Doctoral School

Reading date: 31/07/2026

• HARBA, MOHANAD: Musculoskeletal simulations Integrating a smooth knee contact pressure model
  
  Author: HARBA, MOHANAD
  Programme: DOCTORAL DEGREE IN MECHANICAL, FLUIDS AND AEROSPACE ENGINEERING
  Department: Department of Mechanical Engineering (EM)
  Mode: Normal
  Deposit date: 07/07/2026
  Reading date: 31/07/2026
  Reading time: 10:00
  Reading place: Sala Polivalent de l'EEBE
  Thesis director: SERRANCOLÍ MASFERRER, GIL
  Thesis abstract: Knee osteoarthritis is one of the most common musculoskeletal conditions in the world, affecting hundreds of millions of people and causing chronic pain, reduced mobility, and disability. Understanding how mechanical forces are distributed across the knee joint during activities like walking is key for better diagnosis, treatment planning, and implant design. However, directly measuring these forces in a clinical setting is not possible, and existing computational models often cannot estimate the full spatial distribution of contact pressure across the joint surface, instead they estimate the resultant forces and moment at the knee joint.This thesis addresses these limitations through three studies. The first study developed a method to reduce the complexity of muscle-tendon length and moment arm parameterizations in musculoskeletal simulations. By applying an error-threshold approach to remove redundant polynomial coefficients, the method reduced the number of required parameters by approximately 50 % for muscles spanning the knee and ankle, while preserving accuracy in joint kinematics and knee contact force tracking. This led to a 15.6 % reduction in average computation time.The second study developed and validated a smooth, mesh-based knee contact pressure model included in a full body movement simulation. The model computes contact pressure at each triangular face of the tibial surface at every time step of the simulation, using smoothed penetration detection and a linear elastic foundation relationship. A sensitivity analysis was carried out to evaluate the influence of key model parameters on computation time, convergence, and tracking accuracy. The model was validated on eight gait trials from the Grand Challenge to predict in ivo knee loads dataset, and was also used in a predictive simulation, demonstrating its ability to generate new movement patterns while accounting for joint contact mechanics.The third study investigated the spatial relationship between tibiofemoral contact pressure and tibial cartilage composition, assessed using quantitative MRI T₂ relaxation time mapping, in a group of 23 subjects including 5 healthy individuals and 18 patients with early osteoarthritis. In healthy subjects, no consistent relationship was found between contact pressure and T₂ values. In the early osteoarthritis group, however, the most heavily loaded cartilage regions tended to show above-average T₂ values, suggesting that early compositional changes in cartilage are spatially linked to mechanical loading. This finding was consistent across all pressure thresholds tested. Together, this work shows that it is possible to estimate how contact pressure is distributed across the knee during walking within a computationally efficient simulation framework. The smooth contact model developed here removes a key barrier that has long prevented pressure estimation from being integrated into full-body movement simulations. These findings bring us one step closer to bridging the gap between computational biomechanics and clinical practice, where understanding the link between joint loading and cartilage health could ultimately inform more targeted strategies for osteoarthritis prevention and management.
  

Reading date: 03/09/2026

• GARCÍA LIMÓN, JOSÉ ALBERTO: Diseño e implementación de un sistema de medida no invasiva de parámetros cardiovasculares en calzado
  
  Author: GARCÍA LIMÓN, JOSÉ ALBERTO
  Programme: DOCTORAL DEGREE IN ELECTRONIC ENGINEERING
  Department: Department of Electronic Engineering (EEL)
  Mode: Change of supervisor
  Deposit date: 29/06/2026
  Reading date: 03/09/2026
  Reading time: 11:00
  Reading place: Auditorio del departamento de Ingeniería Eléctrica del CINVESTAV, México
  Thesis director: CASANELLA ALONSO, RAMON | ALVARADO SERRANO, CARLOS
  Thesis abstract: Cardiovascular diseases remain the leading cause of mortality worldwide, driving the development of technologies for continuous monitoring. Among the most relevant biomarkers are heart rate (HR), heart rate variability (HRV), and pulse transit time (PTT), which is closely related to arterial stiffness and blood pressure. However, accurate estimation of PTT in wearable devices remains a challenge, mainly due to the difficulty of obtaining proximal references unaffected by the cardiac pre-ejection period (PEP).This thesis proposes a novel approach for cardiovascular monitoring based on plantar acquisition of physiological signals through sensors integrated into a shoe insole. A measurement system based on a piezoelectric sensor in the heel region was developed to capture the ballistocardiogram (BCG), a mechanical signal associated with blood ejection. Widely accepted physiological models relate its morphology to pressure gradients in the aorta, enabling the extraction of both proximal and distal information along the aortic segment. The proposed system was validated against the standard weighing scale BCG and employed to estimate several cardiovascular parameters.In addition, a continuous wavelet transform (CWT)-based algorithm was validated for robust J-wave detection in BCG signals. Evaluated using multiple datasets with different sensors, body positions, and mechanical interfaces, it yielded the following sensitivity (Se) and positive predictive value (P+): chair (99.87% and 98.48%), bed (99.64% and 97.60%), and weighing scale (99% and 88%) over a total of 9,266 heartbeats, demonstrating robust performance for HR estimation. This behavior was also observed using the proposed plantar BCG system, achieving P+ and Se values of (97.07% and 84.85%) and (98.20% and 93.42%) during standing and seated conditions, respectively, with coverage factors above 95%. Furthermore, robust estimation of temporal HRV parameters was achieved in the seated position, and a significant correlation (r = 0.8, p < 0.001) was observed between the plantar BCG I–J interval and carotid–femoral pulse transit time, suggesting its potential as an indirect marker of arterial stiffness.As a complementary approach, this work explored impedance plethysmography (IPG) for distal pulse arrival time (PAT) detection at the foot. Results showed that, although photoplethysmography generally provides higher signal quality, its performance is more sensitive to sensor contact pressure and peripheral vasoconstriction. In contrast, the IPG enabled robust pulse detection under realistic operating conditions, including the presence of textile layers, making it particularly suitable for footwear-based systems.Finally, combining BCG and IPG signals enabled the estimation of whole-body pulse transit time (wb-PTT), showing a significant correlation (r = 0.72, p = 0.001) with the reference system based on carotid tonometry and photoplethysmography. This approach overcomes the limitations of conventional wearable devices based on short arterial paths and enables continuous, non-invasive cardiovascular monitoring compatible with daily activities. Overall, the results demonstrate the feasibility of a new generation of smart insoles. Integrating BCG and IPG into a single platform enables a comprehensive characterization of cardiovascular status and opens new opportunities for applications in digital health, preventive medicine, and remote clinical monitoring.
  

