Pasar al contenido principal

CONTROL DE LA SEGURIDAD DE PRESAS CON UMBRALES DE AVISO DINAMICOS COMBINANDO MODELOS NUMERICOS Y APRENDIZAJE AUTOMATICO

PID2021-122661OB-I00

•
Nombre agencia financiadora Agencia Estatal de Investigación
Acrónimo agencia financiadora AEI
Programa Programa Estatal para Impulsar la Investigación Científico-Técnica y su Transferencia
Subprograma Subprograma Estatal de Generación de Conocimiento
Convocatoria Proyectos de I+D+i (Generación de Conocimiento y Retos Investigación)
Año convocatoria 2021
Plan Estatal 2021-2023 Plan Estatal de Investigación Científica y Técnica y de Innovación
Centro beneficiario CENTRE INTERNACIONAL DE METODES NUMERICS A L ENGINYERIA

Resultados relacionados

Resultados totales (Incluyendo duplicados): 9
Encontrada(s) 1 página(s)

RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publicación Artículo científico (JournalArticle)

Avances en Iber para la clasificación de balsas: proyecto ACROPOLIS, Advancements in Iber for the classification of off-stream reservoirs: ACROPOLIS project

  • Sanz-Ramos, Marcos
  • Bladé, Ernest
  • Silva-Cancino, Nathalia
  • Salazar, Fernando
[EN] In Spain there are around 70.000 reservoirs. According to current legislation, users and administrations are obligated to classify them based on the potential risk in the event of a rupture, if they have more than 100.000 m3 or the dam height exceeds 5 m. With the aim of creating a support system for their classification that facilitates and expedites this process, the ACROPOLIS project has been carried out. To this end, the numerical modelling tool Iber played an essential and cross-cutting role. On one hand, the use of Iber allowed for the development of a simplified method for pre-classifying reservoirs based on machine learning techniques. On the other hand, ad hoc methodologies were developed in Iber to streamline and automate processes, such as the automatic extraction of hydraulic and hazard information at vulnerable points, and the parallelization of the code for computation using high-performance computing techniques., [ES] Se estima que en España hay alrededor de 70.000 balsas. Según la legislación vigente, usuarios y administraciones tienen la obligación de clasificarlas en función del riesgo potencial frente una eventual rotura si tienen más de 100.000 m3 o el dique mide más de 5 m. Con el objetivo de construir un sistema de ayuda para su clasificación que facilite y agilice este proceso, se ha llevado a cabo el proyecto ACROPOLIS. A tal fin, la herramienta de modelización numérica Iber fue elemento esencial y de carácter transversal. Por un lado, el empleo de Iber permitió desarrollar un método simplificado para pre-clasificar balsas en base a técnicas de machine learning y, por otro lado, se desarrollaron metodologías ad hoc en Iber que agilizan y automatizan los procesos, como la extracción automática de información hidráulica y de peligrosidad en puntos vulnerables y la paralelización del código para el cálculo mediante técnicas de computación de alto rendimiento., Este trabajo fue parcialmente financiado por el Ministerio de Ciencia, Innovación y Universidades de España a través de los proyectos ACROPOLIS (RTC2019-007343-5), TRISTAN (RTI2018-094785-B-I00) y DOLMEN (PID2021-122661OB-I00), así como por el Ministerio de Economía y Competitividad de España, a través del “Programa Severo Ochoa para Centros de Excelencia en I+D” (CEX2018-000797-S), y por la Generalitat de Catalunya a través del Programa CERCA.
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Detection and spatial localization of dam anomalies using autoencoder residuals

