Doctoral School

Reading date: 10/11/2026

• ALAHMAD, AHMAD: Joint State and Parameter Estimation for Structural System Identification: A Unified Statistical Framework across Static and Modal Data
  
  Author: ALAHMAD, AHMAD
  Programme: DOCTORAL DEGREE IN CONSTRUCTION ENGINEERING
  Department: Department of Civil and Environmental Engineering (DECA)
  Mode: Normal
  Deposit date: 29/09/2026
  Reading date: 10/11/2026
  Reading time: 11:15
  Reading place: UPC Campus Nord, Carrer de Jordi Girona, 1-3, 08034 BarcelonaEDIFICI C1- AULA 116
  Thesis director: TURMO CODERQUE, JOSE | MINGUEZ SOLANA, ROBERTO
  Thesis abstract: Assessing the condition of aging civil infrastructure rests on Structural System Identification (SSI): recovering physical parameters such as element stiffnesses and boundary conditions by reconciling a numerical model with measured responses. Every ingredient of this inverse problem is uncertain: measurements carry noise, loads and supports are imperfectly known, and the model idealizes the real structure. A trustworthy result must therefore do more than return a point estimate, yet least-squares estimation, sensitivity analysis, probabilistic uncertainty treatment, and observability theory have largely been pursued separately in structural identification rather than wired into a single estimator.The power systems community faced this challenge earlier. Since Static State Estimation (SSE) was formulated in the 1970s, it has built around the weighted least-squares (WLS) estimator an integrated statistical apparatus refined over more than five decades of operational use: pre-estimation observability analysis, covariance-based uncertainty quantification, bad-data detection through the largest normalized residual, and χ² testing of model adequacy. This thesis adapts that apparatus to structural identification and extends it from state-only to joint state and parameter estimation, with observability as a pre-estimation guard and the residual diagnostics as post-estimation guards, in a single WLS interface accepting static and operational-modal measurements.The methodology comprises five developments, validated in two experimental campaigns. A Jacobian-based observability criterion fixes parameter identifiability from the rank of the measurement Jacobian, replacing recursive symbolic methods at a fraction of their cost. The estimator is then equipped, for the first time in SSI to the author's knowledge, with covariance-based confidence intervals, χ² model-adequacy testing, and normalized-residual bad-data detection. Model-form error is treated directly: unknown loads and boundary conditions become estimable variables, removing the systematic bias they would induce, while principal-component analysis of the residuals localizes what remains. A resolution-matrix conservation diagnostic certifies in advance which property groups admit reliable integrated-stiffness recovery. Finally, static and modal residuals combined over a shared parameter vector localize a stiffness change as accurately as either modality alone, while an additive Fisher-information decomposition breaks the mass-stiffness degeneracy modal data cannot resolve in isolation.The framework is validated on a common laboratory specimen instrumented with low-cost MEMS sensors. A static campaign under an automated multi-load-case protocol confirms that the χ²-calibrated noise model transfers unchanged to an independent test, that sparse rotation measurements under diverse loading resolve a spatially detailed stiffness profile, and that the conservation integral recovers a reinforced section's stiffness to within a fraction of a percent of theory. A second campaign re-instruments the same specimen for output-only modal testing; with only the modal residual rows added, the same solver and diagnostics recover the same reinforcement, and the static and modal estimates agree with the theoretical reference to within about two percentage points. It also maps a limit of the self-diagnosis: an unmodeled support compliance, nearly collinear with the boundary stiffness, biases the boundary parameters without a signature in the diagnostics, surfacing only when the supports are modeled explicitly.
  

• BASAR, IMREN: Sustainable Repair Mortars for Historic Masonry: Design, Characterization and Performance Assessment of Natural and Recycled Materials
  
