Application of time series models for forecasting demand and prices in an international transport company
Authorship
D.A.F.
Master's Degree in Statistical Techniques
D.A.F.
Master's Degree in Statistical Techniques
Defense date
02.09.2026 12:30
02.09.2026 12:30
Summary
This Master’s Thesis presents a comprehensive methodology for logistic demand forecasting at Trucksters, using advanced time-series modeling techniques. The central objective is to anticipate shipment volumes in a market defined by high volatility and a strong dependence on seasonal patterns. The research is developed under a comparative approach between the classic ARIMA methodology and Prophet generalized additive models, in order to determine which architecture most accurately captures the complexity of demand. To this end, the modeling strategy is articulated through a dual methodology: on the one hand, the utility of strategic customer segmentation through clusters is contrasted, and on the other, the performance of the models is evaluated on the original database versus a series cleared of outliers. This analysis demonstrates that the prior treatment of information and disaggregated modeling reduce statistical noise and mitigate the impact of anomalies. As a result, a more stable and robust predictive framework is obtained that optimizes Trucksters operational planning and commercial profitability.
This Master’s Thesis presents a comprehensive methodology for logistic demand forecasting at Trucksters, using advanced time-series modeling techniques. The central objective is to anticipate shipment volumes in a market defined by high volatility and a strong dependence on seasonal patterns. The research is developed under a comparative approach between the classic ARIMA methodology and Prophet generalized additive models, in order to determine which architecture most accurately captures the complexity of demand. To this end, the modeling strategy is articulated through a dual methodology: on the one hand, the utility of strategic customer segmentation through clusters is contrasted, and on the other, the performance of the models is evaluated on the original database versus a series cleared of outliers. This analysis demonstrates that the prior treatment of information and disaggregated modeling reduce statistical noise and mitigate the impact of anomalies. As a result, a more stable and robust predictive framework is obtained that optimizes Trucksters operational planning and commercial profitability.
Direction
AMEIJEIRAS ALONSO, JOSE (Tutorships)
Ginzo Villamayor, María José (Co-tutorships)
AMEIJEIRAS ALONSO, JOSE (Tutorships)
Ginzo Villamayor, María José (Co-tutorships)
Court
AMEIJEIRAS ALONSO, JOSE (Coordinator)
DE UÑA ALVAREZ, JACOBO (Chairman)
TARRIO SAAVEDRA, JAVIER (Secretary)
GONZALEZ DIAZ, JULIO (Member)
AMEIJEIRAS ALONSO, JOSE (Coordinator)
DE UÑA ALVAREZ, JACOBO (Chairman)
TARRIO SAAVEDRA, JAVIER (Secretary)
GONZALEZ DIAZ, JULIO (Member)
Application of statistical techniques for the reduction of computation time in photovoltaic system simulations
Authorship
A.B.A.
Master's Degree in Statistical Techniques
A.B.A.
Master's Degree in Statistical Techniques
Defense date
02.06.2026 10:15
02.06.2026 10:15
Summary
In this work, different statistical techniques are applied to reduce the time required to perform simulations of photovoltaic systems. An example of these techniques is the use of an unsupervised machine learning algorithm (clustering) to group input data in order to reduce its dimensionality. The objective is to reduce the time needed to carry out the simulations as much as possible, while keeping the error resulting from the simplifications under control.
In this work, different statistical techniques are applied to reduce the time required to perform simulations of photovoltaic systems. An example of these techniques is the use of an unsupervised machine learning algorithm (clustering) to group input data in order to reduce its dimensionality. The objective is to reduce the time needed to carry out the simulations as much as possible, while keeping the error resulting from the simplifications under control.
Direction
GONZALEZ RODRIGUEZ, BRAIS (Tutorships)
GONZALEZ RODRIGUEZ, BRAIS (Tutorships)
Court
AMEIJEIRAS ALONSO, JOSE (Coordinator)
Carpente Rodríguez, Maria luisa (Chairman)
SAAVEDRA NIEVES, PAULA (Secretary)
Mosquera Rodríguez, Manuel Alfredo (Member)
AMEIJEIRAS ALONSO, JOSE (Coordinator)
Carpente Rodríguez, Maria luisa (Chairman)
SAAVEDRA NIEVES, PAULA (Secretary)
Mosquera Rodríguez, Manuel Alfredo (Member)
Physics-Informed Neural Networks (PINNs) for Solving Optimal Control Problems in Industrial Applications
Authorship
M.D.L.H.R.
