Definition of emergency thresholds for dam safety improvement using Artificial Intelligence and a non-Euclidean metric for affinity evaluation.
The overall objective of the project is to improve dam safety by means of the application of artificial intelligence techniques and the development of the concept of affinity among dams based on a non-euclidean metric. Dams are critical infrastructures because they play an essential role in water supply and energy production systems, together with the potential consequences of their malfunctioning or failure. According to the most recent data from MAGRAMA, there are about 300 Dam Emergency Plans still to be elaborated and implemented in Spain. Some of the aspects to be analysed in these studies lack objective criteria of application, and in other cases they can be improved. Specifically, it is necessary to define emergency thresholds for monitoring devices, which are values of certain parameters representative of the behavior of the structure that once exceeded involve the declaration of some kind of emergency. Nowadays, these values are determined from simple statistical analysis, along with the engineering judgment of the dam owner.
In other fields of science some powerful tools have been developed, which might be useful to understand the behavior of complex engineering systems such as dams. This is the case of artificial intelligence tools which are already used in fields such as commerce and biomedicine: Neural Networks, Random Forests, Bayesian Networks or Complex Networks, among others.
The project aims to develope computational tools to solve two practical problems related to dam safety:
1. – More realistic definition of emergency thresholds for monitoring devices.
2. – The estimation of dam behaviour during first impounding and the early years of functioning.
This goal is ambitious and important in order to improve dam safety as faults and failures are more frequent during the first stage of the dam life cycle, when there are not enough records to build a data-based predictive model.
The achievement of this second objective requires the analysis of monitoring data from similar dams. To do this, we will define dimensionless parameters to characterise the dam behavior, as well as an appropriate metric to calculate the degree of similarity or affinity between them.
To reach these goals, the project brings together both specialists in computer science (CIMNE) and dam safety and engineering (Prof. Toledo’s group at UPM). Both teams have previously worked together in research projects in the field of dam engineering and safety, with high degree of interaction between their members.
This R&D project has been funded by: