Doctoral School

Reading date: 14/05/2026

• AREIAS FANZERES, LEONARDO: Sound-to-Image Translation Through Direct Cross-Modal Learning: An Exploratory and Architectural Study
  
  Author: AREIAS FANZERES, LEONARDO
  Thesis file: (contact the Doctoral School to confirm you have a valid doctoral degree and to get the link to the thesis)
  Programme: DOCTORAL DEGREE IN SIGNAL THEORY AND COMMUNICATIONS
  Department: Department of Signal Theory and Communications (TSC)
  Mode: Normal
  Deposit date: 16/04/2026
  Reading date: pending
  Reading time: pending
  Reading place: pending
  Thesis director:
  Thesis abstract: Environmental sound conveys rich semantic and contextual information about events, objects, and spatial dynamics. However, prevailing computational approaches to environmental audio analysis, such as Acoustic Event Detection (AED), typically reduce this complexity to discrete textual labels. While effective for automated monitoring tasks, such representations oversimplify acoustic scenes and become inadequate when auditory information must be communicated across modalities. Sound-to-image (S2I) translation offers an alternative approach in which a model synthesizes images that visually depict sound-emitting sources and their surrounding environments.This thesis introduces and advances direct sound-to-image translation, a paradigm that establishes a connection between audio and visual modalities without relying on textual mediation, class supervision, or cluster-based alignment during training. The central hypothesis is that higher-level abstractions learned by deep neural networks provide a shared semantic space in which heterogeneous modalities can connect directly, enabling the generation of images that are interpretable and semantically coherent with the source sound. Such outputs are referred to as informative, meaning that they visually communicate meaningful aspects of the acoustic event.The first part of the thesis presents, to the best of our knowledge, the first study dedicated to direct S2I translation. A densely connected generative adversarial network (GAN), conditioned on audio embeddings, is developed to synthesize images directly from sound. Because multiple plausible images may correspond to a single acoustic event, translation quality cannot be evaluated through pixel-level reconstruction. To address this challenge, an informativity-based evaluation framework is proposed, employing classifiers to determine whether generated images are interpretable and semantically coherent with the source audio. Experiments reveal that, despite the inherent difficulty of the task, the model generalizes to unseen sounds and produces informative outputs for a meaningful portion of translations. Analysis further reveals that latent bottleneck dimensionality influences translation behavior, exposing a trade-off between pixel-space convergence and informativity.Building on this foundation, the second part investigates whether attention mechanisms can strengthen cross-modal alignment. Self-attention and cross-attention modules are integrated into the generator and evaluated across multiple configurations. Results show that attention improves translation performance when applied at early stages of the network, increasing informativity relative to the purely convolutional baseline while preserving the fully direct audio–visual connection.Together, these contributions formally define, validate, and extend direct S2I translation as a distinct research paradigm. Beyond architectural design and training strategies, this thesis advances methodological principles for quantitative evaluation of cross-modal translation in the absence of deterministic visual ground truth. The findings contribute to a broader understanding of multimodal representation learning and highlight the potential of direct S2I translation for applications in multimodal interaction and accessibility-oriented technologies, particularly for enhancing situational awareness in deaf and hard-of-hearing individuals.
  

• BADAKHSHAN, EHSAN: Thermo-Hydro-Mechanical Modeling of Unsaturated Vegetated Slopes
  
  Author: BADAKHSHAN, EHSAN
  Thesis file: (contact the Doctoral School to confirm you have a valid doctoral degree and to get the link to the thesis)
  Programme: DOCTORAL DEGREE IN GEOTECHNICAL ENGINEERING
  Department: Department of Civil and Environmental Engineering (DECA)
  Mode: Article-based thesis
  Deposit date: 16/04/2026
  Reading date: pending
  Reading time: pending
  Reading place: pending
  Thesis director: VAUNAT, JEAN
  Thesis abstract: The top layer of soil on a slope is highly dynamic, influenced by atmospheric forces and the presence of vegetation. The vegetation condition plays a crucial role in determining the amount of water transpiration and evaporation from the slope surface, thus affecting slope performance by altering temperature, water content, and pore water pressure within the soil mass. Many numerical models often neglect the vegetation impact on soil performance. To capture these effects, this study employs a finite element model in CODE_BRIGHT that integrates vegetation and climate-driven interactions, aiming to bridge the gap between reality and simulation. A specialized boundary condition is developed to simulate soil–vegetation–atmosphere (SVA) processes with hydraulic hysteresis, linking canopy resistance to solar radiation, vapor pressure deficit, and soil saturation. Model validation using three years of field data from the Agropolis slope in Barcelona shows strong agreement, confirming its ability to reproduce vegetation effects on slope hydrothermal behavior. Results reveal pronounced daily temperature variations in the root zone, higher temperatures on south-facing slopes, and stronger drying under vegetation during hot, dry periods.Parametric analysis identifies leaf area index (LAI), root density, and vegetation fraction as key factors influencing soil moisture. Root density and LAI most strongly affect water retention, dense roots lower summer saturation by up to 40%, while high LAI reduces surface drying by 30%. Vegetation fraction enhances winter storage but intensifies summer drying.To better represent unsaturated soils, a hysteretic soil–water retention model is implemented in CODE_BRIGHT, coupling suction and void ratio changes to improve accuracy in scenarios such as rainfall-induced landslides.Finally, an enhanced Barcelona Basic Model (BBM-VEG) is formulated within a thermo-hydro-mechanical (THM) framework. The model introduces a strain-dependent reinforcement factor (Rpveg), correlated with root mass fraction and activation strain, allowing soil stiffness and strength to evolve dynamically. Validated with triaxial and tensile tests, it accurately captures the mechanical response of rooted soils. Slope simulations under hydraulic (infiltration rates of 0.001–0.003 and 0.0005 kg/s) and thermal (15–60 °C) cycles show that vegetation limits infiltration, enhances evapotranspiration, expands the unsaturated zone, and reduces deformation by up to 70% compared to bare slopes.
  

Reading date: 15/05/2026

• ARIAS CUEVAS, JOSÉ GABRIEL: Proyectos de recuperación de zonas vulnerables con materiales de ciclo cerrado. Casos de estudio, proyectos URBE.
  
  Author: ARIAS CUEVAS, JOSÉ GABRIEL
  Thesis file: (contact the Doctoral School to confirm you have a valid doctoral degree and to get the link to the thesis)
  Programme: DOCTORAL DEGREE IN ARCHITECTURAL, BUILDING CONSTRUCTION AND URBANISM TECHNOLOGY
  Department: Department of Architectural Technology (TA)
  Mode: Normal
  Deposit date: 05/03/2026
  Reading date: 15/05/2026
  Reading time: 10:00
  Reading place: EPSEB (Escuela Politécnica Superior de Edificación BCN) - Sala de ActosAv. Doctor Marañón, 44-50 - 08028 - Barcelona
  Thesis director: BOSCH GONZÁLEZ, MONTSERRAT
  Thesis abstract: The Dominican Republic, specifically Santo Domingo, has faced decades of unplanned urbanization in highly vulnerable areas, such as the banks of the Ozama River. This generates a socio-environmental risk intensified by overpopulation and a lack of specific public policies. Simultaneously, the absence of an integrated system for managing Construction and Demolition Waste (CDW), often disposed of improperly, causes critical environmental impacts. This research addresses the integration of CDW management and closed-loop materials into urban redevelopment projects promoted by the State in critical areas. Using La Nueva Barquita and Domingo Savio (URBE Projects) as case studies, the research serves as a starting point for future interventions in the country.The central objective is to propose construction alternatives to existing ones for urbanization projects in flood-prone areas through the systematic use of closed-loop materials and CDW recovery and valorization systems. It seeks to provide strategic knowledge so that these urban interventions can become "waste sinks," driving sustainable, socially committed, and viable development.The research follows a mixed methodology combining: a documentary study and critical analysis of local regulations; international references; and the state of the art regarding vulnerability and river intervention projects. It includes the analysis of the case study settlements, the construction sector and its main stakeholders, and the regulatory framework, alongside successful experiences of urban interventions using recycled products. Field research involved site visits and surveys of both residents of the Nueva Barquita project and key stakeholders in Santo Domingo's construction sector. Finally, a technical-economic analysis of work items was conducted using quantification tools such as TCQ/BEDEC. This triple approach allowed for a comprehensive understanding of local barriers and opportunities.The results are framed within three transformation vectors:Regulatory/Institutional Vector: Although a legal framework exists, technical instruments and specific contractual clauses are needed to integrate CDW management as a mandatory requirement in public works.Socio-Economic Vector: The research identified an active informal reuse market, demonstrating latent demand. Formalizing this sector can generate a new competitive and formalized economic fabric.Technical/Territorial Vector: The real feasibility of replacing conventional materials with CDW in redevelopment projects is validated, optimizing resilience on riverbanks through solutions combined with nature-based solutions (NBS).The findings demonstrate that while the Dominican State has a visible social commitment to relocating vulnerable populations, a critical gap persists in regulatory application and technical CDW integration. The main contribution of this thesis lies in having integrated a technical, regulatory, and territorial diagnosis that validates the hypothesis that urban projects can function as "waste sinks" in the Dominican context.The research establishes technical, institutional, and market foundations for the Dominican Republic to adopt a circular economy model in construction. The sector's robust growth presents an exceptional opportunity to implement this model, attracting sustainable investment and strengthening international competitiveness. This transforms the waste challenge into a strategic lever for resilient urban development and national economic prosperity.
  

