APRÈN. Portal de la producció docent del professorat de la UPC

This project aims to optimize the calculation of the zero-fuel weight used in Vueling's flight plans, which is currently based on standard weights. A more precise methodology is proposed, based on real passenger reservations, along with the use of machine learning models to more accurately predict the actual weight of the aircraft. This would reduce the difference between the planned and actual weight, resulting in fuel savings and a reduction in CO2 emissions. The methodology involves the collection and preprocessing of real data, the selection of predictive models such as linear regression and Random Forest, and the validation of their accuracy through cross-validation. The results show a significant improvement in predictions, reducing the weight difference by an average of 170 kg and saving approximately 60 kg of fuel per flight. This project not only provides economic and operational improvements for the company but also contributes to environmental sustainability. Additionally, it opens up the possibility of automating this process in the future, further enhancing operational efficiency.

The main objective of this final degree project is to develop software that allows a drone to identify and classify objects in real time through the use of a camera. To achieve this, advanced computer vision techniques and artificial intelligence algorithms have been used, ensuring the correct integration between the developed software and the drone control system. A specific use case is to guide the movement of the drone following a route marked by objects strategically located on the ground. The methodology used to meet the objectives set out focused on the implementation of an object detection module based on the YOLO (You Only Look Once) algorithm, a convolutional neural network optimised for object detection in real time. The module was developed in Python, and its integration into the Drone Engineering Ecosystem (DEE), a drone control platform, enabled the identification of objects and subsequent decision-making by the drone. During the development process, different YOLOv8 models (v8n, v8s, v8m, v8l, v8x) were selected and evaluated, and then retrained using a proprietary dataset that included classes such as banana, ball, box and backpack. Several tests were performed, both in simulated environments and in a laboratory with a real drone, to measure the accuracy and efficiency of the system. The results were satisfactory, achieving an improvement in object detection compared to pre-trained models, with accuracy increases of up to 53% in some cases. Despite the achievements, the project had limitations, such as the impossibility of implementing object detection on the drone's Raspberry Pi due to technical problems with the library used, which restricted image processing to the ground equipment. In addition, the resolution of the drone's camera was not optimal for detecting small objects, and some false positives were observed that diverted the drone from its route at times. In conclusion, the project demonstrated the effectiveness of integrating an advanced object detection system into the DEE, opening the door to future improvements in model accuracy and drone functionality. Future lines of development are suggested to optimise the system, such as the reduction of false positives and the integration of processing on the Raspberry Pi.

In this master's thesis, three systems for an unmanned surface vehicle (USV) are designed, developed, and tested. These systems include a collision avoidance system, a method to create 3D models of underwater structures, and an integration system for a USV and a remotely operated vehicle (ROV). The first system utilizes LiDAR technology along with four RGB cameras to detect the location of objects around the USV. Based on this information, received via an MQTT channel, the USV defines protection zones for each target. If the USV's protection zone intersects with that of a detected object, it enters a collision alert state and calculates an evasive manoeuvre in accordance with COLREGs. The COLREGs (Convention on the International Regulations for Preventing Collisions at Sea) are international rules established by the International Maritime Organization (IMO) to prevent collisions at sea. This system enhances the autonomy of the USV by enabling it to avoid obstacles in its path. The second system facilitates the scanning of submerged structures using an ROV and then reconstructing them into 3D models. Three methods were investigated for this purpose: stereoscopy, structure from motion (SFM), and neural radiance fields (NeRF). Testing various programs and cameras revealed that both SFM and stereoscopy are effective options for creating 3D models of underwater structures. The third system involves the transportation and remote control of an ROV using a USV, leveraging the USV's onboard computer. This system includes a winch and a framework that secures and tows the ROV safely. The entire system is operated remotely using the USV's remote control. These three systems have been developed for a smaller USV, with the goal of future implementation in a larger USV to improve performance. This approach allows the testing and refinement of these systems on a currently robust USV. This work covers a wide range of topics, including data processing, MQTT communication, image processing, and both hardware and software development. By the end of the project, the three targeted systems were successfully developed, meeting the initial objectives. A collision avoidance system for the USV, a 3D modeling system using marine photogrammetry, and a docking system for the ROV to the USV were all completed.

The purpose of this project can be separated into two main objectives, first we want to study a way to virtualize a 5G network, and the second objective is to carry out the deployment of this network automatically. The project has been developed under the guidance of Neutroon Technologies S.L. as a part of a future proof of concept (PoC) investigation. To carry out the virtualization of the network, it is necessary to virtualize the different elements that make up a network formed by a Core 5G, an Edge computing that houses the application, a Firewall and a switch that allows to interconnect all the elements and separates the data traffic. In this way, all the functionalities of a network will be developed virtually in a single physical device. It should be noted that the Radio Access Network (RAN) part of the 5G network is not considered to virtualize, since it is preferable to keep this past in an isolated unit and based over hardware. The second objective of the work is to automate the deployment process so that human intervention is minimal and to be able to optimise the time required to carry out the task. It must be considered that the intention is to generate a tool that allows the entire deployment of the network by a person without the need to have deep knowledge about the required technologies. To facilitate the correct debugging of the process, checks are made with error codes and files are generated to store log messages. This document will discuss the necessary steps to be able to carry out the virtualization of the functionalities of the network mentioned previously with the use of OpenStack and the OSM orchestrator. It also includes a brief explanation of the structure and functioning of scripts that allows the automation of the network deployment.

