Doctoral School

Reading date: 29/10/2026

• CABRERA REDONDO, ÓSCAR: From Optimisation to Artificial Intelligence Solutions for Modern Power Systems
  
  Author: CABRERA REDONDO, ÓSCAR
  Programme: DOCTORAL DEGREE IN ELECTRICAL ENGINEERING
  Department: Department of Electrical Engineering (DEE)
  Mode: Normal
  Deposit date: 30/09/2026
  Reading date: 29/10/2026
  Reading time: 10:00
  Reading place: ETSEIB - UPC
  Thesis director: ARAGÜÉS PEÑALBA, MÒNICA
  Thesis abstract: Modern power systems are being transformed by decarbonisation, renewable generation, distributed energy resources, electrification, digitalisation, and the participation of new agents. These changes increase operational complexity and require faster, adaptive, and distributed decision-making tools. In this context, optimisation provides the structure to enforce physical and operational constraints, while artificial intelligence contributes prediction, approximation, adaptation, and privacy-preserving learning capabilities.This thesis investigates the transition from optimisation-based formulations towards artificial-intelligence-assisted solutions for modern power systems through three contributions at grid, community, and user levels. At the grid level, a physics-informed graph-based Lagrangian neural network is proposed for direct current optimal power flow approximation. By combining a graph neural network with an augmented Lagrangian loss, the model improves physical consistency under limited training. The approach is evaluated on benchmark electrical networks from 3 to 118 buses, including normal and stressed variants. Under normal conditions, the mean equality-constraint residual is reduced from 0.0219 p.u. to 0.0137 p.u., corresponding to a 37.44% improvement over a purely data-driven graph neural network, while achieving average speed-ups of 94 and 15 times over a widely used nonlinear optimisation solver under normal and extreme conditions, respectively.At the community level, the thesis analyses the smart operation of local energy communities under organisational and operational constraints. The results show that collective self-consumption can create substantial aggregate value, but that user composition and allocation rules strongly influence benefit distribution. In the evaluated case studies, configurations with 50% of users owning private photovoltaic generation reach the highest average total savings, 97.42€, while surplus reallocation increases total savings by up to 45.14€. However, fairness-aware mechanisms remain necessary to reduce negative individual outcomes.At the user level, the thesis proposes personalised federated learning strategies for household electricity demand forecasting. The proposed model-based clustering approaches improve personalisation without centralising private smart meter data. Compared with conventional federated learning, FedEC and FedKM reduce the mean squared error by 0.27% and 0.67%, respectively, with FedKM achieving the lowest aggregate error among the personalised methods considered.Overall, this thesis shows that optimisation and artificial intelligence are complementary components of modern power systems. Their responsible integration requires methods that are accurate, efficient, physics-aware, regulation-aware, and user-centred.
  

