Doctoral School

Reading date: 08/05/2024

• ALONSO ALONSO, JESUS: Dynamic Terrain Modeling
  
  Author: ALONSO ALONSO, JESUS
  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 COMPUTING
  Department: Department of Computer Science (CS)
  Mode: Normal
  Deposit date: 18/03/2024
  Reading date: 08/05/2024
  Reading time: 12:00
  Reading place: Sala de Teleensenyament del 'ETSETB, edifici B3, Barcelona Diagonal NORD
  Thesis director: JOAN ARINYO, ROBERT
  Committee:
       PRESIDENT: CHOVER, MIGUEL
       SECRETARI: ARGUDO MEDRANO, OSCAR
       VOCAL: BOSCH GELI, CARLES
  Thesis abstract: This work explores terrain modelling techniques that provide a comprehensive experience in terms of graphical representation, physics interaction and dynamic updates in real-time. In particular, our focus revolves around the creation of a system able to: 1) capture any possible feature we find on terrains, 2) maintain an accurate level of detail, 3) offer rendering and navigation in real-time, 4) include the option of performing dynamic updates in real-time, and 5) support physical interactions of entities also in real-time.Once previous models from the literature are reviewed, two models are proposed that take a digital elevation model as the base structure. The former follows a strategy in which we mimic the geotectonic events we find in nature. The latter uses a sculpting approach with convex polyhedra as a carving tool. To this end, several works are presented.While the first option introduces some gains with limits, the second option is a proposal that accomplishes the five required constraints. On the one hand, it can model tunnels, caves and overhangs, and terrain features can be captured with pixel-perfect accuracy. On the other hand, it is not demanding regarding processing and storage requirements and offers scalability. Finally, rendering, physics and dynamic updates can be performed in real-time.As a result, this work represents a significant contribution, offering an integrated solution capable of addressing the most challenging aspects of dynamic terrains. Our approach introduces a novel terrain model comprising diverse data structures and a suite of algorithms designed to capture a wide range of terrain formations accurately. A scene composed of tens of millions of triangles can be continuously updated to the extent of simulating a completely devastated terrain, rendered, and subjected to real-time physics computations involving tens of thousands of physical entities. The proposed model holds great potential for computer graphic applications, particularly in scenarios such as simulators and games, where dynamic landscapes play a paramount role.
  

• FERRIOL GALMÉS, MIQUEL: Network modeling using graph neural networks
  
  Author: FERRIOL GALMÉS, MIQUEL
  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 COMPUTER ARCHITECTURE
  Department: (DAC)
  Mode: Normal
  Deposit date: 10/04/2024
  Reading date: pending
  Reading time: pending
  Reading place: pending
  Thesis director: CABELLOS APARICIO, ALBERTO | BARLET ROS, PERE
  Committee:
       PRESIDENT: PESCAPÈ, ANTONIO
       SECRETARI: ARIAS VICENTE, MARTA
       VOCAL: RÉTVÁRI, GÁBOR
  Thesis abstract: Network modeling is central to the field of computer networks. Models are useful in researching new protocols and mechanisms, allowing administrators to estimate their performance before their actual deployment in production networks. Network models also help to find optimal network configurations, without the need to test them in production networks. Arguably, the most prevalent way to build these network models is through the use of discrete event simulation (DES) methodologies which provide excellent accuracy. State-of-the-art network simulators include a wide range of network, transport, and routing protocols, and are able to simulate realistic scenarios. However, this comes at a very high computational cost that depends linearly on the number of packets being simulated. As a result, they are impractical in scenarios with realistic traffic volumes or large topologies. In addition, and because they are computationally expensive, they do not work well in real-time scenarios.Another network modeling alternative is Queuing Theory (QT) where networks are represented as inter-connected queues that are evaluated analytically. While QT solves the main limitation of DES, it imposes strong assumptions on the packet arrival process, which typically do not hold in real networks.In this context, Machine Learning (ML) has recently emerged as a practical solution to achieve data-driven models that can learn complex traffic models while being extremely accurate and fast. More specifically, Graph Neural Networks (GNNs) have emerged as an excellent tool for modeling graph-structured data showing outstanding accuracy when applied to computer networks. However, some challenges still persist:1. Queues and Scheduling Policies: Modeling queues, scheduling policies, and Quality-of-Service (QoS) mappings within GNN architectures poses another challenge, as these elements are fundamental to network behavior.2. Traffic Models: Accurately modeling realistic traffic patterns, which exhibit strong autocorrelation and heavy tails, remains a challenge for GNN-based solutions.3. Training and Generalization: ML models, including GNNs, require representative training data that covers diverse network operational scenarios. Creating such datasets from real production networks is unfeasible, necessitating controlled testbeds. The challenge lies in designing GNNs capable of accurate estimation in unseen networks, encompassing different topologies, traffic, and configurations.4. Generalization to Larger Networks: Real-world networks are often significantly larger than testbeds. Scaling GNNs to handle networks with hundreds or thousands of nodes is a pressing challenge, one that requires leveraging domain-specific network knowledge and novel architectural approaches.This dissertation represents a step forward in harnessing Graph Neural Networks (GNN models) for network modeling, by proposing a new GNN-based architecture with a focus on addressing these critical challenges while being fast and accurate.
  