This report describes the process of creating a game that is intended to be a graphic novel following the story of a young man who begins to discover the outside world around him at a stage of full development for him: the entrance to the university. This game will be created in collaboration with two students of the EETAC and a student of audiovisual education. Bringing together both disciplines is understood as a key factor to be able to approach the various problems that may appear from a different point of view and to create a better quality experience. The focus of the report will be fully on the logical aspect of the game since the visual part corresponds to another of the colleagues who collaborate in this project. It will address the various options that are available for solving the multiple problems that can occur and how each of the elements that make up the game is justified by the needs of the project itself. Throughout the report we will develop the reasons for the decisions taken for the creation of the game and a website that automates one of the parts that will allow us to parameterize the behavior of the interactions of the players with the non-playable characters. During the memory we justify as we said before each of the decisions that we can take to make things in the most agile way possible. We want this report to be a more or less good reference for those who do not know where to place themselves in the narrow world that mixes web development with video game development, and that the analysis of our problems will serve as an approach to address the particular problems that may arise in their projects.

The project focuses on the creation of a pixel art style video game using Unity 2D. It is based on the collaboration between students from different fields that we were at the same point of the degree, all with the aim of merging their knowledge and skills. The game is based mainly on the university life of a young man named Icarus, which has multiple scenarios and characters to interact with. The project has several main objectives, which include the creation of attractive pixel art style graphics, strengthening some of the skills that programming provides, interdisciplinary collaboration and improving the end user experience. All based and surrounded by a learning environment in Unity 2D and presenting a complete and functional final product. The game seeks to have players explore through college life, interact with characters and enjoy a captivating story. The narrative focuses on the emotions and personal relationships of the main character, Icarus, providing an immersive and realistic experience. The project is a pixel art style video game that combines retro aesthetics with modern elements. It offers an interactive and emotional experience that allows players to experience college life and make decisions that influence the story.

This project combines GNSS-R technology and UAVs in the context of the aerial remote sensing field, from the design of the assembly of the UAVs and GNSS-R to the execution of the different tests, as well as the data collection and analysis. As it will be seen in the document, this analysis has allowed to improve the original design of the system. The remote sensing technology based on GNSS reflected signals is simply called GNSS-R, and the greatest advantage of UAV as a mounting platform is the low cost, high system availability, and fast installation. Thus the main task of this project is to experimentally verify the possibility of CTTC's own developed GNSS-R working with UAVs. In this report, first I explain some basic principles of GNSS and GNSS-R, as they are important technical backgrounds for the whole project. Secondly, I describe the application of the RTKPOST software, which is an important processing and analyzing tool in the entire project, mainly describing the positioning methods used in the different processing modes in RTKPOST, also the parameter needed for analysis. Then, the design of the UAV-GNSS-R assembly is discussed, with different options based on the characteristics of the UAV, such as weight, payload space, and so on. The whole experimental process is also explained, and all the test data are analyzed and processed by using the Static mode or Kinematic mode in RTKPOST, and the data results are analyzed and compared in terms of satellite visibility, signal-to-noise ratio, elevation angle, Standard Deviation and Root Mean Square. The experimental process is roughly divided into three parts. The initial verification test is used to collect signal data as a basis, through which it is found that drone interferes with GNSS-R to a greater extent, and that the reflector antenna cannot receive the signal if the drone is placed too close to the ground. The second part of the experiment is a stationary test, where the UAV is placed on a tripod, and it's found that there is a difference in accuracy when the UAV is working on different land surfaces, which a higher accuracy can be obtained on rock than on grassy field. It is also found that the interference problem can be mitigated when the receiver is placed farther away from the UAV body. In the last part of the flight test, the optimization of the interference problem is further confirmed by the comparison of the two flight tests, and the positioning error is reduced from about 50m to 6m. At the end of the report, suggestions are given for the future development of this project, as it is clear that the meter-level accuracy does not meet the needs of many applications, and that the integration of the UAS and GNSS-R needs to be further optimized to reduce interference problems and thus achieve higher accuracy positioning.

The aim of this project has been to investigate the use of anomaly detection techniques for the detection of FODs on airport runways. The main approach consisted of evaluating different anomaly detection algorithms and proposing a model developed from scratch, in order to investigate anomaly detection in different operating conditions and scenarios. To achieve this goal, a detailed review of the existing literature in the field of anomaly detection has been carried out, investigating the most recent improvements and the most promising techniques in this field. In addition, experiments have been designed and performed using specific FOD datasets, evaluating the results obtained using established efficiency metrics. Experimental results reveal the effectiveness of the proposed models to detect foreign objects accurately and efficiently on airport runways, even under challenging conditions.