  • Nogara, Caterina|||0000-0003-1493-566X
  • Salazar González, Fernando|||0000-0002-5566-3253
© 2026 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/)., Monitoring the structural health of dams is essential, as these structures become more susceptible to damage over time. Damage detection methods are often based on Finite Element Models (FEM) or Machine Learning (ML) regression approaches, which require comparison with measured data to identify anomalies. Alternatively, ML algorithms can be employed for classification, but this strategy requires both healthy and damaged data for training. In contrast, this study introduces an unsupervised methodology based on an autoencoder trained exclusively on healthy data, in which anomaly detection relies on the individual reconstruction residuals of each variable, compared against statistically defined thresholds derived from normal operating conditions. The performance of the method is evaluated against the traditional reconstruction error approach using synthetic displacement data generated from the FEM of a reference double-curvature arch dam. Eight damage scenarios are analyzed, considering two damage locations and different levels of stiffness reduction. The results show that the proposed technique is able to consistently identify damage-related anomalies, with good detection performance across precision and recall metrics. In addition, the analysis of individual residuals provides useful information for damage-effect localization by highlighting the variables most affected., This work was funded by the Grant PID2021-122661OB-I00 awarded by MCIN/AEI/ 10.13039/501100011033 and ‘‘ERDF A way of making Europe’’. The publication was also supported by the Generalitat de Catalunya through the CERCA Program., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

SOLDIER: SOLution for Dam behavior Interpretation and safety Evaluation with boosted Regression trees

  • Salazar González, Fernando|||0000-0002-5566-3253
  • Irazábal González, Joaquín
  • Conde Vázquez, André
Decision making in dam safety is fundamentally based on the comparison between the predictions of a behavior model and the records of the monitoring system. Traditionally, simple linear regression models have been used. Recently, models based on machine learning are being explored, which generally offer greater precision –therefore, greater capacity for detecting anomalies –, higher flexibility and versatility. We have developed an interactive application based on R-Shiny to generate models based on boosted regression trees, evaluate their accuracy and analyze the effect of predictor variables on the system response. This allows for identifying changes in dam behavior, detecting potential anomalies and better understanding the effect of the loads on the structure. The availability of the software will contribute to the penetration of machine learning techniques in the dam engineering sector and will open the door to its use in structural health monitoring for other civil infrastructures., This work is funded by the Grant PID2021-122661OB-I00 funded by MCIN/AEI/ 10.13039/501100011033 and “ERDF A way of making Europe”. The publication is also associated to the grants: TED2021-129969B-C33, funded by MCIN/AEI/ 10.13039/ 501100011033 and the “European Union NextGenerationEU/PRTR”; CEX2018-000797-S funded by MCIN/AEI /10.13039/501100011033, and the Generalitat de Catalunya through the CERCA Program ., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

ACROPOLIS: a graphical user interface for classification of risk for off-stream reservoirs using machine learning

  • Silva Cancino, Nathalia
  • Salazar González, Fernando|||0000-0002-5566-3253
  • Bladé i Castellet, Ernest|||0000-0003-1770-3960
Potential risk identification due to off-stream reservoir failure typically requires the use of two-dimensional hydraulic models, which demands considerable effort in terms of expertise, time and financial resources. Unfortunately, not all reservoir owners have access to these resources, and the process of assessing hazard classifications for administrations can be burdensome. ACROPOLIS was developed to address this challenge, employing a Machine Learning model to provide a preliminary risk classification according to Spanish regulations for off-stream reservoirs without the necessity of building a hydraulic model. This approach has been integrated into a user-friendly interface, simplifying the process for users., This work was partially funded by the Spanish Ministry of Science, Innovation and Universities through the Projects ACROPOLIS (RTC2019-007343-5) and DOLMEN (PID2021-122661OB-I00), as well as by the Spanish Ministry of Economy and Competitiveness, through the “Severo Ochoa Programme for Centres of Excellence in R & D” (CEX2018-000797-S)., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Interpretation of a machine learning model for short-term high streamflow prediction

