Available MSc Thesis topics

Modeling, control and optimization of Airborne Wind Energy Systems

We have several topics available for MSc theses on modeling, control and optimization of airborne wind energy systems. Please contact Lorenzo Fagiano for details.

Completed MSc and PhD Theses

67 entries « ‹ 1 of 4 › »

LEONESIO, MARCO

Physics-enhanced machine learning methods for industrial process modeling and optimization PhD Thesis

2025.

Abstract | Links | BibTeX

@phdthesis{alma99693010008776,

title = {Physics-enhanced machine learning methods for industrial process modeling and optimization},

author = {MARCO LEONESIO},

url = {https://hdl.handle.net/10589/237538},

year  = {2025},

date = {2025-05-05},

urldate = {2025-05-05},

abstract = {The recent and significant developments in the field of digital technologies and Artificial Intelligence (AI) suggest their pervasive application even in the sector of capital goods and production systems. In particular, enablers such as smart sensors with local intelligence, machine networking, data mining techniques, and machine learning support the development of "Intelligent Machines". These will provide services in terms of production process monitoring, parameter optimization, and collaborative interaction with the operator, making them more autonomous, flexible, and efficient. This perspective is part of a well-defined National and European strategy that sees AI as one of the main tools for implementing the so-called Industry 4.0 Transition of production systems (PNRR). On the other side, according to the newer paradigm of Industry 5.0, Intelligent Machines are expected to be not only sustainable and resilient, but also "human-centric". Regarding this latter characteristic, we think there is ample room for developing automation approaches that involve sharing knowledge and control capabilities between the operator (more or less experienced) and the Intelligent Machine. This sharing is facilitated through a series of enabling technologies related to the world of AI. On these premises, this thesis is focused on modeling and optimization approaches for industrial processes that try to combine the generality and extrapolative capability of first-principle models (including the domain knowledge of experts) with the adaptive capabilities of data-driven methods, typical of machine learning. This would allow for achieving acceptable levels of accuracy even with limited data availability. These approaches can be framed in the paradigm of physics-informed, or physics-enhanced AI. Investigating Random Forest as a promising machine learning method suited to generate surrogate models in the case of small datasets, the problem of the global optimization of an objective function represented by this kind of model came to our attention. In particular, an original method to obtain an approximate global minimum at low computational complexity has been developed, resorting to the inherent structure of a Random Forest, which is traceable to a non-parametric model that partitions the feature space in convex orthotopes by applying binary splits on training data. Our approximate method shows optimality performances that are comparable to other exact approaches based on the solution of a Mixed Integer Linear Program, which entails a combinatorial complexity and cannot be applied to a large Random Forest operating in a high-dimensional feature space (curse of dimensionality). Then, seeking a way to increase the accuracy of the model in predicting the production quality class, a novel physics-informed learning approach for this problem is proposed. The approach relies on a hierarchical semi-supervised classification, where the training data, classified on the basis of the quality intervals of interest, are divided in a certain number of sub-clusters with respect to the process input parameters (primary features) and enhanced with the classification prediction provided by a physics-based model (apriori knowledge injection). To evaluate the effectiveness of the above-mentioned methodological achievements in the context of a real manufacturing setting, the centerless grinding production process has been considered. In the absence of proper experimental data, a high-fidelity model has been developed to generate a synthetic dataset, which is augmented with the predictions of a low-fidelity model representing the apriori physics-based knowledge about the process. The resulting dataset has been used both to grow a Random Forest and optimize its output, as well as to test the performance of the proposed Semi-supervised physics-informed classifier. Other state-of-the-art approaches to generating gray-box models have been evaluated, mostly based on Feed Forward Neural Networks. The results show the effectiveness of the proposed random forest optimization approach and quality classifier, especially dealing with a high-dimensional variable space. On the other side, the overall absolute performance of the hybrid models, in comparison with the pure data-driven counterpart, suffers from the smallness of the considered dataset with respect to the target behavioral complexity: the first principle predictions are not accurate enough to compensate for the lack of data.},