• FORERO ORTIZ, EDWAR ANDRÉS: Long-term renewal framework for urban water distribution networks: an application in the Barcelona Metropolitan Area
  
  Author: FORERO ORTIZ, EDWAR ANDRÉS
  Thesis file: (contact the Doctoral School to confirm you have a valid doctoral degree and to get the link to the thesis)
  Programme: DOCTORAL DEGREE IN CIVIL ENGINEERING
  Department: Barcelona School of Civil Engineering (ETSECCPB)
  Mode: Normal
  Deposit date: 17/03/2026
  Reading date: 15/05/2026
  Reading time: 12:00
  Reading place: ETSECCPB, Sala Tesines C1-002 (C/Jordi Girona 1-3, mòdul C1, Campus Nord, Barcelona).
  Thesis director: SANCHEZ JUNY, MARTI | MARTÍNEZ GOMARIZ, EDUARDO
  Thesis abstract: Context and BackgroundManaging urban water infrastructure in the 21st century involves balancing systemic asset ageing with rising service expectations and increasing climate stressors. In established metropolitan networks, traditional renewal policies based on reactive maintenance or simple age-based heuristics are becoming less effective at preventing cascading risks of structural failure. Ensuring the sustainability of Water Distribution Networks (WDNs) now requires a shift to proactive stewardship frameworks that can optimise capital allocation over multiple decades while addressing the complex interdependencies of the water-energy nexus and urban mobility.Problem StatementA fundamental disconnect persists between asset-specific stochastic deterioration modelling and system-wide strategic renewal planning. Existing failure prediction models often operate independently of operational decision-support systems, failing to translate probabilistic outputs into actionable long-term policies. Additionally, the integrity of infrastructure datasets is compromised by administrative fragmentation of pipe records and survival bias, which can lead to a significant underestimation of failure risk. There is a critical methodological gap in combining high-fidelity Machine Learning (ML) with Survival Analysis (SA) to project Level-of-Service (LoS) trajectories that incorporate social and environmental externalities as equally important factors in planning.Research ObjectivesThe main aim of this doctoral thesis is to develop and validate a multi-layered computational framework, the Renovation Strategy Simulator (comparator) 'REPIPE', designed to systematically evaluate different pipe renovation strategies for the drinking water distribution system in the Barcelona Metropolitan Area. Specifically, the thesis aims to: (i) improve short-term failure classification through business-oriented hyperparameter optimisation; (ii) create stratified survival models that consider dual temporal censoring and recurrent failure events; and (iii) design an iterative algorithmic framework to convert asset-level prognoses into internally consistent network performance trajectories over a 25-year planning period.MethodologyThis research establishes a unified "data-to-decision" pipeline supported by three key computational pillars. First, a novel Genealogical Tree-Based Inventory (GTBI) was designed to trace pipe segments back to their Original Pipe Section (OPS) identifiers, resolving historical gaps in the asset register. Second, a hybrid predictive framework was developed, combining an eXtreme Gradient Boosting (XGBoost) algorithm for tactical screening with a stratified Cox Proportional Hazards (CPH) model for multi-decadal prognosis. Third, these models were integrated into a simulation engine that employs MIVES-based value functions and conditional survival probabilities to project ten LoS indicators across six annual investment scenarios (€0M–€30M).Main ResultsThe implementation of the GTBI corrected 22% of accumulated failures that databases had previously misattributed, strengthening the evidence base for failure risk modelling. The study applied a 5% annual renewal threshold to keep all scenarios directly comparable and to support a consistent calculation of the failure-avoidance rate under a fixed yearly renovation capacity. Under this constraint, the XGBoost classifier, tuned through a brute-force hyperparameter search using a business-oriented Lorenz metric, achieved a reported 32% relative increase in failure-prevention effectiveness over default settings, capturing 30.2% of expected failures within the 5% renewal limit.
  

