Abstract: Claystone degradation under the effect of freezing and thaw cycles represents a key aspect in the production of erodible material in mountain ranges. Erosion processes have numerous potential consequences such as dam filling, increment in slope instability, increase of badlands, among others. Due to the possible effects related to erosion, numerous research processes have been carried out following primarily experimental approaches based on both soil physical characteristics and mechanical properties. However, these have been based on a macroscopic scale. This document presents the development of a research process based on the implementation of numerical models aimed at representing the processes involved in freezing and thawing phenomena on porous media and their effect on potential changes in soil microfabric. A calibration and validation process of the proposed numerical models was carried out based on experimental measurements of the thermal behavior of soil samples subjected to freezing and thawing cycles. Using the numerical relationships found, a methodology for determining the soil water retention curve (SWRC) was established (as a first approximation) based on a quick and relatively simple freezing and thawing test under undrained conditions. Simultaneously, and following the main objective of the research, the microfabric state during cyclic freezing and thawing was established following the relationship between SWRC and pore size distribution (PSD), as well as the possible hysteresis between freezing and thawing processes. The significant effect of these thermodynamic processes on the state of the terrain, was identified particularly in regard to increased erodibility. One possible cause for particle detachment necessary for erosion processes to occur lies in the development of macroporosity along with the existence of gaps between pore families in the PSD. This behavior was observed through the developed numerical processes. The combined effect of freezing and thawing cycles leads to an increase in soil erodibility. Finally, though a constitutive model-based approach, the cryogenic suction paths associated with freezing and thawing processes under the modification of the soil’s microstructure were established. A final perspective to soil's mechanical degradation is presented in terms of the hydraulic work associated with the cryogenic suction paths developed within the soil mass. Both the progressive evolution of the microstructure (through PSD) and the values of hydraulic work associated with each cycle show an attenuation in the incremental change as the cycles continue. Due to the numerical nature of the approximations and the trends found, further cyclical laboratory tests should be carried out. However, these represent a novel approach to the degradation phenomenon as well as the effects associated with the soil water retention curve.

Abstract: This thesis proposes an update to the NS-047 technical standard of the aqueduct and sewerage company of Bogota for storm drain design, introducing the comprehensive and modern UPC methodology. The research addresses the limitations of the current standard, emphasizing the importance of critical parameters such as flow depth and velocities for pedestrian and vehicle safety during storms. By considering a wide range of hydrologic and hydraulic factors, the UPC method provides a realistic representation of stormwater behavior and enables better system understanding and optimization. The adoption of UPC is expected to enhance stormwater management, promoting safety and sustainability in urban flood control. The application of the UPC methodology to a real-life case study, the CASU road project in Bogotá, demonstrates its effectiveness in optimizing storm drain design. The analysis identified areas for improvement, resulting in a revised approach with strategically located storm drains that effectively controlled runoff while reducing the number of structures. This optimization not only saves costs but also minimizes risks for road users. The comparison between EAAB and UPC methodologies highlights the superiority of the UPC method in terms of its comprehensive approach, graphical outputs, and ability to incorporate multiple roadways. To further improve storm drains designs, experimentation is recommended to determine capture coefficients for different structures, especially in low points critical for drainage efficiency and flood prevention. Obtaining accurate capture coefficients through real-world experiments will enhance the overall stormwater management performance and inform design decisions. By complementing the EAAB standard with the UPC methodology and conducting experimentation for capture coefficients, this research aims to provide a clearer, practical, and more effective storm drain design approach for infrastructure projects in Bogotá and other areas in Colombia.

Abstract: The need for measurable data from physical assets to actively feed a living Digital Twin (DT) is paramount. The requirements and needs that the gathered data should fulfill in order to be practically implemented in the stream data pipeline are heterogeneous, some of them general and other case-specific. This article summarizes a set of identified challenges and requirements for a seamless infusion of well-established Structural Health Monitoring (SHM) systems within DT platforms, without the objective of solve all of them. This identification is performed based on a review of traditional SHM systems with a vast array of information sources as well as on the review of techniques for the systematic digitalization of existing assets. On the other hand, ten real demo cases belonging to Ashvin, an H2020 Research and Innovation project, are providing real world testing beds for an active development of SHM infused in DT systems. Multiple information sources are studied in those sites, which also enriches with more realism, the identification of requirements and challenges presented herein. These assets provide a perspective to researchers about practical implications of these needs.

