Long-term behaviour analysis of Santa Luzia dam affected by concrete swelling reactions using machine learning techniques

Abstract

This paper aims to analyze the structural effects of concrete swelling reactions in a large dam using Machine Learning (ML) techniques. The analysis is based on geodetic displacements data collected from the beginning of dam operation in 1942 until now, covering the entire lifetime of the dam, totaling a period analysis over 80 years. The well-known Hydrostatic-Seasonal-Time (HST) analysis was used to compare the insights obtained from different data-based models. The paper details the results obtained by two ML techniques, Boosted Regression Trees (BRT) and Multilayer Perceptron Neural Networks (NN), which are compared with the results obtained from Multiple Linear Regression (MLR) technique. The purpose of applying these new techniques is to assess the evolution of the swelling process and its development patterns throughout the dam's lifetime. The case study is the Santa Luzia dam, designed by André Coyne and located in the center of Portugal, consisting of a main arch dam 75 m high and a gravity arch that closes the upper section of the left bank. The dam is subjected to a process of deleterious concrete swelling caused mainly by the alkali-silica reactions (ASR). Evidence of the structural effects of this type of deleterious process are the progressive displacements, upwards of the crest and upstream of the arches, as well as concrete cracking.