keywords = {},

pubstate = {published},

tppubtype = {phdthesis}

}

The recent and significant developments in the field of digital technologies and Artificial Intelligence (AI) suggest their pervasive application even in the sector of capital goods and production systems. In particular, enablers such as smart sensors with local intelligence, machine networking, data mining techniques, and machine learning support the development of "Intelligent Machines". These will provide services in terms of production process monitoring, parameter optimization, and collaborative interaction with the operator, making them more autonomous, flexible, and efficient. This perspective is part of a well-defined National and European strategy that sees AI as one of the main tools for implementing the so-called Industry 4.0 Transition of production systems (PNRR). On the other side, according to the newer paradigm of Industry 5.0, Intelligent Machines are expected to be not only sustainable and resilient, but also "human-centric". Regarding this latter characteristic, we think there is ample room for developing automation approaches that involve sharing knowledge and control capabilities between the operator (more or less experienced) and the Intelligent Machine. This sharing is facilitated through a series of enabling technologies related to the world of AI. On these premises, this thesis is focused on modeling and optimization approaches for industrial processes that try to combine the generality and extrapolative capability of first-principle models (including the domain knowledge of experts) with the adaptive capabilities of data-driven methods, typical of machine learning. This would allow for achieving acceptable levels of accuracy even with limited data availability. These approaches can be framed in the paradigm of physics-informed, or physics-enhanced AI. Investigating Random Forest as a promising machine learning method suited to generate surrogate models in the case of small datasets, the problem of the global optimization of an objective function represented by this kind of model came to our attention. In particular, an original method to obtain an approximate global minimum at low computational complexity has been developed, resorting to the inherent structure of a Random Forest, which is traceable to a non-parametric model that partitions the feature space in convex orthotopes by applying binary splits on training data. Our approximate method shows optimality performances that are comparable to other exact approaches based on the solution of a Mixed Integer Linear Program, which entails a combinatorial complexity and cannot be applied to a large Random Forest operating in a high-dimensional feature space (curse of dimensionality). Then, seeking a way to increase the accuracy of the model in predicting the production quality class, a novel physics-informed learning approach for this problem is proposed. The approach relies on a hierarchical semi-supervised classification, where the training data, classified on the basis of the quality intervals of interest, are divided in a certain number of sub-clusters with respect to the process input parameters (primary features) and enhanced with the classification prediction provided by a physics-based model (apriori knowledge injection). To evaluate the effectiveness of the above-mentioned methodological achievements in the context of a real manufacturing setting, the centerless grinding production process has been considered. In the absence of proper experimental data, a high-fidelity model has been developed to generate a synthetic dataset, which is augmented with the predictions of a low-fidelity model representing the apriori physics-based knowledge about the process. The resulting dataset has been used both to grow a Random Forest and optimize its output, as well as to test the performance of the proposed Semi-supervised physics-informed classifier. Other state-of-the-art approaches to generating gray-box models have been evaluated, mostly based on Feed Forward Neural Networks. The results show the effectiveness of the proposed random forest optimization approach and quality classifier, especially dealing with a high-dimensional variable space. On the other side, the overall absolute performance of the hybrid models, in comparison with the pure data-driven counterpart, suffers from the smallness of the considered dataset with respect to the target behavioral complexity: the first principle predictions are not accurate enough to compensate for the lack of data.

ZAGATI, ALEX

Estimation of center of gravity and inertia tensor of cars starting from their technical data and images Masters Thesis

2025.

BibTeX

POLISINI, LEONARDO

Multi-objective flight path optimization for a Fly-Gen Airborne Wind Energy via harmonic balance Masters Thesis

2025.

BibTeX

Petulicchio, Lorenzo

A distributed framework for autonomous graph-based mapping for a multi agent collaborative system Masters Thesis

2025.

BibTeX

Rositani, Marco

Physics-informed learning and data-driven control of an Airborne Wind Energy system Masters Thesis

2025.

BibTeX

Løwenstein, Kristoffer Fink

Physics-informed online learning of gray-box models by Moving Horizon Estimation and efficient numerical methods PhD Thesis

2025.

Abstract | Links | BibTeX

@phdthesis{alma99688692708776,

title = {Physics-informed online learning of gray-box models by Moving Horizon Estimation and efficient numerical methods},