Abstract: The forthcoming industrial environments will require a high level of automation to be flexible. In this sense, Industry 4.0 encourages the integration of new autonomous systems digitally interconnected that can interact with humans. This thesis is developed at CIM 4.0 in the frame of the FIXIT project aiming to produce an autonomous system integrated by a UAV and a UGV. As a result, a combined system that complies with the Industry 4.0 requirements is created and mainly serves an operator by performing inspection tasks. This work is focused on redesigning the UAV airframe. Normally these structures are manufactured with some conventional techniques like molding which limits building complex structures and presents high costs in most cases. In this sense, the scope is to create a personalized structure at a lower price than the standard airframes by implementing one of the industry 4.0 pillars: additive manufacturing. The topology optimization method is integrated into the design process to create a mass-customized structure with optimum structural properties. The role of Additive manufacturing in this work is crucial to obtain an innovative, customized system with an optimum cost considering the prices on the market. The thesis is performed in different phases, starting from the definition of the UAV configuration and the propulsion system. Then a multi-material comparison and the analysis of different airframe designs are performed considering the cost and structural performance. Following this is the detailed design, in which other necessary features like the landing system, devices’ supports, and protection structures are designed. Finally, is performed the manufacturing phase in which a functional prototype is printed through FDM technology. The result is a lightweight customized airframe with good mechanical properties and optimum cost that is well integrated with the electronic components and landing system, enabling its autonomous performance characterized by repeatability and effectiveness. Esta tesis se desarrolla en el marco del proyecto FIXIT en el Centro de Manufactura Integrada (CIM) 4.0 con el objetivo de crear un sistema autónomo integrado por un Vehículo Aéreo No Tripulado (UAV) y un Vehículo Terrestre No Tripulado (UGV). Resultado: Como resultado, se crea un sistema combinado que cumple con los requisitos de la Industria 4.0 y que principalmente sirve a un operador para realizar tareas de inspección. En este sentido, el objetivo de este proyecto de tesis es crear una estructura personalizada a un precio inferior al de las estructuras estándar mediante la implementación de uno de los pilares de la Industria 4.0: la fabricación aditiva. El método de optimización de la topología se integra en el proceso de diseño para crear una estructura personalizada con propiedades estructurales óptimas. El papel de la fabricación aditiva en este trabajo es crucial para obtener un sistema innovador y personalizado con un costo óptimo, teniendo en cuenta los precios del mercado. La tesis se realiza en diferentes fases, comenzando desde la definición de la configuración del UAV y el sistema de propulsión. Luego se realiza una comparación de materiales múltiples y se analizan diferentes diseños de la estructura del UAV teniendo en cuenta el costo y el rendimiento estructural. A continuación, se lleva a cabo el diseño detallado, en el que se diseñan otras características necesarias como el sistema de aterrizaje, los soportes de dispositivos y las estructuras de protección. Finalmente, se realiza la fase de fabricación en la que se imprime un prototipo funcional mediante la tecnología de Fabricación por Deposición Fundida (FDM). El resultado es una estructura personalizada y ligera con buenas propiedades mecánicas y un costo óptimo que se integra bien con los componentes electrónicos y el sistema de aterrizaje, lo que permite su funcionamiento autónomo caracterizado por la repetibilidad y la efectividad.

Abstract: There are too many reinforced concrete (RC) columns built before mid-1970s without sufficient transverse reinforcement. By now, we understand quite well the importance of transverse reinforcement in allowing a column to maintain its integrity under large displacement reversals in the nonlinear range of response. Poorly confined RC columns undergo a fast decay in resistance due to formation of criss-crossing inclined cracks, which can cause an abrupt failure or more gradual disintegration and trigger collapse of the structure. Those columns need to be strengthened to increase their drift capacity. Although there are several alternatives to retrofit RC columns, they often require specialized workmanship and equipment, and involved installation procedures. An easy-todesign and easyto-implement retrofit technique is examined here. It consists of external posttensioned clamps fastened around the column. Results of tests on full-scale RC columns furnished with the proposed clamps suggest the clamps can be effective in increasing column shear strength and drift capacity. Resumen: Muchas columnas de concreto reforzado construidas antes de mediados de la década de 1970 tienen insuficiente refuerzo transversal. Hace solo un par de décadas que se entendió la importancia del refuerzo transversal para permitir que una columna mantenga su integridad bajo grandes desplazamientos en el rango de respuesta no lineal. Las columnas de concreto reforzado mal confinadas sufren una rápida disminución de la resistencia debido a la formación de grietas, que pueden causar una falla abrupta o una desintegración más gradual y desencadenar el colapso de la estructura. Es necesario reforzar dichas columnas para aumentar su capacidad de deriva. Aunque existen varias alternativas para reforzar columnas de concreto reforzado, a menudo requieren mano de obra y equipo especializados, además de procedimientos de instalación complicados. Aquí se examina una técnica de reforzamiento fácil de diseñar e implementar. La técnica consta de abrazaderas presforzadas externas fijadas alrededor de la columna. Los resultados de laboratorio de las columnas de concreto reforzado a gran escala equipadas con las abrazaderas propuestas sugieren que las abrazaderas pueden ser efectivas para aumentar la resistencia al corte y la capacidad de deriva de la columna.

