Analysis and interpretation of observed dynamic behaviour of a large concrete dam aided by soft computing and machine learning techniques

Abstract

The nowadays-available dynamic monitoring equipment integrating sensitive low-noise sensors creates an opportunity to implement continuously operating dynamic monitoring systems in dams and validate the suitability of these systems to monitor such massive structures with the goal of detecting damage. The continuous characterisation of the dam modal properties during important variations of the water level and temperature is a unique experimental result, which is particularly interesting for the calibration of numerical models that consider water–structure interaction.

Using a quite rare database collected in a large concrete dam, the Baixo Sabor dam in this case study, a methodology based on machine learning techniques and soft computing is proposed for the analysis and interpretation of observed dynamic behaviour of concrete dams based on models HST (hydrostatic, seasonal, time). For this model, two methodologies are applied, Multiple Linear Regression and MultilLayer Perceptron Neural Network, to characterise the water level effect and the thermal effect related to the seasonal variation of temperature during one year period. A spectral analysis based on wavelet transform is also presented to characterise the thermal effect of daily temperature variations.

The Baixo Sabor dam is a concrete double-curvature arch dam, 123 meters high, located in the northeast of Portugal, which is being monitored by a dynamic monitoring system that comprises 20 uniaxial accelerometers. The results are compared and discussed. The results of this study show that the methodology proposed is suitable for a better understating of the observed dynamic behaviour and opens new opportunities for dam safety control activities.