• MULERO JIMÉNEZ, LORENA: Bosc i Sostenibilitat: proposta educativa a l’entorn dels serveis ecosistèmics del bosc i els ODS amb pedagogia de ciència oberta a secundària
  
  Author: MULERO JIMÉNEZ, LORENA
  Programme: DOCTORAL DEGREE IN NATURAL RESOURCES AND THE ENVIRONMENT
  Department: Department of Mining, Industrial and ICT Engineering (EMIT)
  Mode: Normal
  Deposit date: 18/06/2026
  Reading date: 03/09/2026
  Reading time: 12:00
  Reading place: Sala de Juntes de l'EPSEM
  Thesis director:
  Thesis abstract: We live in a society that normalizes crises. These crises affect the very pillars of civilization, and education is the engine that moves the world. Furthermore, what would become of society and the education of new generations without caring for the environment and nature?This thesis develops a series of actions related to forests and their contribution to sustainability through their connection with the Sustainable Development Goals (SDGs). This is a process of raising awareness among secondary school students regarding their essential role in the journey toward a sustainable and respectful society. Likewise, it analyzes the impact of implementing the "Forest and Sustainability Project," applying the innovative teaching methodology known as Open Science Schooling. The study focuses on how this approach fosters students' awareness of their role as key agents in changing society and steering it toward sustainability.A case study is conducted using teaching materials created and developed within this thesis. The activities are implemented in secondary schools across Catalonia. Additionally, activities are carried out within the framework of the Exploratori dels Recursos de la Natura that aims to promote science and technology among youth and conducts activities addressed to both secondary school teachers and students.
  

Reading date: 04/09/2026

• BANDARRINHA MONTEIRO, CARLOS ALEXANDRE: Pseudo-Boiling Wall-Bounded Turbulence in Supercritical Flows: Direct Numerical Simulation & Flow Physics
  
  Author: BANDARRINHA MONTEIRO, CARLOS ALEXANDRE
  Programme: DOCTORAL DEGREE IN MECHANICAL, FLUIDS AND AEROSPACE ENGINEERING
  Department: Department of Mechanical Engineering (EM)
  Mode: Article-based thesis
  Deposit date: 14/07/2026
  Reading date: 04/09/2026
  Reading time: 11:30
  Reading place: Sala Polivalent, Edifici A, Campus Diagonal-Besòs
  Thesis director: JOFRE CRUANYES, LLUÍS
  Thesis abstract: This thesis presents the development and application of a high-fidelity computational framework for the direct numerical simulation (DNS) of high-pressure transcritical turbulent flows. These flows, characterized by strong thermophysical property variations in the vicinity of the pseudo-boiling region, exhibit complex interactions between turbulence, thermodynamics, and transport processes that cannot be accurately captured using classical modeling approaches. The primary objective of this work is to advance both the numerical simulation tool and the physical understanding required to reliably simulate and analyze such regimes. To this end, the multiphysics flow solver, RHEA (Reproducible Hybrid-architecture flow solver Engineered for Academia), is redesigned into a GPU-resident version, RHEA-X. The solver combines low-dissipation spatial discretizations with strong-stability-preserving Runge–Kutta time integration and advanced thermodynamic and transport models, including real-fluid equations of state. Its modular, object-oriented design enables the treatment of compressible, high-pressure transcritical, particle-laden, and immersed-boundary flows within a unified framework. The implementation leverages MPI and OpenACC to achieve persistent GPU data residency and direct inter-device communication, enabling efficient execution on modern heterogeneous high-performance computing systems. Performance analyses demonstrate near-ideal strong and weak scaling and substantial reductions in time-to-solution relative to CPU-based implementations up to a large GPU count. Using this framework, the thesis first investigates the grid resolution requirements for DNS of high-pressure transcritical wall-bounded turbulence. It is shown that classical resolution criteria, developed for incompressible flows, are insufficient in the presence of strong thermophysical gradients. In particular, the pseudo-boiling region induces localized amplification of turbulence intensity and energy dissipation, requiring refined resolution to accurately capture up to second-order statistics. New resolution guidelines are proposed based on hydrodynamic and thermodynamic scaling arguments. The physical behavior of transcritical flows is then examined in microconfined square-duct configurations spanning subcritical to supercritical regimes. The results reveal that, under transcritical conditions, flows can transition from laminar to turbulence-like states without changes in geometry or bulk velocity scales, driven by strong density gradients associated with the pseudo-boiling region. This leads to significant amplification of velocity fluctuations, intensified secondary motions, and substantial enhancement of heat transfer. Finally, the role of buoyancy is analyzed in transcritical turbulent square-duct flows. It is demonstrated that buoyancy influences the flow primarily through the modulation of secondary motions, which act as the dominant mechanism for redistributing momentum and restructuring turbulence. The case-dependent interaction between buoyancy, thermophysical gradients, and secondary flows leads to strong spatial heterogeneity, non-equilibrium turbulence behavior, and, consequently pronounced variations in wall shear stress and transport processes.
  

