Machine Learning and models in Hydro-Environmental Engineering
Machine Learning in Civil Engineering
Principal Investigator
Fernando Salazar
Overview
Research
Staff
Projects
Publications
This group develops ML-based tools to solve complex engineering problems in hydraulic, geomechanical, and environmental fields, combining sensor data, numerical modelling, and practical software solutions for predictive analysis and decision support.
The Machine Learning in Civil Engineering Research Group at CIMNE focuses on solving complex engineering problems by integrating machine learning (ML) techniques with data from sensors, numerical simulations, and physical modelling.
The group has extensive experience in applying ML methods to hydraulic infrastructures, including dams, spillways, and water supply systems, with a strong track record in structural health monitoring, anomaly detection, and predictive maintenance.
Beyond hydraulics, the group also explores applications in geomechanics, environmental monitoring, and industrial processes. Research activities span the full ML breath, from data preprocessing and algorithm development to uncertainty quantification and interpretability.
Techniques include ensemble models, deep learning, and hybrid strategies combining ML with numerical models such as CFD and DEM. The group maintains a strong practical orientation, developing custom software solutions with user interfaces for real-world deployment, and expanding into areas such as water quality prediction and wastewater disinfection.
Research areas
Research activities involving Machine Learning techniques
Machine learning for dam behavior prediction
Development of methodologies and software for analysis of dam monitoring data, including generation of ML predictive models and their interpretation, with the final objective of supporting decision making in dam safety. Related to: DOLMEN Project.
Software for dam safety assessment through ML: screenshots of SOLDIER application. GitHub repo.
Advanced machine learning for anomaly detection and localization
We explore the possibilities of Deep Learning and other advanced ML algorithms such as Autoencoders to curate monitoring data, to detect anomalies and to localize potential structural damages.
Autoencoder structure (left) and detection of anomalous monitoring data (right). Source: https://doi.org/10.1007/s13349-025-00910-4.
New computational tools for reliability-based dam safety assessment
Use of ML models to support FEM analysis to predict dam response including uncertainty and risk analysis. Related to: TRISTAN Project
Anomaly detection in dams: example of monitoring network (left) and numerical model to simulate anomalous events (right). Source: https://doi.org/10.3390/w13172387
Analysis of hydraulic structures
Analysis of the hydraulic performance of dam spillways and bottom outlets combining numerical methods (PFEM, Free-Surface) and ML techniques.
Spillway hydraulic performance: example of geometry (left) and relationship between observed and predicted values from ML models of discharge capacity. Source: https://doi.org/10.3390/w11030544
Smart optimization of industrial processes
Support and optimization of rotational metal deformation design processes. Use of FEM-based Digital Twin framework combined with ML classification techniques. Related to: OPTIPRO Project
Metal-forming processes analysis: industrial equipment (left) and GUI for process parametrization (right)
Water quality and water treatment techniques
Application of ML models for the prediction of water quality status in water bodies and assessment of advanced water pollutant removal treatments. Related to: DIGIT4WATER Projecte
Prediction of advanced wastewater treatments with ML techniques. More info: https://doi.org/10.1016/j.jenvman.2024.123537
Air quality forecasting
ML-based prediction of Tropospheric O3 concentration in the PIKSEL platform.
Identification of hazards due to dam failure with ML surrogates
Automatic estimation of potential damages in case of failure of off-stream reservoirs. Related to: ACROPOLIS Project.
Screenshots of the ACROPOLIS software. Available in GitHub.
Calibration of numerical models with ML
Calibration of DEM parameters of clay behavior: numerical model to simulate clay behavior tests (left), and calibration analysis through ML (center and right). Source: https://doi.org/10.1007/s40571-022-00550-1
Research activities involving numerical methods
Thermo-mechanical behavior of concrete dams
Simulation of concrete dam behavior during construction and operation stages integrating high-detailed thermo-mechanical loads. Related to: ACOMBO Project; Software application ‘DamApplication’ (integrated in Kratos framework).
Concrete dam modelling: construction stage simulation (left), displacements field (center) and stresses field (right). Source: https://doi.org/10.1007/s11831-020-09439-9
Design of wedge-shaped block spillways
CFD simulation through Eulerian FEM modelling and block stability simulation through DEM modelling. Related to: PABLO Project.
Wedge shaped block spillways simulation: hydraulic analysis (left), block design (center) and block stability analysis (right)
Industrial design of dam fuse gates
Fluid-solid interaction simulations through PFEM+DEM modelling to calculate the following processes: discharge flow for different gate positions, gate falling velocity and gate-wall impact force. Related to: COFRE Project.
Fuse gates simulation: geometry design (left), fluid-solid interaction 2D simulation (center) and 3D simulation (right)
CFD analysis of hydraulic structures: highly convergent spillways, stilling basin and drainage systems modelling
Simulation of complex 3D hydraulic phenomena, through FEM and PFEM models, such as free surface position of hydraulic jump, pressure and velocity fields and identification of zones with erosion risk.
CFD analysis of hydraulic structures: highly convergent spillways (left) and stilling basin simulation (right)
Analysis of railway ballast behavior with the Discrete Element Method (DEM)
Simulation of railway infrastructures against climate change actions and evaluation of railway ballast response through DEM model. Related to: RESILTRACK Project.
Railway infrastructure simulation: ballast behavior simulation (left), calibration analysis (center) and railway infrastructure simulation (right)
Numerical modelling of Water Distribution Networks (WDN)
Development of numerical models for leakage simulation through advanced pressure-driven solvers. Related to: SMILER Project.
Methodological scheme of advanced numerical utility to simulate massive cases of leakage scenarios in WDNs
Hybrid approaches
Flexible and accurate prediction of dam behavior
Smart approaches to increase flexibility and accuracy of dam behavior prediction combining outputs from numerical methods and ML models using monitoring data. Related to: project DOLMEN.
Prediction of arch dam displacement with hybrid approaches.
Short-term streamflow prediction with hybrid models
Methodologies for combining outputs from 2D numerical models (based on IBER software) with ML-based models considering rainfall and past streamflow values to obtain accurate and fast 3-hour ahead predictions of streamflow in flood events.
Observed hydrograph versus predictions with hybrid approaches. Source: https://doi.org/10.1080/02626667.2024.2426720
Ongoing projects
Finished projects
Search
Related news
CIMNE experts use Machine Learning to streamline reservoir risk assessment
Researchers at CIMNE's Machine Learning in Civil Engineering Research Group and the Flumen Institute have developed a new machine-learning based tool to classify off-stream reservoirs based on their potential risk of rupture. The work, carried out as part of the...
CIMNE’s new machine learning-based software improves dam structural safety
Researchers from the Machine Learning in Civil Engineering group at CIMNE have developed a new machine-learning based software to predict structural behaviour of dams, allowing for enhanced decision-making and minimizing safety risks of these critical infrastructures....