconocimiento-NoComercial 4.0 Internacional