• DIAZ MOTTA, ARMANDO: Large Scale Wind-Powered Offshore Ammonia Production
  
  Author: DIAZ MOTTA, ARMANDO
  Programme: DOCTORAL DEGREE IN ELECTRICAL ENGINEERING
  Department: Department of Electrical Engineering (DEE)
  Mode: Normal
  Deposit date: 30/07/2026
  Reading date: 29/10/2026
  Reading time: 12:00
  Reading place: Aula 28.8 Sala Conferències - ETSEIB-UPC
  Thesis director: DÍAZ GONZÁLEZ, FRANCISCO | VILLA ARRIETA, MANUEL RICARDO
  Thesis abstract: Hydrogen and renewable ammonia can support the decarbonization of sectors where direct electrification is not feasible, with green ammonia serving as a hydrogen carrier between renewable production areas and distant demand centers.Among the large-scale pathways proposed in recent years, green ammonia production based on renewable electricity and water electrolysis has attracted increasing attention. However, limited attention has been given to the long-term feasibility of off-grid offshore ammonia production based on SOEC electrolysis and powered entirely by offshore wind. This thesis specifically addresses this knowledge gap by investigating whether large-scale, off-grid offshore green ammonia production based on solid oxide electrolysis (SOEC), a Haber-Bosch synthesis loop (HB), and offshore wind power can become a feasible pathway by 2050.This research aims to assess the technical and long-term economic feasibility of offshore wind-based ammonia production. To this end, the thesis combines a state-of-the-art assessment with a thermodynamic steady-state model of a 3 000 tpd SOEC-Haber-Bosch ammonia plant and an integrated pseudo-dynamic framework that couples the plant, offshore wind generation, battery storage, and hydrogen buffer storage. This framework includes an economic model and a capacity optimization algorithm to minimize the levelized cost of ammonia (LCOA). The pseudo-dynamic model simulates flexible plant operation over an extended load range under site-specific wind conditions. Three offshore locations with high-quality wind resources, namely La Guajira (Colombia), the North Sea (Denmark), and the Gulf of Roses (Spain), are assessed in this thesis. The results show that, at steady state and full load, the plant requires 1 030 MW and reaches an overall efficiency of 63% on an LHV-NH3 basis, while the SOEC electrolyzer achieves 87% on an LHV-H2 basis. The Haber-Bosch loop supplies 57% of the thermal power required for feedwater evaporation in the electrolysis process. Under partial-load operation, the SOEC unit reaches minimum loads of 61% and 72% for steam conversion scenarios of 70% and 85%, respectively, with stack temperature variations within ±30 °C. Under offshore operating conditions, the overall efficiency reaches 60%, and the plant capacity factor reaches 80% at the best-performing site, about 10% below that of a conventional grid-connected ammonia plant.Regarding economic feasibility, the results indicate that La Guajira and the North Sea could achieve the lowest levelized cost of ammonia by 2050, at 472 and 436 USD t-1, respectively, whereas the Gulf of Roses remains less competitive at around 720 USD t-1.A comparison across the selected sites further shows that inflexible Haber-Bosch operation, i.e., operation at 100% load, results in only marginal production cost differences relative to flexible operation over the 10-100% load range, but increases Haber-Bosch hot-standby events, revealing a trade-off between operational stability and storage requirements.The thesis provides a techno-economic assessment framework for offshore green ammonia production, while also supporting the capacity planning of offshore renewable resources. Furthermore, the results indicate that SOEC-based offshore ammonia could become cost-competitive by 2050, provided that costs decline for floating wind and solid oxide electrolysis, and that exogenous factors such as supportive market conditions and favorable policy frameworks emerge.
  

• MUKHOPADHYAY, SHRODDHA: Nonlinear optics at the nanoscale: harmonic generation in resonant metasurfaces
  
