No abstract

In its lifetime, a dam can be exposed to significant water level variations and seasonal environmental temperature changes. The structural safety control of a concrete dam is supported by monitoring activities and is based on models.In practice, the interpretation of recorded concrete dam displacements is usually based on HST (hydrostatic, seasonal, time) statistical models. These models are widely used and consider that the thermal effect can be represented by a seasonal function. The main purpose of this paper is to present an HTT (hydrostatic, thermal, time) statistical model to interpret recorded concrete dam displacements. The idea is to replace the seasonal function with the use of recorded temperatures that better represent the thermal effect on dam behavior. Two new methodologies are presented for constructing HTT statistical models, both based on principal component analysis applied to recorded temperatures in the concrete dam body. In the first method, principal component analysis is used to choose the thermometers for the construction of the HTT model. In the second method, the thermal effect is represented by the principal components of temperature of selected thermometers.The advantage of these methods is that the thermal effect is represented by real temperature measured in the concrete dam body. The HTT statistical models proposed are applied to the 110 m high Alto Lindoso arch dam, and the results are compared with the HST displacement model.

No abstract

To improve the effectiveness of concrete dam safety control in real time, a method is presented for the construction of decision rules for the early detection of developing failure scenarios. The decision rules are based on the use of linear discriminant models developed with data obtained through mathematical models of the dam's behaviour. The aim is to combine the physical quantities measured by the automated monitoring system of the dam, appropriately weighted, into a new single index allowing the classification of the observations into one of two classes (normal behaviour and development of a failure scenario).

The protection of critical engineering infrastructures is vital to today's society, not only to ensure the maintenance of their services (e.g., water supply, energy production, transport), but also to avoid large-scale disasters. Therefore, technical and financial efforts are being continuously made to improve the safety control of large civil engineering structures like dams, bridges and nuclear facilities. This control is based on the measurement of physical quantities that characterize the structural behavior, such as displacements, strains and stresses. The analysis of monitoring data and its evaluation against physical and mathematical models is the strongest tool to assess the safety of the structural behavior. Commonly, dam specialists use multiple linear regression models to analyze the dam response, which is a wellknown approach among dam engineers since the 1950s decade. Nowadays, the data acquisition paradigm is changing from a manual process, where measurements were taken with low frequency (e.g., on a weekly basis), to a fully automated process that allows much higher frequencies. This new paradigm escalates the potential of data analytics on top of monitoring data, but, on the other hand, increases data quality issues related to anomalies in the acquisition process. This chapter presents the full data lifecycle in the safety control of large-scale civil engineering infrastructures (focused on dams), from the data acquisition process, data processing and storage, data quality and outlier detection, and data analysis. A strong focus is made on the use of machine learning techniques for data analysis, where the common multiple linear regression analysis is compared with deep learning strategies, namely recurrent neural networks. Demonstration scenarios are presented based on data obtained from monitoring systems of concrete dams under operation in Portugal.