  Author: BASAR, IMREN
  Programme: DOCTORAL DEGREE IN CONSTRUCTION ENGINEERING
  Department: Department of Civil and Environmental Engineering (DECA)
  Mode: Normal
  Deposit date: 29/09/2026
  Reading date: 10/11/2026
  Reading time: 12:00
  Reading place: ETSECCPB, Sala C2-212 (C/Jordi Girona 1-3, mòdul C2, Campus Nord, Barcelona).
  Thesis director: FERNANDEZ CARRASCO, LUCIA | TORRENS MARTÍN, DAVID
  Thesis abstract: This thesis presents a comparative and conservation-oriented investigation of lime-based mortars employed in architectural heritage, with particular emphasis on material compatibility, durability, and long-term performance. Historical mortars sampled from two representative sites in Catalonia: the Roman archaeological site of Altafulla and the Edifici de l’Hivernacle in Barcelona, were studied alongside commercial restoration mortars and laboratory-designed lime-based formulations in order to evaluate technological choices, performance limits, and conservation suitability.A comprehensive multi-analytical methodology was adopted, integrating mineralogical, granulometric, mechanical, physical, and visual characterization techniques. This approach enabled the assessment of binder–aggregate relationships, microstructural features, mechanical response, and aesthetic behaviour, with special attention to parameters relevant to heritage conservation practice. The characterization of historical mortars revealed distinct technological strategies in raw material selection, aggregate grading, and binder composition, closely linked to the functional requirements and environmental exposure conditions of each site.Comparative analysis demonstrated that standardized commercial repair mortars, despite exhibiting higher mechanical strength, do not necessarily ensure improved compatibility with historic masonry substrates. In contrast, laboratory-prepared lime mortars, particularly CL90 air-lime-based systems, showed greater adaptability in balancing mechanical performance and deformability when curing conditions and environmental exposure were carefully controlled.Durability-related conditioning tests highlighted the critical influence of carbonation regime, moisture interaction, salt presence, and thermal cycling on mortar behaviour over time, underscoring the necessity of long-term evaluation beyond early-age mechanical performance. In situ qualitative observations conducted at the Monestir de Sant Joan les Fonts further confirmed satisfactory workability, adhesion, and visual integration of both reference and PDMS-modified lime mortars during the initial post-application period.Taken together, the results support an integrated and context-sensitive approach to the design and assessment of compatible lime-based repair mortars, providing experimental evidence that emphasizes the importance of aesthetic or integration considerations, material compatibility, durability, and environmental interaction in the conservation of architectural heritage.Keywords: Lime-based mortars, heritage conservation, material compatibility, durability, long-term performance, carbonation, CL90 air lime, repair mortars, microstructure, environmental exposure, thermal cycling, salt interaction, aesthetic considerations, architectural heritage. 
  

Reading date: 11/11/2026

• PONS RECASENS, GERARD: Evolving Knowledge Graphs: Continuous Knowledge Acquisition, Adaptation, and Exploitation
  
  Author: PONS RECASENS, GERARD
  Programme: DOCTORAL DEGREE IN COMPUTING
  Department: Department of Computer Science (CS)
  Mode: Normal
  Deposit date: 01/10/2026
  Reading date: 11/11/2026
  Reading time: 15:00
  Reading place: Sala d'Actes FIB - Edifici B6 Campus Nord
  Thesis director: QUERALT CALAFAT, ANNA | BILALLI, BESIM
  Thesis abstract: Modern systems continuously generate vast amounts of heterogeneous data. While this data contains valuable information, its full potential remains difficult to realize when it is treated as a collection of isolated observations lacking meaning and context. Unlocking its full value therefore requires enriching data with explicit semantics that capture concepts, relationships, and contextual information, transforming isolated observations into structured knowledge. This thesis argues that such semantic enrichment constitutes a fundamental step towards building more effective intelligent systems. In this context, Knowledge Graphs (KGs) provide a natural framework for representing and organizing this knowledge, while Knowledge Graph Embeddings (KGEs) transform it into latent representations that can be flexibly exploited by modern Machine Learning methods to address evolving analytical needs.As knowledge is not static, realizing this vision requires addressing challenges that span its entire life cycle. First, data originating from heterogeneous and unstructured sources must be incorporated into a continuously evolving knowledge structure, where it is enriched with semantics. Second, as this knowledge structure evolves, its latent representations must be continuously updated while preserving previously acquired knowledge. Finally, the evolving knowledge structure and its learned representations must be effectively exploited to support adaptive behavior in intelligent systems.This thesis addresses these challenges through the framework of evolving KGs. For knowledge acquisition from unstructured sources, it focuses on textual information and proposes a schema-aware entity disambiguation method that combines Large Language Models (LLMs) with the hierarchical structure of KGs, improving adaptability across domains without requiring model retraining. To maintain latent representations over time, it investigates the continual adaptation of KGEs. Within this scope, it identifies a previously overlooked phenomenon that contributes to the degradation of previously acquired knowledge and proposes an adjusted evaluation protocol to accurately capture its effects. Additionally, it introduces an initialization strategy for newly added embeddings that enhances knowledge acquisition while mitigating this performance degradation. Finally, for knowledge exploitation, it demonstrates how continuously evolving KGs and their learned representations can support adaptive recommendation and intent-aware assistance in experiment-driven data analytics environments. Together, these advancements contribute towards enabling evolving KGs to continuously acquire, adapt, and exploit knowledge.
  