Reading date: 09/05/2024

• MESA GÓMEZ, ADRIANA MARÍA: Analysis and modelling of natech accidents originated by strong winds
  
  Author: MESA GÓMEZ, ADRIANA MARÍA
  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 CHEMICAL PROCESS ENGINEERING
  Department: Department of Chemical Engineering (EQ)
  Mode: Change of supervisor
  Deposit date: 11/04/2024
  Reading date: pending
  Reading time: pending
  Reading place: pending
  Thesis director: CASAL FABREGA, JOAQUIM | MUÑOZ GIRALDO, FELIPE | SANCHEZ SILVA, EDGAR MAURICIO
  Committee:
       PRESIDENT: PLANAS CUCHI, EULALIA
       SECRETARI: PALACIOS ROSAS, ADRIANA
       VOCAL: DEMICHELA, MICAELA
  Thesis abstract: In recent decades, there has been an increase in the frequency of natural events, coinciding with the simultaneous development of industrial activities in many countries. Consequently, the frequency of Natech accidents, which are technological disasters triggered by natural hazards, has also risen. This trend has spurred researchers to explore new risk analysis methods to prevent and mitigate potential damage to populations, the environment, and industrial facilities. There is a growing awareness in the literature about the impact of natural events, particularly when they occur concurrently, cascade, or accumulate over time.This thesis proposes a research initiative to conduct a risk assessment that includes the Natech risk associated with strong winds. The primary objective is to develop a methodology for analyzing Natech risk in storage units in coastal zones that are particularly vulnerable to extreme weather events.Firstly, the thesis introduces the integration of natural events, specifically strong winds, into a quantitative Natech risk analysis methodology. This integration represents a significant advancement in assessing the potential impacts of technological accidents triggered by natural events. By incorporating strong winds as a hazard, the methodology offers a more comprehensive approach to evaluating the vulnerability of industrial facilities, especially storage tanks, to natural-technological events. This integration enables stakeholders to better understand and quantify the risks posed by Natech events involving strong winds, facilitating the implementation of targeted mitigation measures and enhancing preparedness. Ultimately, it contributes to improving the resilience of industrial facilities and surrounding communities to the risks posed by natural events.Secondly, the thesis describes the development of two models for environmental and socioeconomic risk assessment, respectively. These models provide a comprehensive framework for evaluating the potential environmental and socioeconomic impacts of Natech events, thereby enhancing the understanding of the overall risk landscape. By incorporating previously overlooked vulnerable elements, such as cultural heritage sites, sensitive environmental areas, water catchment sites, and so on, the models offer a more holistic perspective on Natech risks, ensuring that mitigation strategies can protect not only human safety and infrastructure, but also socioeconomic and environmental assets.Thirdly, the thesis outlines the development of a computational tool designed to facilitate the implementation of these models. This tool streamlines the risk assessment process, enabling stakeholders to analyze and manage Natech risks efficiently.Overall, the generation of these models and the accompanying computational tool represents a significant advancement in Natech risk management. By integrating environmental and socioeconomic considerations into the risk assessment process, these models provide a more robust foundation for decision-making and emergency preparedness, ultimately contributing to the resilience of communities and ecosystems in the face of Natech events. Finally, the methodology is applied in a case study to verify its applicability
  