Master's Degree in Industrial Mathematics
M.D.L.H.R.
Master's Degree in Industrial Mathematics
Defense date
02.05.2026 11:00
02.05.2026 11:00
Summary
This work studies an optimal control problem applied to a nonlinear dynamical system, combining classical control approaches with neural network based methods. A method based on Pontryagin’s Minimum Principle is analyzed, which allows the computation of optimal trajectories with high accuracy through an open loop approach. However, its computational cost becomes high when multiple state evaluations are required. In addition, Model Predictive Control (MPC) methods are discussed, which introduce feedback at the expense of increased computational cost and algorithmic complexity. As the main contribution of this work, an alternative approach based on the Hamilton Jacobi Bellman (HJB) equation is proposed, in which the value function is approximated using a physics informed neural network (PINN). The novel idea consists of combining, within the loss functional, the residual of the HJB equation and the terminal conditions of the problem with data obtained from optimal trajectories computed via Pontryagin’s method, all within the framework of the Howard algorithm (policy iteration). This hybrid strategy significantly improves the approximation of the value function and, most importantly, of its gradient, which constitutes the key element for constructing the optimal feedback control law. Numerical results show that the network trained with this methodology is able to reproduce optimal trajectories with high accuracy, achieving low relative errors while substantially reducing computational times compared to Pontryagin and MPC based approaches. As a result, the HJB PINN framework emerges as a promising alternative for real time optimal control and as a foundation for future research aimed at generalizing the method to broader families of control problems.
This work studies an optimal control problem applied to a nonlinear dynamical system, combining classical control approaches with neural network based methods. A method based on Pontryagin’s Minimum Principle is analyzed, which allows the computation of optimal trajectories with high accuracy through an open loop approach. However, its computational cost becomes high when multiple state evaluations are required. In addition, Model Predictive Control (MPC) methods are discussed, which introduce feedback at the expense of increased computational cost and algorithmic complexity. As the main contribution of this work, an alternative approach based on the Hamilton Jacobi Bellman (HJB) equation is proposed, in which the value function is approximated using a physics informed neural network (PINN). The novel idea consists of combining, within the loss functional, the residual of the HJB equation and the terminal conditions of the problem with data obtained from optimal trajectories computed via Pontryagin’s method, all within the framework of the Howard algorithm (policy iteration). This hybrid strategy significantly improves the approximation of the value function and, most importantly, of its gradient, which constitutes the key element for constructing the optimal feedback control law. Numerical results show that the network trained with this methodology is able to reproduce optimal trajectories with high accuracy, achieving low relative errors while substantially reducing computational times compared to Pontryagin and MPC based approaches. As a result, the HJB PINN framework emerges as a promising alternative for real time optimal control and as a foundation for future research aimed at generalizing the method to broader families of control problems.
Direction
RODRIGUEZ GARCIA, JERONIMO (Tutorships)
RODRIGUEZ GARCIA, JERONIMO (Tutorships)
Court
VAZQUEZ CENDON, MARIA ELENA (Coordinator)
VIAÑO REY, JUAN MANUEL (Chairman)
Perales Perales, José Manuel (Secretary)
Lopez Bonilla, Luis Francisco (Member)
VAZQUEZ CENDON, MARIA ELENA (Coordinator)
VIAÑO REY, JUAN MANUEL (Chairman)
Perales Perales, José Manuel (Secretary)
Lopez Bonilla, Luis Francisco (Member)
Modeling, characterization, and performance optimization of Rydberg-atom-based radio-frequency quantum sensors
Authorship
C.L.A.
Master's Degree in Industrial Mathematics
C.L.A.