• CARRINO, CASIMIRO PIO: Multilingual and Cross-Lingual Machine Reading Comprehension in Few-Resource Settings: Synthetic Data, Cross-Lingual Transfer, and Domain-Specific Evaluation
  
  Author: CARRINO, CASIMIRO PIO
  Programme: DOCTORAL DEGREE IN SIGNAL THEORY AND COMMUNICATIONS
  Department: Department of Signal Theory and Communications (TSC)
  Mode: Normal
  Deposit date: 06/07/2026
  Reading date: 04/09/2026
  Reading time: 12:00
  Reading place: Aula MERIT D5010, Campus Nord, Barcelona
  Thesis director: RODRIGUEZ FONOLLOSA, JOSE ADRIAN
  Thesis abstract: Machine Reading Comprehension (MRC) is a fundamental benchmark for evaluating natural language understanding systems, requiring models to generate answers grounded in context. Despite advances with transformer-based architectures and large language models (LLMs), challenges persist in multilingual and cross-lingual scenarios, especially under few-resource and domain-specific constraints. This thesis addresses these challenges through three principal contributions, spanning the progression from encoder-based extractive to generative MRC, and covering data regimes from large-scale in-language supervision to zero-shot cross-lingual transfer in domain-specific settings.The first contribution introduces the Translate-Align-Retrieve (TAR) methodology, which preserves answer span alignment during translation of extractive QA datasets. By integrating neural machine translation, unsupervised word alignment, and alignment-based answer retrieval, TAR enables systematic construction of high-quality multilingual training data for low-resource languages. This approach produced SQuAD-es v1.1, the first large-scale Spanish extractive QA dataset (87,599 question-answer pairs). Models trained on this resource achieve or surpass established baselines on Spanish QA benchmarks, demonstrating that alignment-based translation is effective for mitigating data scarcity.The second contribution develops a self-knowledge distillation framework with cross-lingual sampling to enhance transfer in few-resource settings with limited parallel data. This work addresses Cross-Lingual Transfer (XLT) and Generalised Cross-Lingual Transfer (G-XLT), where question and context are in different languages. The method integrates cross-lingual sampling as data augmentation with self-knowledge distillation as regularisation, introducing a mean Average Precision at k (mAP@k) coefficient to modulate distillation loss by teacher model quality. Evaluations on MLQA, XQuAD, and TyDiQA-GoldP demonstrate consistent improvements over standard fine-tuning, providing a robust alternative to machine translation-based augmentation when only limited parallel resources exist.The third contribution presents a semi-automatic methodology for constructing privacy-preserving multilingual QA benchmarks in sensitive, data-scarce domains, exemplified by the Human Resources (HR) domain with JobResQA. This combines data de-identification, LLM-based synthesis, and human-in-the-loop translation with MQM error annotations and post-editing. JobResQA comprises 105 synthetic résumé–job description pairs and 581 QA items across five languages and four question types. Baseline evaluation using the G-Eval LLM-as-judge metric across nine open-weight models reveals persistent cross-lingual challenges, with only the largest models attaining high accuracy across all languages. The methodology also enables bias and fairness investigations in high-stakes HR applications.Collectively, these contributions show that data scarcity, cross-lingual transfer, and domain specificity persist as central challenges across architectural paradigms, particularly in few-resource settings. All resources, including datasets, models, and code, are released as open-source. The TAR methodology has seen widespread adoption with over 100 citations, producing datasets for diverse languages and scripts. While performance has improved with LLMs, the core challenges addressed in this thesis remain central to advancing multilingual and cross-lingual MRC.