Abstract: Complex networks emerge as a natural framework to describe real-life phenomena involving a group of entities and their interactions, i.e., a social net-work. Furthermore, other problems involve a collection of networks, such as multi-layer, temporal, or brain networks (connectomes). With the increasing availability of graph-shaped data and associated meta-data, i.e., node or edge attributes, machine learning (ML) techniques on networks have risen in popularity. However, there are still several challenging issues to be addressed concerning feature engineering, graph representation, visualization, and general graph mining applications. This thesis contributes to bridging the gap between ML and complex net-works. We investigate dynamics on networks such as discrete and continuous- time random walks, in two directions. First, we develop a multi-scale anomaly detection algorithm on attributed networks. We exploit the link between graph signal processing and the Markov stability framework used in community detection, for spotting anomalous nodes in multiple contexts of the network. Second, we introduce a generalization of assortativity on networks operating on scalar and categorical node attributes. These coefficients turn out to be useful descriptors to build network fingerprints, so that we can perform supervised graph classification, i.e., predicting the toxicity of molecules or classifying Reddit discussion threads. Besides, we develop an unsupervised approach to learn graph embeddings from a collection of networks. By learning a non-linear mapping from input graphs to a lower-dimensional feature space, we obtain useful graphs representations used in graph visualization, clustering, and classification, e.g., predicting people’s gender based on their structural connectome. Finally, we propose a principled approach to identify stable biomarkers for schizophrenia diagnosis in the human connectome. From an ML perspective, the problem is stated as an embedded feature selection scheme for graph classification. By doing so, we identify the affected-core in schizophrenia, highlighting increasing performance when structural and functional neural features are combined. Resumen: Las redes complejas emergen como marco general en la descripción de problemas reales que incluyen entidades y sus interacciones, ex., redes sociales. Además, otros problemas incluyen colecciones de redes tales como redes multi nivel, redes temporales o redes del cerebro. Con el incremento de la disponibilidad de datos en forma de grafos y metadatos asociados, ex., atributos en nodos o ejes, técnicas de aprendizaje automático en redes han incrementado en popularidad. Sin embargo, múltiples desafíos siguen aún sin ser resueltos, relacionados con representación de grafos, ingeniería de características en grafos, visualización y aplicaciones generales de minería de datos sobre grafos. Esta tesis doctoral contribuye a cerrar la brecha entre dos campos del conocimiento emergentes: aprendizaje automático y ciencia de redes. En esta tesis se investigan sistemas dinámicos sobre redes en la extracción de características estructurales en redes y se introduce una generalización de asortatividad en redes. Dicha asortatividad permite definir descriptores de grafos útiles en la construcción de huellas digitales de redes utilizadas en problemas de aprendizaje supervisado en clasificación de grafos, ex., predicción de toxicidad en moléculas o clasificación de redes sociales Reddit. Además, esta tesis propone un nuevo método no supervisado para crear embeedings de grafos. Gracias a la estimación de una función no lineal, se puede mapear grafos enteros en espacios euclideanos, preservando la similaridad entre grafos de origen. Finalmente, se propone un algoritmo para el diagnóstico de esquizofrenia en el cerebro humano usando connectomes, destacando las regiones del cerebro afectadas por la enfermedad mostrando resultados de alta confiabilidad utilizando distintas representaciones neuronales.