• RALLIS, KONSTANTINOS: Novel Nanoelectronic Circuits and Systems
  
  Author: RALLIS, KONSTANTINOS
  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 ELECTRONIC ENGINEERING
  Department: Department of Electronic Engineering (EEL)
  Mode: Normal
  Deposit date: 13/03/2024
  Reading date: 09/05/2024
  Reading time: 11:00
  Reading place: ETSETB - Multimedia room Building B3 at Campus Nord UPC
  Thesis director: RUBIO SOLA, JOSE ANTONIO | SIRAKOULIS, GEORGIOS
  Committee:
       PRESIDENT: JIMÉNEZ JIMÉNEZ, DAVID
       SECRETARI: ABADAL CAVALLÉ, SERGI
       VOCAL: CUCU LAURENCIO, NICOLETA
  Thesis abstract: Lately, in the rise of the era of 2D materials, Graphene is one of the materials that has been extensively investigated for its possible integration in computing devices and thus computing circuits. This is mainly attributed to its very wide set of appealing properties. The combination of its electronic properties with others, such as mechanical, optical or chemical properties, can extend the range of use of computing devices and lead to groundbreaking interdisciplinary applications. However, this integration of Graphene in switching and computing elements is not easy. In this dissertation, the Non-Equilibrium Green's Function method (NEGF), along with the Tight Binding Hamiltonians, are fitted on experimental data from fabricated Graphene devices. Although as a computational method, NEGF is appropriate for the simulation of small-scale devices in the regime of nanometers, its ability to be efficiently expanded for the description of larger devices is presented. The aforementioned electronic properties of the material are highly related to its shape and structure. Consequently, it requires a very precise fabrication method that can guarantee the minimum presence of defects on the Graphene grid. For that reason, the effect of defects is deeply investigated. The NEGF method is further enhanced in order to be able to incorporate lattice defects. The most common lattice defects are included, meaning the single and double vacancy. A framework has thus been created, so that for the first time the user can select areas of interest on the grid, in which the defects will be concentrated. Those concentrations can also be variable. Moreover, an extensive study is conducted on defective grids with different concentrations of single and double vacancies. The investigated grids are non-rectangular and have regions with different widths. The effect of those vacancies on the electronic properties of Graphene is investigated, and more specifically their effect on the conductance and the energy gap of the device, as well as the effect on circuit-centered characteristics such as the leakage current and ON/OFF current ratio. Having a functional, robust, versatile, and accurate model, the focus of this thesis is extended to the level of circuits. The model is imported into SPICE through Verilog-A. In this part, the thesis emphasizes on the investigation of the switching capabilities of L-shaped Graphene Nanoribbons (GNRs). These structures have been proven to be able to operate as switches, without the use of a back gate, and here, the properties that are dependent on their dimensions are explored and optimized for the first time. The optimized structures are then used for the realization of a set of computing topologies. Initially, a novel area-optimized 2-branch comb-shaped topology is introduced for the realization of a universal computing set that consists of an AND, OR, NOT gate, and a Buffer. All these logic operations can be mapped on the same topology through appropriate biasing. Then, an extension of this, the 3-branch comb-shaped topology is proposed, which is able to operate as a 2-XOR, 3-XOR and 3-MAJ gate. The circuit of a 1-bit full adder, is also presented. For the evaluation of the performance of the topologies, several related metrics are employed such as the area, delay, power dissipation and the power-delay product. The operation of these topologies relies of the principles of Pass Transistor Logic (PTL) and reconfigurable computing. Finally, in an attempt to go beyond the conventional Boolean logic, the compliance of Graphene with Multi-Valued Logic (MVL) circuits and applications is investigated. The ability of a Graphene Quantum Point Contact (G-QPC) device to encode the digits of the radix-4 numeral system is presented and as a proof of concept, the operation of an arbitrary radix-4 adder is explained.
  

