Author: LIU, YUMING
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 ELECTRICAL ENGINEERING
Department: (DEE)
Mode: Article-based thesis
Deposit date: 06/11/2023
Reading date: pending
Reading time: pending
Reading place: pending
Thesis director: RIBA RUIZ, JORDI ROGER | MORENO EGUILAZ, JUAN MANUEL
Committee:
     PRESIDENT: BERGAS JANE, JOAN GABRIEL
     SECRETARI: CAPELLI, FRANCESCA
     VOCAL: KADECHKAR, AKASH
Thesis abstract: In recent years, there has been a significant increase in the growth of electricity demand. This electricity demand requires retrofitting of lines or exploiting the maximum capacity of existing power lines. In addition, substation connectors are the critical components of power systems since the failure of substation connectors can lead to serve power outages and significantly affect the power transmission efficiency. Therefore, it is vital to have real-time information on key elements of electrical systems, such as connectors and conductors, in order to ensure the reliability and efficiency of power transmission systems. To this end, the Smartconnector project aims to combine ICT (Information and Communication Technologies), IoT (Internet of Things) and data-driven technologies to estimate the state of health of substation connectors and take advantage of the maximum capacity of lines. Over the past few years, several research projects have been carried out to develop a smart high voltage connectors prototype, Smartconnector, in order to collect and wirelessly transmit information from power connectors in real time. Moreover, a prediction model has been proposed to utilize the collected data to separately predict the remaining useful life of connectors. However, improvements, experimental validation and field application are still needed to verify the reliability and feasibility of the Smartconnector prototype. In addition, field application of the IoT device is required for both connectors and conductors.This thesis is dedicated to the development, improvement, and experimental validation of the IoT prototype and the extension of its application to further increase the efficiency of power transmission systems. It is divided into two cores, which include the optimization of the Smartconnector prototype, and the extension of its application. This thesis improves the power management system, which helps to prolong the lifetime of the device. This improvement is based on thermal energy harvesting together with the energy balance strategy. Also, it develops statistical filtering algorithms for data processing. The proposed algorithms are finally implemented on the embedded system of the Smartconnector device, ensuring the accuracy of the continuous measurements. This thesis also focuses on the application of the Smartconnector for the dynamic monitoring of the line capacity considering weather conditions. In conclusion, this thesis aims to provide improvements and developments for the Smartconnector, as well as to open its application to other fields.

Author: PARAREDA ORIOL, SERGI
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 MATERIALS SCIENCE AND ENGINEERING
Department: (CEM)
Mode: Article-based thesis
Deposit date: 02/10/2023
Reading date: 04/12/2023
Reading time: 11:00
Reading place: EEBE, Sala Polivalent de l'Edifici A, planta baixa, Campus Diagonal-Besòs
Thesis director: MATEO GARCIA, ANTONIO MANUEL | CASELLAS PADRO, DANIEL
Committee:
     PRESIDENT: BUSQUETS MATAIX, DAVID JERONIMO
     SECRETARI: LLANES PITARCH, LUIS MIGUEL
     VOCAL: CARRERAS BLASCO, LAURA
Thesis abstract: Fatigue failures in materials have been studied for centuries, with early pioneers like August Wöhler noting that repeated loading, even below the static strength of the material, could lead to structural deterioration. Despite decades of research by notable authors, fatigue remains a complex and challenging issue, accounting for most service failures in metallic and composite structures. Therefore, designing structures to withstand cyclic loads without compromising integrity is crucial. But this process requires conducting numerous fatigue tests to define the appropriate design stress levels for each material and condition.However, determining the fatigue behaviour of metallic alloys and composites through standardised testing methods is often costly and time-consuming. While various techniques have been proposed to expedite testing and enhance the optimisation of materials and components for fatigue resistance, they have not gained wide industry adoption due to limitations in equipment or complex data treatment. Thus, there is an industry need for a testing method that rapidly determines material fatigue resistance, especially in the automotive sector where new designs and developments require results in a short time.To address this challenge, the development of a new testing method for characterising the fatigue resistance of metallic alloys and composites has become essential, as current solutions like ultra-high frequency testing machines or the rapid testing methods using temperature variations are not universal solutions for all these materials. In this thesis, a novel fatigue testing method, named the stiffness method, is introduced to rapidly assess the fatigue resistance of both metallic and composite materials with minimal specimens and in a short timeframe. This approach involves monitoring fatigue damage using different variables, such as inelastic strain in metallic alloys and compliance in composites. These measurements overcome the limitations of other methods by using common extensometers like digital image correlation techniques and contact extensometers.The results obtained through the stiffness method are not only convincing but also more accessible for interpretation and discussion compared to other monitoring techniques, such as temperature dissipation. The effectiveness of this approach has been validated across nineteen metallic materials, including titanium and aluminium alloys, carbon steels, stainless steels, and one carbon-fibre composite. The estimated fatigue limit and high cycle fatigue curve (S-N curve) obtained through the stiffness method align excellently with values derived from standardised tests. This underscores the method as a powerful and efficient tool for swiftly assessing the fatigue behaviour of both metallic alloys and composite materials.Furthermore, this research investigates the fatigue reduction observed in high-strength steels when surface defects are introduced during manufacturing processes such as shearing. This reduction in fatigue resistance is explained by the fatigue notch sensitivity of the material. The results establish a robust correlation between fracture toughness, assessed within the framework of fracture mechanics, and fatigue notch sensitivity in high-strength steels. As a result, fracture toughness coupled with the stiffness method can be a valuable toolkit for selecting materials with superior fatigue resistance.In summary, this work presents an innovative and efficient approach to evaluate the fatigue behaviour of metallic alloys and composite materials, offering significant advantages in terms of time and resource savings. Additionally, it introduces fracture toughness as a valuable indicator for material selection in high-strength steel applications, ultimately contributing to improved fatigue performance.