Abstract: Infrastructure plays an essential role in the economy and has a positive impact on social stability. In this regard, bridges are a fundamental component of infrastructure; however, most of the bridges currently under operation were constructed a long time ago and have even passed their useful life. Certainly, Flaig & Lark (2000) noted that most of the current bridges in the UK were built between 1,950 and 1,970. Furthermore, Steve Kroft (2014) indicated that almost 70,000 bridges (one out of every nine bridges) in America are structurally deficient. Consequently, it is necessary to develop fast and efficient inspection techniques that meet current requirements, and that could potentially replace the highly- subjective visual observation methods. This research presents a critical comparison among different Machine-Learning algorithms that are currently used in image classification problems and applied them into the automated recognition of cracks in reinforced concrete images so that they can be used as an alternative for the existing inspection methods. Moreover, ensembles with different combinations of the evaluated models were tested as an incentive to improve the performance of individual models. In total, four algorithms that based their operation in mathematical formulations and 10 different architectures of Convolutional Neural Networks were included. The results showed that the Random Forest classifier outperformed the other mathematical classifiers and performed similarly to the best of the neural networks. In addition, it was found that the ensembled models achieved little or no improvement concerning the best of the classifiers included in the stack. Resumen: La infraestructura desempeña un papel esencial en la economía de los países y tiene un efecto positivo en la estabilidad social de las personas. A este respecto, los puentes son un componente fundamental de la infraestructura; sin embargo, la mayoría de los puentes actualmente en funcionamiento se construyeron hace mucho tiempo e incluso han pasado su vida útil. Ciertamente, Flaig y Lark (2000) observaron que la mayoría de los puentes actuales del Reino Unido se construyeron entre 1.950 y 1.970. Además, Steve Kroft (2014) indicó que casi 70.000 puentes (uno de cada nueve puentes) en USA son estructuralmente deficientes. Por consiguiente, es necesario desarrollar técnicas de inspección rápidas y eficientes que cumplan los requisitos actuales y que puedan sustituir a los métodos de inspección visual altamente subjetivos. Esta investigación presenta una comparación crítica entre los diferentes algoritmos de aprendizaje a máquina que se utilizan actualmente en los problemas de clasificación de imágenes y los aplica al reconocimiento automatizado de grietas en las imágenes de concreto reforzado para que puedan utilizarse como alternativa a los métodos de inspección existentes. Además, se ensayaron conjuntos con diferentes combinaciones de los algoritmos de aprendizaje evaluados como incentivo para mejorar el rendimiento de los modelos individuales. En total, se incluyeron cuatro algoritmos que basaban su funcionamiento en formulaciones matemáticas y 10 arquitecturas diferentes de Redes Neuronales Convolucionales. Los resultados mostraron que el clasificador Random Forest superó a los otros clasificadores matemáticos y tuvo un rendimiento similar a la mejor de las redes neuronales. Además, se encontró que los modelos ensamblados lograron poca o ninguna mejora en lo que respecta al mejor de los clasificadores incluidos en el conjunto.

We explore how we can use synthetically generated RGB-D training data from a near photo-realistic game engine to train modality-specific person detectors. We perform ablation studies on a challenging, real-world dataset which we recorded using a Kinect v2 RGB-D sensor in multiple warehouse environments. Through extensive use of domain randomization techniques, we synthesize a realistic and highly varied training set of challenging intralogistics scenarios as observed from a mobile robot, comprising persons in confined and cluttered indoor spaces. We then train the detector layers of a YOLOv3 model from scratch on our synthetic RGB and jet-encoded depth images. While for the RGB case, we still observe a domain gap of 6 points in mAP compared to a pretrained COCO model, results indicate that by exploiting simulation, an immense manual labeling effort needed to prepare large-scale datasets such as MS COCO might be unnecessary for the depth modality. We further find that filtering of highly occluded groundtruth boundin boxes during training, as well as modeling of time-of-flight sensor noise characteristics has a positive impact on model performance. We also provide an initial set of qualitative results on our real-world dataset.