Master's Degree in Industrial Mathematics
Defense date
02.06.2026 10:30
02.06.2026 10:30
Summary
The development of quantum technologies has unlocked new possibilities in high-precision metrology, surpassing the limitations of classical techniques in terms of sensitivity, resolution, and disturbance of the measured field. This work studies and models a quantum sensor based on Rydberg atoms for the detection of radiofrequency electric fields. The operational principle of these sensors is based on the Electromagnetically Induced Transparency technique, through which the electric field amplitude is inferred from observable modifications in an absorption spectrum, specifically, the Autler-Townes splitting induced by the radiofrequency field. Numerical and analytical models are employed to describe the system dynamics under realistic experimental conditions. By integrating a noise model that includes the technical specifications of the detection hardware, the parameter space is analyzed to optimize sensor performance and identify the factors limiting its resolution, achieving a sensitivity of 1.6 mV per m per square root of Hz for the 9.707 GHz radiofrequency channel. This Master’s Thesis represents a departure from the conventional methodology found in Rydberg sensor literature, which tends to exhibit a critical disconnection between theoretical models and analyzed data. This work establishes an intrinsic dialogue between theory and experimentation, transforming the theoretical framework into an active decoding tool for automated control optimization. This approach makes it possible to extract metrological information that would remain hidden under conventional analysis, providing the sensor with new capabilities to be integrated in applications such as quantum-enhanced radars or remote satellite detectors. The studies presented not only consolidate the ability of Rydberg quantum sensors to operate with high precision and stability but also propose an optimization framework and a predictive control interface based on an analytical solution 30000 times faster than standard numerical methods. This facilitates their transition from the laboratory to real-world applications in telecommunications and defense.
The development of quantum technologies has unlocked new possibilities in high-precision metrology, surpassing the limitations of classical techniques in terms of sensitivity, resolution, and disturbance of the measured field. This work studies and models a quantum sensor based on Rydberg atoms for the detection of radiofrequency electric fields. The operational principle of these sensors is based on the Electromagnetically Induced Transparency technique, through which the electric field amplitude is inferred from observable modifications in an absorption spectrum, specifically, the Autler-Townes splitting induced by the radiofrequency field. Numerical and analytical models are employed to describe the system dynamics under realistic experimental conditions. By integrating a noise model that includes the technical specifications of the detection hardware, the parameter space is analyzed to optimize sensor performance and identify the factors limiting its resolution, achieving a sensitivity of 1.6 mV per m per square root of Hz for the 9.707 GHz radiofrequency channel. This Master’s Thesis represents a departure from the conventional methodology found in Rydberg sensor literature, which tends to exhibit a critical disconnection between theoretical models and analyzed data. This work establishes an intrinsic dialogue between theory and experimentation, transforming the theoretical framework into an active decoding tool for automated control optimization. This approach makes it possible to extract metrological information that would remain hidden under conventional analysis, providing the sensor with new capabilities to be integrated in applications such as quantum-enhanced radars or remote satellite detectors. The studies presented not only consolidate the ability of Rydberg quantum sensors to operate with high precision and stability but also propose an optimization framework and a predictive control interface based on an analytical solution 30000 times faster than standard numerical methods. This facilitates their transition from the laboratory to real-world applications in telecommunications and defense.
Direction
López Pouso, Óscar (Tutorships)
López Pouso, Óscar (Tutorships)
Court
VAZQUEZ CENDON, MARIA ELENA (Coordinator)
Durany Castrillo, José (Chairman)
Sanchez Villaseñor, Eduardo Jesús (Secretary)
Rodríguez Seijo, José Manuel (Member)
VAZQUEZ CENDON, MARIA ELENA (Coordinator)
Durany Castrillo, José (Chairman)
Sanchez Villaseñor, Eduardo Jesús (Secretary)
Rodríguez Seijo, José Manuel (Member)
Mathematical Optimization in the Field of Energy Transition
Authorship
P.L.L.
Master's Degree in Statistical Techniques
P.L.L.
Master's Degree in Statistical Techniques
Defense date
02.06.2026 09:30
02.06.2026 09:30
Summary
This Master's Thesis, conducted under the collaboration agreement between CITMAga and Repsol, focuses on the optimization of BIO feedstock procurement for the production of advanced biofuels and other low-carbon fuels, in response to the increasing requirements set by new European regulations within the energy transition framework. The core of this work is the modeling and integration of the C43 plant into the general optimization problem. The report details the transition from the mathematical formulation of constraints and optimization properties for the C43 biofuel plant elements toward their technical implementation in an industrial environment. To this end, a functional simplification of the global model is presented which, while maintaining logical rigor, allows for an in-depth analysis of the C43 plant without the overwhelming complexity of the full system. The development integrates theoretical aspects characteristic of optimization problems with practical data engineering tasks, covering source code development, database management, and result visualization. Based on the experience gained during the professional immersion throughout the internship period, the project evaluates the new functionalities added to the model and analyzes the technical limitations still present. Finally, alternatives and future lines of development are proposed to ensure the tool's operability and effectiveness within the current sustainability context.