Author: PICCARDO, STEFANO
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 APPLIED MATHEMATICS
Department: (DECA)
Mode: Change of supervisor
Deposit date: 29/09/2023
Reading date: 04/12/2023
Reading time: 09:00
Reading place: Sala d'Actes FME, Edifici U, Campus Sud.
Thesis director: ERN, ALEXANDRE | HUERTA CEREZUELA, ANTONIO | GIACOMINI, MATTEO
Committee:
     PRESIDENT: PRUD\'HOMME, CHRISTOPHE
     SECRETARI: DIEZ MEJIA, PEDRO
     VOCAL: KRELL, STELLA
Thesis abstract: We consider two-fluid problems involving immiscible Stokes fluids in contact and with surface tension at their interface. We develop, analyze, and compare two approaches for space discretization, the Hybrid High-Order (HHO) and the Hybridizable Discontinuous Galerkin (HDG) methods, both combined with a geometrically unfitted approach to handle fluid interfaces. Both methods do not place discrete unknowns on the interface, and for both methods, we observe optimal error decay and condition number growth. The first part of the Thesis deals with the unfitted HHO solver, where the interface is described by a level-set function and discretized using isoparametric finite elements. We use the methodology to study the equilibrium problem with a shear flow imposed at infinity, and we investigate the relationship between the eccentricity of the ellipse-shaped interface at equilibrium and the ratio of shear to surface tension force Then, we explore settings where the shape of the interface is unknown. We devise a fixed-point iterative procedure that alternates a transport step for the level-set function with the unfitted HHO solver with frozen interface. In the second part of the Thesis, the unfitted HDG solver is combined with a NURBS description of the interface and of the external boundary, enabling a seamless transition from CAD geometries. We use this solver to study one- and two-fluid Stokes problems with applications in microfluidics related to microfluidic mixers and to emulsion flows in porous media. Both applications employ a polynomial-adaptivity error estimator, delivering results with at least two significant digits. Finally, in the third part of the Thesis, the HDG and HHO methods are compared for one- and two-fluid Stokes problems. In the simplest setting, we prove that the HHO and HDG methods can differ only in the choice of the approximation spaces and of the stabilization operator.