• SAYOLS BAIXERAS, NARCÍS: Cognitive Robot Control Strategies for Complex Surgical Environments
  
  Author: SAYOLS BAIXERAS, NARCÍS
  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 BIOMEDICAL ENGINEERING
  Department: Department of Automatic Control (ESAII)
  Mode: Normal
  Deposit date: 11/04/2024
  Reading date: pending
  Reading time: pending
  Reading place: pending
  Thesis director: CASALS GELPI, ALICIA | HERNANSANZ PRATS, ALBERTO
  Committee:
       PRESIDENT: DALL\'ALBA, DIEGO
       SECRETARI: FRIGOLA BOURLON, MANEL
       VOCAL: AVILÉS RIVERO, ANGÉLICA
  Thesis abstract: This thesis aims to contribute to the development of robotics autonomy in complex tasks based on the cognitive control paradigm. Cognition is a multidisciplinary approach aimed to provide robotic systems with intelligent and autonomous behaviour that should learn and reason about how to respond in front of complex tasks and environments.Cognition involves aspects as perception, awareness, interpretation of human actions, learning, planning, anticipating and dynamic response to changes in the working conditions and in the interaction with humans. Autonomy is intended to partially substitute and/or complement the human faculties at the level of perception, analysis and execution. Increasing the level of autonomy of robots allows focusing the humans cognitive load on high level decisions and actions, in aspects where the human factor is essential: contextualisation of information, specific expertise, medical knowledge and complex decision-making among others. Furthermore, robots improve the properties of humans in certain aspects such as precision, repeatability, absence of fatigue or response efficiency in terms of time and accuracy.This thesis addresses different key aspects of robotic autonomy: perception, planning and dynamic execution of actions and, finally, the control structures required for efficient control and their integration in robotic systems.This thesis combines a global theoretical approach supported by practical applications based on the field of robot-assisted minimally invasive surgery. This field has been chosen for two main reasons: the social impact involved in the improvement of surgery and, secondly, because this field of application is highly demanding from both, human and robotic perspective.The experimental phases have focused on various surgical robotic. First, a teleoperated platform with a single robot has been used aimed at minimally invasive fetal surgery in which a cognitive system offers a certain level of autonomy to generate trajectories in collision-free spaces, increasing patient safety and decreasing the cognitive load of surgeons in navigation and interaction tasks within the intra-uterine region. Second, a multi-robot architecture to execute auxiliary actions in a human-robot cooperative system: the main surgeon performs the surgical actions while the auxiliary robots perform, autonomously, auxiliary surgical tasks. With this configuration the experimentation focuses on minimally invasive radical prostatectomy surgery.Thus, the thesis addresses the perception of the anatomical environment, considering the limitations of data acquisition in terms of quality and quantity, as well as the absence of anatomical markers. The next topic that the thesis addresses is the dynamic planning of actions. Different application paradigms have been studied, such as direct human-robot interaction using haptic guidance, movement planning in pseudo-structured environments and, active planning and control in dynamic environments. These proposed environments respond to different surgical scenarios within minimally invasive techniques.Finally, cognitive control applied to robotic platforms is addressed. The followed approach is based on the multi-level decomposition of complex tasks (e.g. surgical procedure) defining all potential states and transitions. This decomposition translates into the use of deterministic and robust control structures that restrict falling into uncontrollable or unexpected situations that put at risk, in the application case, the patient, the surgeons or the auxiliary personnel.Control structures also consider human-robot interaction, robots coordination and cooperation, interaction with the work environment and restrictions imposed by surgery and patient safety.The integration of all these modules: perception, planning and cognitive control, demonstrates the advances achieved in cognitive robotics and their applicability towards a more autonomous robotic surgery.