  Author: MUKHOPADHYAY, SHRODDHA
  Programme: DOCTORAL DEGREE IN COMPUTATIONAL AND APPLIED PHYSICS
  Department: Department of Physics (FIS)
  Mode: Normal
  Deposit date: 30/09/2026
  Reading date: 29/10/2026
  Reading time: 15:30
  Reading place: Sala de conferencias, planta baja, edificio TR1, Calle Colon 1, Campus Terrassa
  Thesis director: COJOCARU, CRINA MARIA | TRULL SILVESTRE, JOSE FRANCISCO
  Thesis abstract: This thesis aims to contribute to two important research areas in photonics: nonlinear optics and nanophotonics (in particular metasurfaces).These interconnected fields underpin the development of next‑generation nonlinear nanophotonic devices. As modern photonic technologies increasingly rely on nanomaterials, their reduced dimensionality gives rise to new physical regimes, and we require refined theoretical tools and experimental strategies to understand and harness the behavior of sub‑wavelength structures. The work integrates concepts from nonlinear optics, condensed‑matter physics, and nanophotonics. While nonlinear optics was historically explored in bulk crystals under phase-matching conditions, recent advances in nanofabrication have shifted the focus toward nanoscale platforms where light–matter interactions are dominated by strong confinement, surface effects, and nonlocal corrections. In this regime, under ultrafast and intense light excitation, traditional bulk descriptions of nonlinear polarization are insufficient, and microscopic models become necessary. The Maxwell-hydrodynamic theoretical framework used throughout this thesis fulfills this need for a unified description. This theoretical foundation is complemented by nonlinear simulations performed using FDTD and FEM numerical methods. Experimentally, the thesis implements table-top optical setups for high-precision measurements of the efficiencies of the nonlinear harmonic generation processes, and their enhancements in the abovementioned material platforms. In this regards, we reach detection deep into the UV, a limit of far field optical measurements, as well as report two and three orders of magnitude enhancements in the harmonics generated, direct implications of resonant light localization in investigated metasurfaces.Within this framework, this thesis examines harmonic generation in two representative classes of nanomaterials: ultrathin films of semiconductors and metals, and their corresponding one‑ and two‑dimensional periodic metasurfaces. Although the geometries and fabrication strategies differ across these systems, their nonlinear optical response can be interpreted and validated using our combined experimental-theoretical methodology, which applies seamlessly across the different materials and geometries investigated. This consistent approach across disparate platforms has been a guiding principle of the work, as it enables a systematic understanding of how free and bound electrons, surface contributions, and nonlocal effects jointly shape nonlinear emission at the nanoscale.A major component of the work is the study of resonant metasurfaces in plasmonic, dielectric and hybrid-plasmonic platforms, understanding of the material properties at the nanoscale along with the strong electromagnetic field enhancement to compensate for the reduced interaction volume at the nanoscale. How these resonances arise, how they hybridize, and how they shape nonlinear emission has been central to the progression of this work. The interplay between material dispersion, geometry, and electromagnetic coupling has revealed a rich landscape of phenomena that bridge classical and emerging regimes of nanophotonics. By unifying nonlinear optical theory with nanoscale material physics, this work provides a foundation for the design and optimization of nonlinear metasurfaces, compact frequency converters, and emerging quantum photonic platforms. The results demonstrate that ultrathin films and nonlinear metasurfaces are not simply miniaturized versions of bulk materials, but fundamentally distinct systems governed by additional physical mechanisms that dominate at nanometric scales. By integrating theory, simulation, and experiment, this work provides both conceptual clarity and practical methodologies for understanding and engineering nonlinear processes in next‑generation nanophotonic devices.
  

• OBRADOR REY, SERGI: Physics-based state estimation algorithms for battery management systems
  
  Author: OBRADOR REY, SERGI
  Programme: DOCTORAL DEGREE IN ENVIRONMENTAL ENGINEERING
  Department: Department of Civil and Environmental Engineering (DECA)
  Mode: Normal
  Deposit date: 18/09/2026
  Reading date: 29/10/2026
  Reading time: 11:00
  Reading place: CAMPUS BESOSSala Polivalent de l'Edifici A.EEBEAv. Eduard Maristany, 16
  Thesis director: TRILLA ROMERO, LLUÍS | CANALS CASALS, LLUC
  Committee:
       PRESIDENT: OYARBIDE URQUIZU, MIKEL
       SECRETARI: AMANTE GARCIA, BEATRIZ
       VOCAL: AUGUST JOHANSSON, SVEN AAKE
  Thesis abstract: For real sustainability, especially in the electric mobility sector, the trend of oversizing batteries and moving the computational capabilities of management systems to the cloud needs to be reduced. Such practices are harmful to the alignment of these systems with the eco-design principles that these technologies should follow, thus dramatically increasing the resources needed to produce and operate them. The real solution is to maximize the life and efficiency of the resources we do have.As a first step toward gaining better control, the state estimation algorithms of the cells within the battery must be improved. Hence, equipping the management system with capabilities derived directly from battery physics and electrochemistry is presented as the key concept of this thesis. Instead of states being estimated through externally processed empirical data, the deployment of advanced physics-based models directly onto the battery's embedded electronics is proposed. By this means, the internal dynamics and aging of the cells can be understood, predicted, and adapted in real-time with enough accuracy.This allows them to take control of their own operations and avoid early internal decay. The research undertaken in this thesis finally demonstrates that battery systems can be converted into more durable, efficient energy assets that are more in line with the principles of eco-efficiency only by directly integrating electrochemical knowledge into the battery’s embedded electronics.
  