This Master's Thesis, conducted under the collaboration agreement between CITMAga and Repsol, focuses on the optimization of BIO feedstock procurement for the production of advanced biofuels and other low-carbon fuels, in response to the increasing requirements set by new European regulations within the energy transition framework. The core of this work is the modeling and integration of the C43 plant into the general optimization problem. The report details the transition from the mathematical formulation of constraints and optimization properties for the C43 biofuel plant elements toward their technical implementation in an industrial environment. To this end, a functional simplification of the global model is presented which, while maintaining logical rigor, allows for an in-depth analysis of the C43 plant without the overwhelming complexity of the full system. The development integrates theoretical aspects characteristic of optimization problems with practical data engineering tasks, covering source code development, database management, and result visualization. Based on the experience gained during the professional immersion throughout the internship period, the project evaluates the new functionalities added to the model and analyzes the technical limitations still present. Finally, alternatives and future lines of development are proposed to ensure the tool's operability and effectiveness within the current sustainability context.
Direction
GONZALEZ DIAZ, JULIO (Tutorships)
GONZALEZ RODRIGUEZ, BRAIS (Co-tutorships)
GONZALEZ DIAZ, JULIO (Tutorships)
GONZALEZ RODRIGUEZ, BRAIS (Co-tutorships)
Court
AMEIJEIRAS ALONSO, JOSE (Coordinator)
Carpente Rodríguez, Maria luisa (Chairman)
SAAVEDRA NIEVES, PAULA (Secretary)
Mosquera Rodríguez, Manuel Alfredo (Member)
AMEIJEIRAS ALONSO, JOSE (Coordinator)
Carpente Rodríguez, Maria luisa (Chairman)
SAAVEDRA NIEVES, PAULA (Secretary)
Mosquera Rodríguez, Manuel Alfredo (Member)
Topological Deep Learning
Authorship
S.M.J.
Master's Degree in Mathematics
S.M.J.
Master's Degree in Mathematics
Defense date
02.09.2026 11:00
02.09.2026 11:00
Summary
This paper studies the theoretical framework of combinatorial complexes, a higher-order topological domain introduced as a generalization of graphs, simplicial complexes, cellular complexes, and hypergraphs. Cochain spaces are defined on these domains to serve as data support, along with operators that allow information to be transformed between cells of different dimensions. Based on this structure, neural networks on combinatorial complexes (CCNNs) are analyzed. These networks are constructed using elementary tensor operators and can be represented through tensor diagrams. A message-passing paradigm for combinatorial complexes is also presented, generalizing from those defined in the literature. Finally, it is shown that the calculations performed by a CCNN can be reduced to message-passing schemes on graphs. This work is based on the article: Mustafa Hajij et al., Topological Deep Learning: Going Beyond Graph Data, arXiv:2206.00606, 2023.
This paper studies the theoretical framework of combinatorial complexes, a higher-order topological domain introduced as a generalization of graphs, simplicial complexes, cellular complexes, and hypergraphs. Cochain spaces are defined on these domains to serve as data support, along with operators that allow information to be transformed between cells of different dimensions. Based on this structure, neural networks on combinatorial complexes (CCNNs) are analyzed. These networks are constructed using elementary tensor operators and can be represented through tensor diagrams. A message-passing paradigm for combinatorial complexes is also presented, generalizing from those defined in the literature. Finally, it is shown that the calculations performed by a CCNN can be reduced to message-passing schemes on graphs. This work is based on the article: Mustafa Hajij et al., Topological Deep Learning: Going Beyond Graph Data, arXiv:2206.00606, 2023.
Direction
Gómez Tato, Antonio M. (Tutorships)
Gómez Tato, Antonio M. (Tutorships)
Court
FERNANDEZ TOJO, FERNANDO ADRIAN (Coordinator)
LADRA GONZALEZ, MANUEL EULOGIO (Chairman)
FERNANDEZ TOJO, FERNANDO ADRIAN (Secretary)
GARCIA RIO, EDUARDO (Member)
FERNANDEZ TOJO, FERNANDO ADRIAN (Coordinator)
LADRA GONZALEZ, MANUEL EULOGIO (Chairman)
FERNANDEZ TOJO, FERNANDO ADRIAN (Secretary)
GARCIA RIO, EDUARDO (Member)
Conservative methods for wave and electromagnetism problems based on a dual complex with splines
Authorship
A.P.D.V.R.