author = {Kristoffer Fink Løwenstein},

url = {https://hdl.handle.net/10589/233832},

year  = {2025},

date = {2025-03-03},

urldate = {2025-03-03},

abstract = {Advanced human-engineered systems will inevitably play a more dominant role in the future and are expected to operate with increased autonomy benefiting society as a whole. Major challenges lie ahead to ensure that such complex systems operate in a safe manner while satisfying strict performance requirements. A compelling idea is to use optimization-based methodologies such as Moving Horizon Estimation (MHE) and Model Predictive Control (MPC) for autonomous decision-making as they allow incorporating complex objectives and critical measures through a cost function and constraints on system states and inputs. A fundamental requirement for MHE and MPC is a reliable, but computationally light, model enabling online evaluation. Acquiring such a model and, especially, adapting it to temporal variations of the underlying system dynamics, surrounding environment, or part-to-part variations found in any system is a challenging endeavor. This issue is addressed by introducing parametric gray-box models: Relying on physics-based modeling facilitates the integration of function approximators such as Feed Forward Neural Networks (FFNN) or Recurrent Neural Networks (RNN) of rather limited size as the data-driven component in the gray-box models, and also benefits from increased model interpretability compared to pure black-box models. By deploying these gray-box models in an MHE scheme, a novel MHE framework for physics-informed learning is introduced. The learning is made safe through constraints consistent with physical laws seamlessly integrated into the optimization problem. The proposed MHE scheme is inherently suitable for online application and provides concurrent state estimation and model adaptation with uncertainty quantification. Thus, the proposed MHE framework can support any advanced decision-making process that relies on accurate state and parameter estimates, being autonomous or human-assisted, extending beyond just MPC. Furthermore, an offline training algorithm capitalizing on the proposed MHE scheme tailored for gray-box models is presented, which provides a viable alternative to classical training algorithms for such a class of models. The utility of the methodology is demonstrated through two numerical experiments, showing promising results for estimation, prediction, closed-loop Nonlinear MPC (NMPC), and online model adaptation. For real-time implementation of MHE and MPC, the underlying optimization algorithms are of paramount importance. In linear MHE and MPC the problem boils down to a Quadratic Program (QP) while Sequential Quadratic Programming (SQP), a cherished solution procedure for Nonlinear MHE (NMHE) and NMPC, is based on solving a series of related QPs. To meet this demand, QPALM-OCP is introduced: QPALM-OCP is a Proximal Augmented Lagrangian Method (P-ALM) tailored for the multi-stage structure arising in MHE and MPC. The algorithm relies on a semi-smooth Newton method to solve the inner ALM problem allowing multiple active set changes between iterates and benefits from warm-starting. Due to specialized low-rank matrix modifications, the iterates remain cheap while providing fast execution times. Contrary to conventional Quadratic Programming solvers (QP solvers), QPALM-OCP comes with guarantees of R-linear convergence to a stationary point of nonconvex QPs, making it an ideal candidate for SQP methods, especially when using exact second-order information. A prototype implementation, despite being non-optimized, outperforms state-of-the-art general-purpose QP solvers and is slightly faster than state-of-the-art Optimal Control Problem (OCP)-specific solvers in numerical experiments, demonstrating great potential for further development.},

keywords = {},

pubstate = {published},

tppubtype = {phdthesis}

}

Advanced human-engineered systems will inevitably play a more dominant role in the future and are expected to operate with increased autonomy benefiting society as a whole. Major challenges lie ahead to ensure that such complex systems operate in a safe manner while satisfying strict performance requirements. A compelling idea is to use optimization-based methodologies such as Moving Horizon Estimation (MHE) and Model Predictive Control (MPC) for autonomous decision-making as they allow incorporating complex objectives and critical measures through a cost function and constraints on system states and inputs. A fundamental requirement for MHE and MPC is a reliable, but computationally light, model enabling online evaluation. Acquiring such a model and, especially, adapting it to temporal variations of the underlying system dynamics, surrounding environment, or part-to-part variations found in any system is a challenging endeavor. This issue is addressed by introducing parametric gray-box models: Relying on physics-based modeling facilitates the integration of function approximators such as Feed Forward Neural Networks (FFNN) or Recurrent Neural Networks (RNN) of rather limited size as the data-driven component in the gray-box models, and also benefits from increased model interpretability compared to pure black-box models. By deploying these gray-box models in an MHE scheme, a novel MHE framework for physics-informed learning is introduced. The learning is made safe through constraints consistent with physical laws seamlessly integrated into the optimization problem. The proposed MHE scheme is inherently suitable for online application and provides concurrent state estimation and model adaptation with uncertainty quantification. Thus, the proposed MHE framework can support any advanced decision-making process that relies on accurate state and parameter estimates, being autonomous or human-assisted, extending beyond just MPC. Furthermore, an offline training algorithm capitalizing on the proposed MHE scheme tailored for gray-box models is presented, which provides a viable alternative to classical training algorithms for such a class of models. The utility of the methodology is demonstrated through two numerical experiments, showing promising results for estimation, prediction, closed-loop Nonlinear MPC (NMPC), and online model adaptation. For real-time implementation of MHE and MPC, the underlying optimization algorithms are of paramount importance. In linear MHE and MPC the problem boils down to a Quadratic Program (QP) while Sequential Quadratic Programming (SQP), a cherished solution procedure for Nonlinear MHE (NMHE) and NMPC, is based on solving a series of related QPs. To meet this demand, QPALM-OCP is introduced: QPALM-OCP is a Proximal Augmented Lagrangian Method (P-ALM) tailored for the multi-stage structure arising in MHE and MPC. The algorithm relies on a semi-smooth Newton method to solve the inner ALM problem allowing multiple active set changes between iterates and benefits from warm-starting. Due to specialized low-rank matrix modifications, the iterates remain cheap while providing fast execution times. Contrary to conventional Quadratic Programming solvers (QP solvers), QPALM-OCP comes with guarantees of R-linear convergence to a stationary point of nonconvex QPs, making it an ideal candidate for SQP methods, especially when using exact second-order information. A prototype implementation, despite being non-optimized, outperforms state-of-the-art general-purpose QP solvers and is slightly faster than state-of-the-art Optimal Control Problem (OCP)-specific solvers in numerical experiments, demonstrating great potential for further development.