Reading date: 12/11/2026

• BENÍTEZ COLOMINAS, POL: Modeling Lattice-Mediated Optoelectronic Response in Anharmonic Semiconductors: A First-Principles and Machine Learning Study
  
  Author: BENÍTEZ COLOMINAS, POL
  Programme: DOCTORAL DEGREE IN COMPUTATIONAL AND APPLIED PHYSICS
  Department: Department of Physics (FIS)
  Mode: Article-based thesis
  Deposit date: 15/10/2026
  Reading date: 12/11/2026
  Reading time: 10:00
  Reading place: Sala Polivalent, Edifici A, Escola d'Enginyeria de Barcelona Est (EEBE), Campus Diagonal Besos
  Thesis director: CAZORLA SILVA, CLAUDIO | SAUCEDO SILVA, EDGARDO ADEMAR
  Thesis abstract: The rapid development of next-generation energy technologies, such as photovoltaics, necessitates the discovery of materials with highly tunable and resilient optoelectronic properties. This dissertation investigates the fundamental role of lattice dynamics in phase stability and thermal renormalization of the optoelectronic properties of anharmonic materials exhibiting strong electron-phonon coupling, through an integrated approach combining first-principles quantum simulations and machine learning.The first part of this work analyzes in detail a family of materials known as antiperovskites, with chemical formula Ag3BC (B: S, Se; C: Cl, Br, I), that is fundamentally very interesting and highly relevant for energy and optoelectronic applications. In particular, by employing crystal structure prediction and free-energy calculations, we demonstrate that the cubic phase identified in experiments is dynamically stabilized at finite temperatures or represents kinetically trapped metastable states. Following this, we quantify a giant thermally induced band-gap renormalization in these systems, revealing that large-amplitude phonon displacements at room temperature can narrow the electronic band gap by more than 50%. This finding not only reconciles theory with experiment but also reveals a sensitive coupling between polar lattice distortions and electronic hybridization. We leverage this insight to propose a novel paradigm for “electro-optoelectronic” tuning, demonstrating that external electric fields can mimic thermal effects to provide non-chemical control over the band gap in centrosymmetric semiconductors.To scale these findings to more complex systems, the second part of the thesis focuses on the development of physics-informed artificial intelligence (AI) workflows. We show that graph neural networks trained on datasets generated via phonon-informed sampling, rather than random displacements, exhibit superior accuracy and generalizability. Through model interpretability analysis, we prove that these networks effectively learn the relative importance of specific chemical bonds in determining electronic responses. Finally, we implement an AI-workflow to evaluate the thermal response of complex solid solutions (Ag3SBr(x)I(1-x)), successfully simulating systems that were previously computationally prohibitively expensive using standard methods. Collectively, this research establishes that the explicit inclusion of lattice dynamics is essential for the accurate description of anharmonic materials and provides a robust, AI-accelerated framework for the “temperature-aware” design of functional semiconductors.
  

• DYHR, SØREN ISTVÁN ADORJÁN: Optimal Metrics in Odd-Dimensional Geometry: Reeb Dynamics and Applications in Fluid Mechanics
  
  Author: DYHR, SØREN ISTVÁN ADORJÁN
  Programme: DOCTORAL DEGREE IN COMPUTATIONAL AND APPLIED PHYSICS
  Department: Department of Physics (FIS)
  Mode: Normal
  Deposit date: 11/09/2026
  Reading date: 12/11/2026
  Reading time: 13:00
  Reading place: Sala de juntes Facultat de Matemàtiques i Estadística, Campus Sud - UPC
  Thesis director: MIRANDA GALCERÁN, EVA | GONZÁLEZ PRIETO, ÁNGEL
  Thesis abstract: Particles described by Hamiltonian mechanics tend to stay on odd-dimensional hypersurfaces of symplectic manifolds. In this thesis we study a generalization of the geometric structures appearing on such submanifolds. They all have a Reeb vector field and, in dimension three, coincide with stable Hamiltonian structures (SHS), a class which includes both cosymplectic and contact structures. We study two problems concerning metrics compatible with such structures.First we consider the Navier-Stokes equations, which describe viscous incompressible fluids. It turns out that Reeb vector fields of cosymplectic structures are stationary solutions for all compatible metrics. We create a cosymplectic plug from a construction by Cardona, Miranda, Peralta-Salas, and Presas, which realizes Turing machines as vector fields. Their construction was originally used in contact geometry to extend these vector fields to stationary solutions of the 3D Euler equations, which model ideal (inviscid) fluids. We extend their result by embedding the plug to obtain cosymplectic Reeb fields, and thus stationary solutions to the 3D Navier-Stokes equations, capable of simulating any Turing machine.Second we study an energy functional on the space of compatible metrics, originally considered for contact manifolds by Blair as well as by Chern and Hamilton. The metrics of minimal energy for contact manifolds in dimension three have been classified. We generalize the functional and the classification to three-dimensional SHSs and find that positive energy critical metrics correspond to the algebraic Anosov flows.Finally we obtain a similar result in higher dimension when there exist certain first integrals satisfying a non-degeneracy condition. We also classify a family of five-dimensional examples of critical metrics, including some not satisfying the non-degeneracy condition.