Master's Degree in Industrial Mathematics
A.P.D.V.R.
Master's Degree in Industrial Mathematics
Defense date
02.02.2026 11:00
02.02.2026 11:00
Summary
Isogeometric methods constitute a subclass of Galerkin methods in which spline based functions are employed to discretize the functional spaces of interest. A notable advantage of this approach lies in the natural construction of discrete spaces that conform the de Rham sequence, a fundamental structural property required in many mathematical models. This requirement is particularly crucial in wave propagation problems, such as linear acoustics and Maxwell’s equations, where two de Rham sequences (commonly referred to as the primal and dual sequences) arise and must be consistently and accurately approximated. This document is devoted to the study of isogeometric methods for wave problems, with a particular focus on electromagnetic applications. The work begins with a comprehensive review of the existing literature, which establishes the theoretical background and contextualizes the contribution of this research. Building upon this foundation, the document subsequently develops new theoretical and computational results that advance the current state of the art.
Isogeometric methods constitute a subclass of Galerkin methods in which spline based functions are employed to discretize the functional spaces of interest. A notable advantage of this approach lies in the natural construction of discrete spaces that conform the de Rham sequence, a fundamental structural property required in many mathematical models. This requirement is particularly crucial in wave propagation problems, such as linear acoustics and Maxwell’s equations, where two de Rham sequences (commonly referred to as the primal and dual sequences) arise and must be consistently and accurately approximated. This document is devoted to the study of isogeometric methods for wave problems, with a particular focus on electromagnetic applications. The work begins with a comprehensive review of the existing literature, which establishes the theoretical background and contextualizes the contribution of this research. Building upon this foundation, the document subsequently develops new theoretical and computational results that advance the current state of the art.
Direction
VAZQUEZ HERNANDEZ, RAFAEL (Tutorships)
VAZQUEZ HERNANDEZ, RAFAEL (Tutorships)
Court
VAZQUEZ CENDON, MARIA ELENA (Coordinator)
ARREGUI ALVAREZ, IÑIGO (Chairman)
Porter , Jeff (Secretary)
FERNANDEZ FERNANDEZ, FRANCISCO JAVIER (Member)
VAZQUEZ CENDON, MARIA ELENA (Coordinator)
ARREGUI ALVAREZ, IÑIGO (Chairman)
Porter , Jeff (Secretary)
FERNANDEZ FERNANDEZ, FRANCISCO JAVIER (Member)
Tracking Marine Litter in Northwestern Spain: Assessing the Influence of Wind Using a Lagrangian Transport Model
Authorship
M.R.O.
Master's Degree in Industrial Mathematics
M.R.O.
Master's Degree in Industrial Mathematics
Defense date
02.02.2026 12:30
02.02.2026 12:30
Summary
Marine debris are responsible for major problems in our oceans, causing serious environmental degradation, detrimental health effects and economic losses in sectors related to the marine environment. In this work, we examine how emissions from the Ulla river at the most inner side of the estuary affect the transport, accumulation, and beaching of floating particles in the Ría de Arousa, an estuary on the northwest coast of the Iberian Peninsula, as a result of wind force. Using Lagrangian simulations of particle tracking under different wind drag coefficients (1%, 3% and 5%), we evaluate the spatial and seasonal patterns of particle concentration, residence time and deposition on the coast. Our results show that wind plays a crucial role in modulating particle behavior. Low wind-driven conditions favor greater near-shore accumulation and longer residence times, especially in the northern and inner regions of the estuary. As wind influence increases, particle dispersion intensifies, leading to higher overall accumulation of particles nearshore. Seasonal differences are also studied, with higher concentrations observed in the north during winter and in the south during summer.
Marine debris are responsible for major problems in our oceans, causing serious environmental degradation, detrimental health effects and economic losses in sectors related to the marine environment. In this work, we examine how emissions from the Ulla river at the most inner side of the estuary affect the transport, accumulation, and beaching of floating particles in the Ría de Arousa, an estuary on the northwest coast of the Iberian Peninsula, as a result of wind force. Using Lagrangian simulations of particle tracking under different wind drag coefficients (1%, 3% and 5%), we evaluate the spatial and seasonal patterns of particle concentration, residence time and deposition on the coast. Our results show that wind plays a crucial role in modulating particle behavior. Low wind-driven conditions favor greater near-shore accumulation and longer residence times, especially in the northern and inner regions of the estuary. As wind influence increases, particle dispersion intensifies, leading to higher overall accumulation of particles nearshore. Seasonal differences are also studied, with higher concentrations observed in the north during winter and in the south during summer.