PICARIELLO, NICOLO'

Automatic gas source localization with UAV: a graph-based approach Masters Thesis

2024.

BibTeX

Talacci, Mattia

Optimal trajectory planning of a magnetic levitation carriage system for advanced manufacturing Masters Thesis

2024.

BibTeX

MOHAMMED, TAREG MAHMOUD HASSAN

Automatic control solutions to enhance the reliabilty of airborne wind energy systems PhD Thesis

2024.

Abstract | Links | BibTeX

@phdthesis{alma99479472608776,

title = {Automatic control solutions to enhance the reliabilty of airborne wind energy systems},

author = {TAREG MAHMOUD HASSAN MOHAMMED},

url = {https://hdl.handle.net/10589/228592},

year  = {2024},

date = {2024-10-14},

urldate = {2024-10-14},

abstract = {Airborne wind energy (AWE) is a technology that captures wind energy and transforms it into electricity using a flying apparatus connected through a tether to the ground. It holds substantial promise in working in conjunction with traditional wind turbines to mitigate CO2 and other greenhouse gas emissions, thus combating global warming. However, more research is required to optimize the use of AWE technology; Given the current status of AWE technology, there are significant challenges to its successful implementation that must be addressed. These challenges include 1) ensuring fully autonomous operation even beyond standard operating conditions and 2) enhancing the system’s robustness and failure tolerance. This thesis introduces strategies to enable autonomous operation beyond nominal operation condition, and to increase system reliability by applying fault tolerance control (FTC) for Airborne Wind Energy Systems (AWES). Specifically, this study introduces FTC algorithms tailored for Vertical Take-Off and Landing (VTOL) pumping cycle AWES, the method has been tested in a verified simulation environment, a control surface failure has been injected at different points during the production phase, and the proposed FTC algorithm effectively maintained system performance. Additionally, a "standby" safety mode is proposed for operating AWES outside its nominal operational conditions; this mode notably enables AWES to function when wind speeds fall below the operational cut-in point, the method has been tested in verified simulator in wind speed from 0 to 12 m/s, the results indicate that the method successfully kept the kite aloft even at 0 m/s wind speed, and shows reliable performance in higher wind speed. Moreover, in this research the developed AWES simulator has been validated using experimental data, which is considered one of the contributions of this research given the limited availability of validated simulators in the AWE domain. This simulator is considered a crucial instrument for exploring AWES failure modes and assessing the corresponding mitigation strategies.},

keywords = {},

pubstate = {published},

tppubtype = {phdthesis}

}

Airborne wind energy (AWE) is a technology that captures wind energy and transforms it into electricity using a flying apparatus connected through a tether to the ground. It holds substantial promise in working in conjunction with traditional wind turbines to mitigate CO2 and other greenhouse gas emissions, thus combating global warming. However, more research is required to optimize the use of AWE technology; Given the current status of AWE technology, there are significant challenges to its successful implementation that must be addressed. These challenges include 1) ensuring fully autonomous operation even beyond standard operating conditions and 2) enhancing the system’s robustness and failure tolerance. This thesis introduces strategies to enable autonomous operation beyond nominal operation condition, and to increase system reliability by applying fault tolerance control (FTC) for Airborne Wind Energy Systems (AWES). Specifically, this study introduces FTC algorithms tailored for Vertical Take-Off and Landing (VTOL) pumping cycle AWES, the method has been tested in a verified simulation environment, a control surface failure has been injected at different points during the production phase, and the proposed FTC algorithm effectively maintained system performance. Additionally, a "standby" safety mode is proposed for operating AWES outside its nominal operational conditions; this mode notably enables AWES to function when wind speeds fall below the operational cut-in point, the method has been tested in verified simulator in wind speed from 0 to 12 m/s, the results indicate that the method successfully kept the kite aloft even at 0 m/s wind speed, and shows reliable performance in higher wind speed. Moreover, in this research the developed AWES simulator has been validated using experimental data, which is considered one of the contributions of this research given the limited availability of validated simulators in the AWE domain. This simulator is considered a crucial instrument for exploring AWES failure modes and assessing the corresponding mitigation strategies.