An accurate understanding of the environment is key for a robot in order to execute tasks safe and efficiently. In the field of perception, after the introduction of deep learning, computer vision tasks have made big leaps, surpassing even the human inference capability. As a trade-off, big amounts of annotated data were required to train such algorithms. On its own, the collection of annotated dataset is a highly time consuming activity prone to human errors, setting a limit to the maximum achievable performance. In this sense, annotations (quality) and samples (quantity) bound the optimization of perception algorithms. One extra challenge encountered when training object detectors for robotics applications is that the sensor setup can be multi-modal, and vary significantly between robots. This work explores how to generate and use synthetic RGB-D training data from a near photo-realistic game engine to train modality-specific person detectors, and perform ablation studies on a challenging, real-world dataset recorded using a reference RGB-D sensor in different intralogistics environments. A virtual RGB-D camera was implemented, leveraging the underlying deferred rendering architecture. Multiple environments were tailored, exploring various data augmentation techniques and enabling the comparison between different types of synthetic data. Detection layers of a pre-trained object detector network have been trained from scratch for the RGB and depth modality, with the latter being transformed by applying a Jet-colormap. Compared to a pre-trained network, a domain gap of 5 mPA points was still present for RGB images. Meanwhile with synthetic (15k) and real (1.5k) depth images, it was already possible to train robust human detectors. Comparing simulation features against data preparation, filtering annotations had a major impact on performance than adopting an explicit time-of-flight sensor model.

Resumo Neste trabalho, é apresentado um modelo de programação estocástica (MPE) para resolver o problema de planejamento da expansão da rede de transmissão multiestágio (PERTM). O MPE proposto considera fontes de energia renovável intermitente (FERIs) e a alocação de sistemas de armazenamento de energia (SAEs). A natureza estocástica da demanda e das FERIs é modelada através de períodos representativos (PRs). Os PRs são períodos caraterís-ticos construídos a partir de séries temporais que permitem representar o comportamento do sistema elétrico de potência (SEP). Variações de curto prazo da demanda e das FERIs, a cor-relação horária entre essas variáveis aleatórias, e a sua caracterização de acordo com a regi-ão do SEP a que pertencem são incluídas nos PRs, fornecendo um aceitável esforço compu-tacional para resolver o problema de PERTM. Com base nos PRs, o MPE proposto é formu-lado como um modelo equivalente determinístico de programação linear inteira mista (PLIM) para um conjunto de cenários. O modelo de PLIM incorpora perdas de potência ativa e contingências N-1. Investimento em construção de linhas de transmissão (LTs) e novos SAEs são otimizados. Um método eficiente para reduzir o espaço de busca combinatório e lidar com a dimensionalidade do problema de PERTM foi desenvolvido. Os sistemas de tes-te: Garver de 6 barras e IEEE de 24 barras; e o sistema de transmissão norte-colombiano foram usados para validar o modelo proposto. Os resultados mostram que a otimização si-multânea de LTs e SAEs fornece melhores e mais baratos planos de expansão do que as abordagens tradicionais. A eficácia dos SAEs para deslocar blocos de energia elétrica ao longo dos PRs, adiar investimentos e aliviar o congestionamento da rede é evidenciada. Resumen: En este trabajo se presenta un modelo de programación estocástica (MPE) para resolver el problema de planeamiento multietapa de la expansión de la red de transmisión (PERTM). El MPE propuesto considera las fuentes de energía renovables intermitentes (FERIs) y la instalación de sistemas de almacenamiento de energía (SAEs). La naturaleza estocástica de la demanda y las FERIs se modela a través de períodos representativos (PRs). Los PRs son periodos característicos construidos a partir de series de tiempo que permiten representar el comportamiento del sistema eléctrico de potencia (SEP). Las variaciones de corto plazo en la demanda y las FERIs, la correlación horaria entre estas variables aleatorias y su caracterización según la región del SEP a la que pertenecen se incluyen en los PRs, proporcionando un esfuerzo computacional aceptable para resolver el problema PERTM. Sobre la base de los PRs, el MPE propuesto se formula como un modelo equivalente determinístico de programación lineal entera mixta (PLIM) para un conjunto de escenarios. El modelo PLIM incorpora pérdidas de potencia activa y contingencias N-1. Se optimizan las inversiones en la construcción de líneas de transmisión (LT) y nuevos SAEs. Se desarrolló un método eficiente para reducir el espacio de búsqueda combinatoria y mejorar la eficiencia del modelo. Los sistemas de prueba: Garver de 6 barras y IEEE de 24 barras; y el sistema de transmisión del norte de Colombia fueros usados para validar el modelo propuesto. Los resultados muestran que la optimización simultánea de LTs y SAEs proporciona planes de expansión mejores y más económicos que los enfoques tradicionales. Se evidencia la efectividad de los SAEs en desplazar bloques de energía en el tiempo, posponer inversiones y aliviar la congestión de la red.