Direction
VAZQUEZ CENDON, MARIA ELENA (Tutorships)
VAZQUEZ CENDON, MARIA ELENA (Tutorships)
Court
VAZQUEZ CENDON, MARIA ELENA (Coordinator)
ARREGUI ALVAREZ, IÑIGO (Chairman)
Varas Mérida, Fernando (Secretary)
FERNANDEZ FERNANDEZ, FRANCISCO JAVIER (Member)
VAZQUEZ CENDON, MARIA ELENA (Coordinator)
ARREGUI ALVAREZ, IÑIGO (Chairman)
Varas Mérida, Fernando (Secretary)
FERNANDEZ FERNANDEZ, FRANCISCO JAVIER (Member)
Model Selection and Conformal Prediction for Reliable Electricity Consumption Forecasting
Authorship
N.R.G.
Master's Degree in Industrial Mathematics
N.R.G.
Master's Degree in Industrial Mathematics
Defense date
02.05.2026 11:30
02.05.2026 11:30
Summary
This work proposes a deployable pipeline for reliable probabilistic forecasting of hourly electricity consumption time series, using standard exogenous variables such as temperature and calendar features, and aiming to generalize across heterogeneous consumption settings without retraining for each case. To handle non-stationarity, it adds an online model-selection layer in an “expert advice” setting, where each expert corresponds to a fixed hyperparameter configuration. It implements Follow the Leader with a sliding evaluation window and Hedge with exponential weights, and compares them using operational metrics such as MSE, NLL, and CRPS; overall, FTL with a 30 day window is typically the strongest performer among the implementable strategies. On top of the base probabilistic forecast (Gaussian predictive distributions), it adds conformal prediction to obtain more reliable prediction intervals. Calibration is performed by context groups, comparing a heuristic group-wise approach updated via a Robbins Monro recursion and a window-based asymmetric conformalization inspired by time-series conformal methods; empirically, conformalization mitigates the systematic undercoverage of nominal Gaussian bands and improves reliability across contexts. Overall, the experiments support that combining online adaptation (expert selection/weighting) with conformal calibration is a practical path toward robust and trustworthy forecasting at scale.
This work proposes a deployable pipeline for reliable probabilistic forecasting of hourly electricity consumption time series, using standard exogenous variables such as temperature and calendar features, and aiming to generalize across heterogeneous consumption settings without retraining for each case. To handle non-stationarity, it adds an online model-selection layer in an “expert advice” setting, where each expert corresponds to a fixed hyperparameter configuration. It implements Follow the Leader with a sliding evaluation window and Hedge with exponential weights, and compares them using operational metrics such as MSE, NLL, and CRPS; overall, FTL with a 30 day window is typically the strongest performer among the implementable strategies. On top of the base probabilistic forecast (Gaussian predictive distributions), it adds conformal prediction to obtain more reliable prediction intervals. Calibration is performed by context groups, comparing a heuristic group-wise approach updated via a Robbins Monro recursion and a window-based asymmetric conformalization inspired by time-series conformal methods; empirically, conformalization mitigates the systematic undercoverage of nominal Gaussian bands and improves reliability across contexts. Overall, the experiments support that combining online adaptation (expert selection/weighting) with conformal calibration is a practical path toward robust and trustworthy forecasting at scale.
Direction
Rapún Banzo, Mª Luisa (Tutorships)
Rapún Banzo, Mª Luisa (Tutorships)
Court
VAZQUEZ CENDON, MARIA ELENA (Coordinator)
VIAÑO REY, JUAN MANUEL (Chairman)
Perales Perales, José Manuel (Secretary)
Lopez Bonilla, Luis Francisco (Member)
VAZQUEZ CENDON, MARIA ELENA (Coordinator)
VIAÑO REY, JUAN MANUEL (Chairman)
Perales Perales, José Manuel (Secretary)
Lopez Bonilla, Luis Francisco (Member)
Hemodynamic Analysis of the Aortic Arch Using CFD Techniques
Authorship
R.R.L.
Master's Degree in Industrial Mathematics
R.R.L.
