Master's Thesis

Modelação do comportamento observado em barragens de betão com base em técnicas de aprendizagem automática

João Francisco Henriques de Sena Cardoso — 2021

Key information

Authors:

João Francisco Henriques de Sena Cardoso (João Francisco Henriques de Sena Cardoso)

Supervisors:

Juan Tomé Caires da Mata; João Carlos De Oliveira Fernandes de Almeida (João Carlos De Oliveira Fernandes de Almeida)

Published in

June 17, 2021

Abstract

The scope of this dissertation entails the structural health monitoring of concrete dams and different machine learning formulations, to support the analysis and interpretation of the observed behavior in concrete dams. In this dissertation, horizontal displacements, spanning 30 years, were analyzed, and interpreted at two points in the central block of Alto Lindoso dam. At each point, three formulations denominated Hydrostatic-Season-Time (HST), Hydrostatic-Temperature-Time (HTT) and Hydrostatic-Temperature-Foundation-Time (HTFT) were tested, the latter being an innovative proposal within the scope of this work. All formulations allow the consideration of the effects of the reservoir water level, temperature, and time. The differences between them lies in: i) HST considers the effect of air temperature by an annual period sinusoidal function; ii) HTT considers the effect of air temperature through the temperature measured in the dam body; and iii) HTFT considers the effect of air temperature through the temperature measured in the dam body and considers the effect of the deformation of the foundation. The following machine learning methods were used in the formulations: multiple linear regression, artificial neural networks, and random forest. The analysis considered 286 observation campaigns categorized according to learning period and prediction period. All the formulations showed a good estimation capacity, with some potential advantages in the use of the HTFT formulation. The machine learning methods proved to be suitable for this type of application, with the random forest method presenting more beneficial insights results when compared to the other methods.

Publication details

Authors in the community:

Supervisors of this institution:

Fields of Science and Technology (FOS)

civil-engineering - Civil engineering

Publication language (ISO code)

por - Portuguese

Rights type:

Embargo lifted

Date available:

April 5, 2022

Institution name

Instituto Superior Técnico