Browsing by Author "Bukenya, Patrick"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Open Access
    Ambient vibration testing of concrete dams
    (2011) Bukenya, Patrick; Moyo, Pilate; Beushausen, Hans-Dieter
    In this thesis, seven techniques namely; rational fractional polynomial, complex exponential, frequency domain decomposition (FDD) based techniques which include; frequency domain decomposition (FDD), enhanced frequency domain decomposition (EFDD), curve fitting frequency domain decomposition (CFDD) and stochastic subspace identification (SSI) methods namely; unweighted principal component (UPC), principal component (PC) and canonical variant analysis (CVA)) have been applied to data from ambient vibration testing of two concrete dams namely; Roode Elsberg and Kouga dams.
  • Thumbnail Image
    Item
    Open Access
    Structural performance evaluation of concrete arch dams using ambient vibration monitoring and GNNS systems
    (2020) Bukenya, Patrick; Moyo, Pilate
    Societies around the world are heavily dependent on civil engineering infrastructures such as concrete dams that provide necessities such as water supply for irrigation, hydroelectric power generation and prevention of floods. As a result, it is important to ensure that concrete dams are protected such that their failure is avoided. To ensure the structural safety of these structures, concrete dams are continuously monitored by sensors installed on the dam to detect any unusual behaviour. Data collected by the sensors include environmental variables (temperature and water levels) and dam responses (deformations, stresses, strains, natural frequencies) which indicate the structural behaviour of dams. This implies that research in the analysis of the collected data is very important. Methods used in the analysis of dam monitoring data include data-driven models, physical-based models and hybrid models. Data-driven models utilise environmental variables as independent factors and dam responses as dependent factors. The trends in the dam responses can be learnt for purposes of monitoring and prediction by understanding the interactions between environmental variables and dam responses. Dam specialists have mainly focused on predicting the static deformations of dam walls using environmental variables through statistical modelling. Dynamic properties such as natural frequencies also provide valuable information on the structural behaviour of dams as they are influenced by the changes in environmental variables. To the best of my knowledge, there is no scientific literature that has studied the influence of environmental variables on natural frequencies through statistical modelling. The increase in the amount of data collected from monitoring devices installed on dams has led to the use of advanced statistical models to extract important information about the behaviour of dams. Machine learning algorithms have been developed to solve problems of large data sets and nonlinearity between variables. In particular, there are no studies that exist in the prediction of natural frequencies using measured environmental variables (water level and temperature). Therefore, the purpose of this study was to understand the effect of water level and temperature on natural frequencies and deformations. The case study used in this thesis is Roode Elsberg dam, a concrete arch dam located in South Africa. Natural frequency data used was collected between December 2014 and June 2017 while deformation data used was between January 2012 and June iii 2016. Observations indicated that water level was the dominant factor driving the variation of natural frequencies with temperature affecting the natural frequency variations in periods of constant water levels. On the other hand, temperature was the driving factor in deformation variations with water level also affecting deformation variation. Due to the nonlinear relationship between environmental variables and dam responses, a machine learning-based algorithm known as Gaussian process regression models were developed to predict natural frequencies. In Gaussian process regression, the choice of a covariance function is very important in producing good results. The ability of the different covariance functions in Gaussian process regression models, to predict natural frequencies and dam deformations, was studied. The performance of Gaussian process regression models was compared with other machine learning algorithms (BRT, SVR and ANN), multivariate adaptive regression splines and the commonly used MLR models in the prediction of natural frequencies and deformations. Results suggested that the GPR model is the most suitable and more accurate in the prediction of dam responses. Finally, robust statistics are introduced in the identification of anomalies in the collected data. Furthermore, univariate methods are used to identify any abnormalities in dam behaviour. Results indicated, there were no abnormalities in the dam behaviour.