Short-Term Scientific Mission

Diagnosis of Pathologies in Dams by Establishing Symptomatic Patterns Defined from Numerical Modeling and Machine Learning

Main theme: Algorithm Selection and Configuration
Grantee: Andre Conde Vazquez, Universidad Politécnica de Madrid, Madrid, Spain
Host: Carola Doerr, Sorbonne Université, Paris, France

Start date: 2025-04-10
End date: 2025-06-25
Awarded: 2025-03-07
Report approved: 2025-08-19

Description

Detection of behavioral anomalies in dams using machine learning algorithms has been investigated and even applied in certain real-world scenarios. However, once an anomaly is detected, considerable time and resources are required by dam experts to analyze it and identify its root cause. This project proposes the integration of artificial intelligence and numerical modeling to explore the feasibility of automatically diagnosing dam pathologies from monitoring data. Benefiting from black box algorithms and optimization techniques, the project aims to improve machine learning approaches to both detect anomalies and diagnose underlying causes. The ultimate goal is not only to identify anomalous behaviors, but also to determine the most likely pathology responsible for them.

Achievements

This STSM investigated the fusion of artificial intelligence and numerical modeling to automate the diagnosis of dam pathologies using monitoring data. A computational model was developed to efficiently simulate dam displacements within realistic time frames. Various machine learning algorithms and sampling methods were evaluated to detect cracks, with the most effective strategies identified. Notable advancements were achieved in optimizing the models, enhancing the accuracy and practicality of the proposed methodologies. The research highlights the potential of combining AI with simulation techniques to improve dam safety monitoring and supports the development of automated, data-driven solutions for structural health assessment in hydraulic infrastructures.