  • López Chacón, Sergio Ricardo
  • Salazar González, Fernando|||0000-0002-5566-3253
  • Bladé i Castellet, Ernest|||0000-0003-1770-3960
Machine learning models are increasingly used for streamflow prediction due to their promising performance. However, their data-driven nature makes interpretation challenging. This study explores the interpretability of a Random Forest model trained on high streamflow events from a hydrological perspective, comparing methods for assessing feature influence. The results show that the mean decrease accuracy, mean decrease impurity, Shapley additive explanations, and Tornado methods identify similar key features, though Tornado presents the most notable discrepancies. Despite the model being trained with events of considerable temporal variability, the last observed streamflow is the most relevant feature accounting for over 20% of importance. Moreover, the results suggest that the model identifies a catchment region with a runoff that significantly affects the outlet flow. Accumulated local effects and partial dependence plots may represent first infiltration losses and soil saturation before precipitation sharply impacts streamflow. However, only accumulated local effects depict the influence of the scarce highest accumulated precipitation on the streamflow. Shapley additive explanations are simpler to apply than the local interpretable model-agnostic explanations, which require a tuning process, though both offer similar insights. They show that short-period accumulated precipitation is crucial during the steep rising limb of the hydrograph, reaching 72% of importance on average among the top features. As the peak approaches, previous streamflow values become the most influential feature, continuing into the falling limb. When the hydrograph goes down, the model confers a moderate influence on the accumulated precipitation of several hours back of distant regions, suggesting that the runoff from these areas is arriving. Machine learning models may interpret the catchment system reasonably and provide useful insights about hydrological characteristics., This work is funded by the Grant PID2021-122661OB-I00 funded by MCIN/AEI/10.13039/501100011033 and “ERDF A way of making Europe”. The publication is also associated to the grants: TED2021-129969B-C33, funded by MCIN/AEI/10.13039/501100011033 and the “European Union NextGenerationEU/PRTR”; CEX2018-000797-S funded by MCIN/AEI/10.13039/501100011033, and the Generalitat de Catalunya through the CERCA Program., Peer Reviewed
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

ACROPOLIS: a graphical user interface for classification of risk for off-stream reservoirs using machine learning

  • Silva Cancino, Nathalia
  • Salazar González, Fernando
  • Bladé i Castellet, Ernest
Potential risk identification due to off-stream reservoir failure typically requires the use of two-dimensional hydraulic models, which demands considerable effort in terms of expertise, time and financial resources. Unfortunately, not all reservoir owners have access to these resources, and the process of assessing hazard classifications for administrations can be burdensome. ACROPOLIS was developed to address this challenge, employing a Machine Learning model to provide a preliminary risk classification according to Spanish regulations for off-stream reservoirs without the necessity of building a hydraulic model. This approach has been integrated into a user-friendly interface, simplifying the process for users., This work was partially funded by the Spanish Ministry of Science, Innovation and Universities through the Projects ACROPOLIS (RTC2019-007343-5) and DOLMEN (PID2021-122661OB-I00), as well as by the Spanish Ministry of Economy and Competitiveness, through the “Severo Ochoa Programme for Centres of Excellence in R & D” (CEX2018-000797-S)., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

SOLDIER: SOLution for Dam behavior Interpretation and safety Evaluation with boosted Regression trees

  • Salazar González, Fernando
  • Irazábal González, Joaquín
  • Conde Vázquez, André
Decision making in dam safety is fundamentally based on the comparison between the predictions of a behavior model and the records of the monitoring system. Traditionally, simple linear regression models have been used. Recently, models based on machine learning are being explored, which generally offer greater precision –therefore, greater capacity for detecting anomalies –, higher flexibility and versatility. We have developed an interactive application based on R-Shiny to generate models based on boosted regression trees, evaluate their accuracy and analyze the effect of predictor variables on the system response. This allows for identifying changes in dam behavior, detecting potential anomalies and better understanding the effect of the loads on the structure. The availability of the software will contribute to the penetration of machine learning techniques in the dam engineering sector and will open the door to its use in structural health monitoring for other civil infrastructures., This work is funded by the Grant PID2021-122661OB-I00 funded by MCIN/AEI/ 10.13039/501100011033 and “ERDF A way of making Europe”. The publication is also associated to the grants: TED2021-129969B-C33, funded by MCIN/AEI/ 10.13039/ 501100011033 and the “European Union NextGenerationEU/PRTR”; CEX2018-000797-S funded by MCIN/AEI /10.13039/501100011033, and the Generalitat de Catalunya through the CERCA Program ., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Interpretation of a machine learning model for short-term high streamflow prediction