Reading date: 30/10/2026

• NI CHUIREAIN, FIONNUALA: Intrinsic Randomness in Quantum Information Theory
  
  Author: NI CHUIREAIN, FIONNUALA
  Programme: DOCTORAL DEGREE IN PHOTONICS
  Department: Institute of Photonic Sciences (ICFO)
  Mode: Normal
  Deposit date: 24/09/2026
  Reading date: 30/10/2026
  Reading time: 10:00
  Reading place: ICFO Auditorium
  Thesis director: ACÍN DAL MASCHIO, ANTONIO
  Thesis abstract: The unpredictability of quantum physics gives rise to an intrinsic form of randomness. We generate it by performing a measurement on a quantum state. In an adversarial scenario, we then quantify the randomness by the probability that a correlated eavesdropper could correctly guess the measurement outcomes. This thesis addresses the following question: given a known quantum state and measurement, how much intrinsic randomness can we generate?.We begin by solving the maximal intrinsic randomness of any state, optimised over all projective measurements, for four figures of merit: the conditional min, max and von Neumann entropies, and the newly introduced unambiguous randomness. The last of these describes an eavesdropper who is never wrong, but can sometimes return an inconclusive outcome. Relaxing the assumption of perfect accuracy, we also consider randomness given a fixed rate of inconclusive outcomes. We solve both of these quantities for any state and projective measurement in dimension two. If the setup is used to choose random bases for a prepare-and-measure quantum key distribution protocol, we show that the unambiguous randomness captures the knowledge gained by an eavesdropper about the secret key, without causing any disturbance. We then turn to extremal measurements beyond the projective case, which may have more outcomes. An unbiased measurement is one whose outcomes all have the same a priori probability. We characterise the randomness generated by any unbiased extremal rank-one measurement acting on any state, solving the problem explicitly in dimension two. Four-outcome measurements of this type are tomographic, such that one can fully determine the state from the outcome probabilities, so these results hold for fully source-device-dependent randomness too. The tetrahedral symmetric informationally complete (SIC) measurement, we find, has the least intrinsic randomness within this class. We also present the biased skewed SIC family of measurements, and use them to prove that 2 log d bits of randomness, the maximal amount, can be generated device-dependently (or source-device-independently) in any dimension d in which there exists a SIC measurement.Reversing the scenario, we optimise over all states to solve the maximal intrinsic randomness of two classes of non-extremal, or noisy, measurements: any two-outcome measurement in dimension two, and projective measurements in any dimension affected by isotropic noise. In a realistic implementation, however, neither the state nor the measurement is completely free of noise. Taking a simple example of a state measured in a basis of any dimension, we compare the case where only one device is affected by noise to that where both devices are noisy, such that an eavesdropper may have joint correlations with the state and the measurement. For a fixed amount of total noise, we find that, contrary to the common practice of attributing all experimental noise to a single device, the joint-noise eavesdropper strictly outperforms her single-noise counterpart, compromising the security of the randomness generated.That the outcomes of a quantum measurement are unpredictable has been known since the inception of quantum theory. This thesis presents a step forward in analytically characterising the intrinsic randomness of quantum physics. On the practical side, our results bound the functionality of quantum random number generators, under various assumptions. On a more foundational level, this work provides a new lens through which to compare quantum states and measurements and, from their vast landscape, to see which of them emerge as the most random.