Author: YÁÑEZ ALVARADO, CARLOS RENÉ
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 OPTICAL ENGINEERING
Department: Department of Optics and Optometry (OO)
Mode: Article-based thesis
Deposit date: 06/11/2023
Reading date: pending
Reading time: pending
Reading place: pending
Thesis director: ROYO ROYO, SANTIAGO
Committee:
     PRESIDENT: ESQUIVIAS MOSCARDÓ, IGNACIO
     SECRETARI: LAZARO VILLA, JOSE ANTONIO
     VOCAL: DABBICCO, MAURIZIO
Thesis abstract: In this doctoral thesis, significant strides have been made in the realm of biofluid flow sensing through the innovative integration of Self-Mixing Interferometry (SMI) with confocal microscopy principles, resulting in a revolutionary technology known as Confocal Self-Mixing Interferometry (CSMI). This groundbreaking approach addresses the critical need for accurate, real-time, non-invasive, and cost-effective methods for evaluating biofluid dynamics within the human body.CSMI demonstrates its exceptional ability to distinguish flow rates across various sample depths within a laboratory setting, opening doors multifaceted applications across diverse domains.By employing a highly scattering fluid simulating blood, the system successfully measured flow rates ranging from 0.2 to 1.6mL/min within microcapillaries. Importantly, it achieved a maximum velocity measurement of approximately 307 mm/s, well within the range of average maximum blood flow velocities in human vessels. The confocal measurements exhibited clear Doppler frequency peaks at the microcapillary center across all pumping rate values, overcoming the limitations faced by non-confocal measurements for higher flow rates.This research produced results that provide insight into depth sectioning. A two-step process, involving lateral scanning and depth sectioning, was employed to reconstruct the velocity profile of the entire microcapillary at a pumping rate of 0.5 mL/min. While some challenges were encountered near the microcapillary walls due to low-frequency interference, the study effectively established the alignment of theoretical Hagen-Poiseuille profiles with CSMI measurements, validating the system's capabilities.Moreover, this research presents a significant achievement in terms of Signal-to-Noise Ratio (SNR) improvement. By implementing a stationary reflector in a secondary optical path, it was possible to maintain an adequate SNR in SMI configurations with reduced optical output power. This advancement holds promising applications in non-invasive measurements of individual living cells, microorganisms, or micro-particles, spanning various fields, including medicine, entomology, botany, cytopathology, and other health sciences.This research reaffirms the potential of CSMI as a transformative tool for diagnosing and forecasting diseases associated with abnormalities in bodily fluids. It not only excels in intrinsically-induced disruptions in blood flow, such as angiogenesis and thrombosis, but also in scenarios involving extrinsic factors like traumas and burns. The versatility of this technology can be extended to cell biology applications and point-of-care (PoC) analysis.

Author: ARIAS DUART, ANNA
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 ARTIFICIAL INTELLIGENCE
Department: Department of Computer Science (CS)
Mode: Normal
Deposit date: 18/10/2023
Reading date: 11/12/2023
Reading time: 11:00
Reading place: Sala de Juntes de la FIB, Edifici B6, planta 1
Thesis director: CORTÉS GARCÍA, CLAUDIO ULISES | GARCÍA GASULLA, DARIO
Committee:
     PRESIDENT: GIANNOTTI, FOSCA
     SECRETARI: BÉJAR ALONSO, JAVIER
     VOCAL: PADGET, JULIAN
Thesis abstract: Machine learning and, more specifically, deep learning applications have grown in number in recent years. These intelligent systems have shown remarkable performance across various domains, including sensitive areas like medicine and justice. Nevertheless, these models remain opaque, and we need a complete understanding of their internal process. Therefore, the deployment of these black box models can pose risks. Firstly, it might not comply with the current legislation. Secondly, it may lead to severe consequences. Let us consider a scenario in which a model used in a medical application is gender-biased, yielding distinct predictions depending on a person's gender. This fact would perpetuate discrimination against certain parts of the population and exacerbate existing inequalities.To better understand the model's behaviour, enabling the detection and mitigation of potential biases and ultimately achieving more trustworthy models, the eXplainable AI (XAI) field is an active research domain which is growing and receiving increasing attention. Various approaches have been proposed in the literature. Nevertheless, the most widely used are the post-hoc methods. These approaches can be applied once the model is trained, thus preserving the model's original performance. By employing these post-hoc explainability methods to gain insights into the model and identify biases within the datasets and models, we realized that two other biases arise: XAI and human biases.While different XAI methods exist, assessing their faithfulness becomes challenging due to the absence of a ground truth determining what the correct explanation is. The uncertainty regarding whether the explanation accurately reflects the model's behaviour can lead to what we refer to as XAI biases. Is the model biased or is it the explainability method that fails to reflect the model's behaviour? Human bias is another of the biases that emerge when applying these explainability methods. How we show these explanations to humans can be misleading or lead to incorrect conclusions. This can be due to confirmation or automation biases. In addition, when domain experts are asked to review all the explanations, the process can be time-consuming and may lead experts to overlook potential biases in the data and models.The main goal of this thesis is to mitigate the influence of these two new sources of biases (i.e., XAI and human) when explainability is used to detect biases in datasets and models. First, we focus on mitigating XAI biases. To do so, we propose a methodology to assess the reliability of XAI methods. Although our primary goal was to use this methodology within the computer vision discipline, we also demonstrated its applicability in other domains, such as the natural language processing field. After selecting the most reliable XAI method according to our proposed approach, we focus on mitigating human biases. With this objective in mind, we present potential methodologies to semi-automate the detection of data/model biases, thereby reducing the noise introduced by humans. Adopting this approach limits the domain expert's intervention to the final step, in which experts assess whether the biases found are harmful or harmless.