  • López Chacón, Sergio Ricardo
  • Salazar González, Fernando
  • Bladé i Castellet, Ernest
Machine learning models are increasingly used for streamflow prediction due to their promising performance. However, their data-driven nature makes interpretation challenging. This study explores the interpretability of a Random Forest model trained on high streamflow events from a hydrological perspective, comparing methods for assessing feature influence. The results show that the mean decrease accuracy, mean decrease impurity, Shapley additive explanations, and Tornado methods identify similar key features, though Tornado presents the most notable discrepancies. Despite the model being trained with events of considerable temporal variability, the last observed streamflow is the most relevant feature accounting for over 20% of importance. Moreover, the results suggest that the model identifies a catchment region with a runoff that significantly affects the outlet flow. Accumulated local effects and partial dependence plots may represent first infiltration losses and soil saturation before precipitation sharply impacts streamflow. However, only accumulated local effects depict the influence of the scarce highest accumulated precipitation on the streamflow. Shapley additive explanations are simpler to apply than the local interpretable model-agnostic explanations, which require a tuning process, though both offer similar insights. They show that short-period accumulated precipitation is crucial during the steep rising limb of the hydrograph, reaching 72% of importance on average among the top features. As the peak approaches, previous streamflow values become the most influential feature, continuing into the falling limb. When the hydrograph goes down, the model confers a moderate influence on the accumulated precipitation of several hours back of distant regions, suggesting that the runoff from these areas is arriving. Machine learning models may interpret the catchment system reasonably and provide useful insights about hydrological characteristics., This work is funded by the Grant PID2021-122661OB-I00 funded by MCIN/AEI/10.13039/501100011033 and “ERDF A way of making Europe”. The publication is also associated to the grants: TED2021-129969B-C33, funded by MCIN/AEI/10.13039/501100011033 and the “European Union NextGenerationEU/PRTR”; CEX2018-000797-S funded by MCIN/AEI/10.13039/501100011033, and the Generalitat de Catalunya through the CERCA Program., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Detection and spatial localization of dam anomalies using autoencoder residuals

  • Nogara, Caterina
  • Salazar González, Fernando
© 2026 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/)., Monitoring the structural health of dams is essential, as these structures become more susceptible to damage over time. Damage detection methods are often based on Finite Element Models (FEM) or Machine Learning (ML) regression approaches, which require comparison with measured data to identify anomalies. Alternatively, ML algorithms can be employed for classification, but this strategy requires both healthy and damaged data for training. In contrast, this study introduces an unsupervised methodology based on an autoencoder trained exclusively on healthy data, in which anomaly detection relies on the individual reconstruction residuals of each variable, compared against statistically defined thresholds derived from normal operating conditions. The performance of the method is evaluated against the traditional reconstruction error approach using synthetic displacement data generated from the FEM of a reference double-curvature arch dam. Eight damage scenarios are analyzed, considering two damage locations and different levels of stiffness reduction. The results show that the proposed technique is able to consistently identify damage-related anomalies, with good detection performance across precision and recall metrics. In addition, the analysis of individual residuals provides useful information for damage-effect localization by highlighting the variables most affected., This work was funded by the Grant PID2021-122661OB-I00 awarded by MCIN/AEI/ 10.13039/501100011033 and ‘‘ERDF A way of making Europe’’. The publication was also supported by the Generalitat de Catalunya through the CERCA Program., Peer Reviewed, Postprint (published version)