• KASULURU, VAISHNAVI: AI-Driven Network Service Management for Efficient and Sustainable Open-RAN Systems in 6G and Beyond
  
  Author: KASULURU, VAISHNAVI
  Thesis file: (contact the Doctoral School to confirm you have a valid doctoral degree and to get the link to the thesis)
  Programme: DOCTORAL DEGREE IN SIGNAL THEORY AND COMMUNICATIONS
  Department: Department of Signal Theory and Communications (TSC)
  Mode: Normal
  Deposit date: 24/02/2026
  Reading date: 15/05/2026
  Reading time: 14:00
  Reading place: ETSETB B3 Teleensenyament, Campus Nord UPC, Barcelona
  Thesis director: BLANCO BOTANA, LUIS | ZEYDAN, ENGIN
  Thesis abstract: This thesis presents an efficient and sustainable AI-driven resource management framework for next-generation Open Radio Access Network (O-RAN) in the context of the emerging 6G era. The framework operates in a cloud-native 6G environment and translates predictive intelligence into reliable, energy-aware orchestration. It combines advanced predictive modeling with optimization-based control to address challenges, like stochastic demand, multitenancy, and computational complexity in O-RAN. The predictive forecasting architecture is the core of the framework, quantifying uncertainty and interdependencies among network resources across multiple tenants. Probabilistic forecasting models generate distributions of future resource demands, enabling service providers to perform more informed and risk-aware resource orchestration in complex multitenant environments.Initially, the framework considered in this thesis considers univariate probabilistic estimators, including Simple-Feed-Forward (SFF), Deep Autoregressive Recurrent network (DeepAR), and Transformers, to predict individual resource demands and support effective provisioning in O-RAN. These models deliver efficient, agile, and uncertainty-aware resource predictions, which are integrated into a novel percentile-based orchestration strategy, Dynamic Percentile Adjustment Approach (DYNp). The proposed method dynamically adjusts the percentage to ensure efficient resource utilization in O-RAN systems. Selecting an appropriate percentile is critical for balancing resource waste and service reliability. However, univariate probabilistic estimates do not capture cross-resource interdependencies, leading to suboptimal decision-making. To address this limitation, the framework incorporates state-of-the-art multivariate probabilistic forecasting models such as Gaussian Process Vector Autoregression (GPVAR) and the Temporal Fusion Transformer (TFT). They jointly process multiple time series and provide robust estimates of future resource demands. These models effectively learn complex interdependencies among different resources and key parameter indicators across network slices and tenants. Furthermore, we have evaluated how low-rank approximation in GPVAR estimator enhances scalability and robustness by reducing the algorithm's training time. One of the main goals of this thesis is to achieve energy efficiency and effective resource management and sharing. By considering predictive intelligence together with power consumption, the proposed techniques proactively optimize the activation and deactivation of radio resources or radio units. This strategy significantly reduces power consumption while maintaining user experience and adhering to Service Level Agreement (SLA) guarantees. Furthermore, another relevant contribution of the thesis is the extension of the traditional cellular O-RAN architecture to include Cell-Free massive Multiple-Input and Multiple-Output (CF-mMIMO) networks, reflecting the architectural evolution for beyond 6G systems. This provides a scalable approach to ultra-dense, energy-efficient O-RAN deployments. Finally, the algorithm tools considered in the dissertation are implemented as modular applications to facilitate deployment across O-RAN. The cloud-native implementation of the forecasting and orchestration pipeline is a notable achievement. Each module has been containerized using Docker, and its functionality is exposed via Representational State Transfer (REST) APIs, such as Swagger. This enables the pipeline to operate as microservices, supporting flexible deployment, scalable execution and seamless integration within O-RAN. The thesis establishes a mathematical and architectural foundation for deploying AI-driven, sustainable, and energy-optimized O-RAN with uncertainty adaptation. It provides a basis for realizing intelligent, autonomous, and stable 6G networks and supports future research and industrial implementation of AI-powered O-RAN ecosystems.