10.

Vallese, Andrea

Air-gap eccentricity estimation in induction motors: the advantages of a physics-informed machine learning approach Masters Thesis

2024.

Abstract | Links | BibTeX

@mastersthesis{alma99475249308776,

title = {Air-gap eccentricity estimation in induction motors: the advantages of a physics-informed machine learning approach},

author = {Andrea Vallese},

url = {https://hdl.handle.net/10589/227717},

year  = {2024},

date = {2024-10-10},

urldate = {2024-10-10},

abstract = {In modern industrial environments, the reliability and efficiency of machinery are paramount, particularly for widely used induction motors. These motors, essential to numerous industrial applications, are susceptible to faults that can result in costly downtime and repairs. Condition Monitoring (CM) and Fault Detection (FD) have become crucial in predicting and diagnosing potential issues before they lead to catastrophic failures. Among the various faults, air-gap eccentricity — where the rotor's rotational axis misaligns with the stator's axis — stands out as one of the most common and damaging. This thesis presents an innovative non-invasive solution to the quantitative estimation of air-gap eccentricity in induction motors. The use of current sensors, which are often already integrated into the safety systems, guarantees a non-invasive approach to be integrated into industrial production systems. Traditional CM techniques, while effective, often struggle with accurately diagnosing and quantifying faults in complex, noisy environments. By integrating physical models with ML, PIML overcomes these limitations, offering a robust solution to the eccentricity estimation problem. Through a comparative analysis, this research demonstrates the superiority of PIML over both fully model-based and pure ML approaches. The proposed PIML method significantly enhances the accuracy of fault severity estimation, as evidenced by a more than 20% improvement in inference performance on experimental data, measured by the R^2 metric. The results confirm that PIML not only outperforms traditional methods but also provides a cost-effective, seamless solution for real-time CM and FD in industrial settings.},

keywords = {},

pubstate = {published},

tppubtype = {mastersthesis}

}

11.

KESSLER, NICOLAS MATTHIAS

Linear matrix inequality conditions for gain-scheduling and model predictive control PhD Thesis

2024.

Abstract | Links | BibTeX

@phdthesis{alma99477554108776,

title = {Linear matrix inequality conditions for gain-scheduling and model predictive control},

author = {NICOLAS MATTHIAS KESSLER},

url = {https://hdl.handle.net/10589/224812},

year  = {2024},

date = {2024-09-17},

urldate = {2024-09-17},

abstract = {This dissertation presents a novel approach to gain-scheduling model predictive control (MPC) for trajectory tracking on uncertain nonlinear systems, leveraging linear parameter-varying (LPV) models. A hierarchical scheme is developed, separating trajectory generation from stabilization using a 2-Degrees-of-Freedom (DoF) design. The focus of this thesis is the design of the feedback action, such that it guarantees tracking of the reference under bound satisfaction. A key innovation is the graph-based gain-scheduling variable, enabling modular feedback application for online decisions. Nonlinearities are taken into account by extending the resulting LPV model with a polytopic uncertainty. Initially, a simple Linear Matrix Inequality (LMI) conditions are proposed to address stabilizability and later extended to address performance in an MPC scheme. Subsequently, it yields a novel method for the systematic design of the terminal ingredients for an LTV MPC. The LTV MPC is then extended to a robust tube-MPC with constraint satisfaction. Efficient offline solvability of the resulting LMI conditions is addressed via the Alternating Direction Method of Multipliers (ADMM) to enable memory-efficient, distributed optimization. The proposed LTV MPC scheme is computationally efficient online, because the optimal control problem is structured as a convex Quadratic Program (QP), that exploits its temporal evolution. Simulation on a Continuously Stirred Tank Reactor (CSTR) and hardware implementation on a CrazyFlie drone demonstrate the approach's capability to stabilize nonlinear systems under disturbances and constraints with limited computing resources. These advancements, combined with efficient offline LMI solving, promise broad applicability for safety-critical industrial systems.},

keywords = {},

pubstate = {published},

tppubtype = {phdthesis}

}