Master's Degree in Industrial Mathematics
Defense date
02.02.2026 12:00
02.02.2026 12:00
Summary
Cerebrovascular accidents (CVA), commonly known as stroke, constitute one of the leading causes of mortality worldwide. In the case of embolic strokes of undetermined source (ESUS), the embolization of atherosclerotic plaque fragments located in the aorta has recently been proposed as a potential cause, transported in a retrograde manner by diastolic retrograde flow. In this context, a computational model of the aorta has been developed in this Master’s Thesis, integrating computational fluid dynamics (CFD) and solid mechanics within a fluid structure interaction (FSI) framework, together with a Lagrangian approach for the intravascular tracking of particles with physical properties representative of human emboli. The analysis of blood flow dynamics in several idealized geometric configurations has confirmed the existence of diastolic retrograde flow capable of transporting such particles in a reverse direction and redirecting them towards the main branches of the aorta supplying the brain.
Cerebrovascular accidents (CVA), commonly known as stroke, constitute one of the leading causes of mortality worldwide. In the case of embolic strokes of undetermined source (ESUS), the embolization of atherosclerotic plaque fragments located in the aorta has recently been proposed as a potential cause, transported in a retrograde manner by diastolic retrograde flow. In this context, a computational model of the aorta has been developed in this Master’s Thesis, integrating computational fluid dynamics (CFD) and solid mechanics within a fluid structure interaction (FSI) framework, together with a Lagrangian approach for the intravascular tracking of particles with physical properties representative of human emboli. The analysis of blood flow dynamics in several idealized geometric configurations has confirmed the existence of diastolic retrograde flow capable of transporting such particles in a reverse direction and redirecting them towards the main branches of the aorta supplying the brain.
Direction
VAZQUEZ CENDON, MARIA ELENA (Tutorships)
VAZQUEZ CENDON, MARIA ELENA (Tutorships)
Court
VAZQUEZ CENDON, MARIA ELENA (Coordinator)
ARREGUI ALVAREZ, IÑIGO (Chairman)
Porter , Jeff (Secretary)
FERNANDEZ FERNANDEZ, FRANCISCO JAVIER (Member)
VAZQUEZ CENDON, MARIA ELENA (Coordinator)
ARREGUI ALVAREZ, IÑIGO (Chairman)
Porter , Jeff (Secretary)
FERNANDEZ FERNANDEZ, FRANCISCO JAVIER (Member)
Clusterization of establishments and Business Applications
Authorship
M.S.C.
Master's Degree in Statistical Techniques
M.S.C.
Master's Degree in Statistical Techniques
Defense date
02.09.2026 11:45
02.09.2026 11:45
Summary
This project presents an analytical solution for the segmentation of establishments in the HORECA channel for the company Hijos de Rivera S.A.U.. To this end, sales data, the product catalog, and a census collecting all establishments in Spain are employed. Following an exhaustive process of data cleaning and treatment, the CLARA algorithm was selected to properly manage high dimensionality and the presence of outliers. Finally, the HORECA channel was segmented into 17 easily interpretable clusters. The results translate into tangible business strategies: a mobility model to elevate the category of current clients and a recommendation system for the efficient acquisition of new establishments in key areas.
This project presents an analytical solution for the segmentation of establishments in the HORECA channel for the company Hijos de Rivera S.A.U.. To this end, sales data, the product catalog, and a census collecting all establishments in Spain are employed. Following an exhaustive process of data cleaning and treatment, the CLARA algorithm was selected to properly manage high dimensionality and the presence of outliers. Finally, the HORECA channel was segmented into 17 easily interpretable clusters. The results translate into tangible business strategies: a mobility model to elevate the category of current clients and a recommendation system for the efficient acquisition of new establishments in key areas.
Direction
AMEIJEIRAS ALONSO, JOSE (Tutorships)
AMEIJEIRAS ALONSO, JOSE (Tutorships)
Court
AMEIJEIRAS ALONSO, JOSE (Coordinator)
DE UÑA ALVAREZ, JACOBO (Chairman)
TARRIO SAAVEDRA, JAVIER (Secretary)
GONZALEZ DIAZ, JULIO (Member)
AMEIJEIRAS ALONSO, JOSE (Coordinator)
DE UÑA ALVAREZ, JACOBO (Chairman)
TARRIO SAAVEDRA, JAVIER (Secretary)
GONZALEZ DIAZ, JULIO (Member)