| Abstract |
Dam safety is highly relevant to our society due to the importance of the services they provide — hydropower generation, water storage, flood protection, among others— and to the serious consequences of their potential malfunctioning or failure.
In addition, many dams worldwide are approaching the end of their expected service life, which results in higher risk of failure and thus more resources needed for adequate safety control.
In response to this situation, the Spanish administration recently published new technical dam safety standards. Among the main elements for ensuring safety and proper dam operation, the new regulations require the definition of relevant indicators of dam safety and associated quantitative thresholds.
The importance of reliable thresholds and the need for research on the topic is also demonstrated by the activities of the Committee on Dam Surveillance of the ICOLD: a specific Theme was defined on this topic for the upcoming Benchmark Workshop, to be held in Slovenia in 2022.
The definition of adaptive and reliable thresholds relies on the existence of accurate predictive models of dam response. The main approaches available are a) physics-based models (FEM) and b) data-driven models (machine learning). The former often lacks accuracy due to the necessary simplifications to be applied and to the uncertainty in the material properties. By contrast, data-driven models based on monitoring data and machine learning algorithms offer better predictive ability and are useful for any dam typology and response variable. Nonetheless, they also have relevant limitations: they require good quality monitoring databases –often not available- and they can only predict the response in situations already recorded in the dam history, e.g., they are not applicable under extraordinary loads.
The main objective of the project is the development of a methodology for generating hybrid predictive models, combining physics-based (FEM) and data-driven models based on machine learning. The resulting approach aims at alleviating the limitations of both methodologies and taking advantage of their benefits.
In addition, a methodology will be developed for defining dynamic, adaptive and reliable warning thresholds based on the resulting hybrid models. It will be applied in a case study of a real arch dam in operation.
Similar approaches have been successfully applied in other fields of science and engineering, and the research team gathers unique expertise on dam safety, numerical modelling and machine learning. This is proven by recent and current research projects of the group, as well as by its publication record.
The scientific and technical impact and the knowledge transfer is ensured by the ongoing collaborations of the PI with the International Committee on Large Dams and the National counterpart, as well as with research groups at international level, e.g., Dr. Juan Mata (LNEC, Portugal, Dr. Alexandre Simon (EDF, France), Prof. Hariri-Ardebili (University of Colorado, USA), Dr. Richard Malm (KTH, Sweden). Likewise, relevant entities in the field showed their interest in the project with support letters: Endesa Generación (hydropower company), LNEC (Research institution responsible for dam safety in Portugal), and the Spanish Committee on Large Dams (SPANCOLD). |
