Abstract:
Dam safety control activities require an accurate knowledge of each specific dam, with the purpose of
defining and justifying the judgment about its safety. This task is mainly supported by the cross validation
between simulation models, measurements provided by the monitoring systems, and the parameters that
characterize the dam's behaviour. The main issue is the assessment of the actual structural behaviour in real
conditions, which can be used to detect any anomaly and/or malfunction in advance.
Over the years, we can verify a significant evolution in the process of interpreting the physical quantities
provided by dam monitoring systems. Nowadays, automated data acquisition systems have become a reality
in several dams. These systems can be used to support the analysis for dam safety assessment in real time,
but also lead to the increase of requirements related to the management, processing and analysis of large
amounts of data.
With the development of information systems to support the activities related to dam safety control,
particularly the management of a large quantity of information, new challenges related to the management
and analysis of information in real time are raised.
The implementation of an Internal Early Warning System (IEWS) based on the automatic analysis of a large
quantity of data in real time allows the early identification and notification of potential abnormal situations and
makes the person responsible able to focus on other activities related with the dam safety control of dams.
This paper addresses a proposal for an IEWS able to generate warnings in real time when non-accordance
between observed and predicted values is verified. Subjects related to the definition of the thresholds for
quantities measured by the monitoring system, as well as the notification process to the person responsible
for the dam's safety control, are also discussed.
This paper presents the actual Portuguese experience. Aspects related with the requirements for the IEWS
are approached. Threshold definition with quantitative interpretation models based on statistical methods,
such as multiple linear regression models, and artificial neural network models, is discussed.