In this work, a study will be carried out on the task of inspecting airport runways using drones and artificial intelligence. Different aspects are evaluated based on verifying the feasibility of the operation. Aspects such as the impact of the drone's characteristics on the operation, and the possibility of training deep learning algorithms in FOD detection have been studied. Simulations have also been carried out emulating the real environment in order to see how the created environment would behave. The results provided by the different stages of the study confirm that the operation could be carried out for the majority of drones on the market, thus demonstrating the potential of performing this operation.

Genetic neuromuscular diseases are a group of disorders characterized by a progressive loss of muscle fibres and their substitution by fibrotic and fat tissue. Magnetic resonance imaging is helpful for the study of these diseases, as it identifies fat replacement in the muscles. Several studies have described patterns of muscle involvement that can guide patient diagnosis. In this project, we present a pipeline for the automatic diagnosis of muscle diseases using magnetic resonance imaging. We applied Deep Learning techniques to identify individual muscles in the images, defined an algorithm for the automated quantification of muscle tissue, and created a Machine Learning diagnosis model. This project was made possible thanks to an international network of 16 centres across 4 continents, which shared the required data for training the different models. We developed multiple tools for the processing and cohesion of the data, creating a large repository of data that will be useful for future projects.

We currently live in a world where technology predominates and advances very quickly, developing applications and devices that help us in daily life to improve our lives and solve everyday problems. One of these technologies that are in full development and evolution are the Drones, or also called RPAS (Remotely Piloted Aircraft System). These systems are unmanned aerial vehicles that offer us infinity applications to cover these needs and problems that we have on a daily basis. On the other hand, another technology that is in full development and that has been seen to have great potential for developing applications is Artificial Intelligence (AI). To quickly define AI, we say that it is the intelligence that humans develop with body, brain and mind but expressed by a machine, processor and software. Thanks to the fact that Drones can fly over places where humans cannot reach, through their cameras we can see what is in those areas, so one of the most useful applications in Drones is Search and Rescue operations. The main focus of this project is to merge RPAS with AI to develope a Search and Rescue application. For this, an executable software has been created for any computer, which will allow to detect people lost in the forest or other places, through a video, either through a Streaming on Youtube or a video saved locally, allowing detection both in real time as well as in deferred time, making the detection done by the machine and the human being able to do other functions while the search is done with the Drone. The objective is to use both, the thermal and visual cameras of a Drone, to record a video or stream the image and send it to the software so that, through Artificial Intelligence, if it finds a person, it detects the human and an alarm sounds. The software has been developed in Python, an open source, cross-platform programming language that can be used for web development, software creation, and data processing. This language is truly useful since it is one of the most used in the world of programming, thus there are multiple libraries, as OpenCV, created by open source users that have allowed the development of this human detection software. To develop the program, it has been necessary to train the machine with images of people. These images have been obtained with real flights in the area of Collserola, Barcelona. This area is close to the airport of Barcelona El Prat, so permissions and coordination are needed to be able to fly completely legally. For this reason, this project has also included all the documentation and legal part necessary to be able to fly in the Barcelona area.

El siguiente proyecto se ha desarrollado en un entorno industrial, bajo un convenio de prácticas con la empresa DAMM. La planta de trabajo es la fábrica ubicada en el Prat de Llobregat, en ella hay distintas naves, con más de 5 líneas de producción, que trabajan todo el día. En estas líneas existen numerosos controles de calidad que actúan en continuo, como por ejemplo los que comprueban la integridad, la limpieza o la presencia de cuerpos extraños y residuos líquidos en el interior, y todos ellos comprometen la calidad final del producto.

The technology has evolved very fast and in this last decade has started a future we don’t know for sure where it will arrive. The emergence of the machine learning has brought to our present days thoughts that were very far away at the beginning of the XXI century. The progress of the artificial intelligence has allowed the development of new intelligent machines capable to learn and improve by themselves. With the help of this technology projects like the autonomous vehicle is continually improving, allowing a user do a displacement where only needs to supervise the vehicle decisions. Moreover, using the deep learning we can design in the medicinal world a machine capable to discover a disease with only the help of a few medical images and thereby, save lives more efficiently. Having in a future, scenarios where the research and the creation of new medicines will be led by artificial intelligence processes. The objective of this project is to investigate the different techniques of image analysis, in more detail the deep learning based on semantic segmentation, to evaluate the implementation viability of a CNN in a UAS for the analysis of agricultural fields with Keras and Azure Machine Learning. Approaching typical problems as the research and the increase of the dataset, the choice of a work environment and the achievement of trainings and parameters adjustments. The obtained result is positive, raising the idea of possible solutions, but there is a big improvement gap with the research of new implementation techniques.

Both Machine learning and Deep learning are two concepts that are present in many areas, in health for medical diagnoses, as well as in marketing when segmenting the market or as in most social networks. In this project, we will see what they can offer at the level of recognition of facial expressions in photos. At present, there is infinity of data that is not used since the volume of data is so large that people are not able to process such quantity. The Deep learning and the Machine learning offer several mechanisms to be able to process this data, order them so that they can obtain information and take predictive models to be able to apply them to other data. The recognition of facial expressions is a part of artificial intelligence whose main objective is to recognize basic forms of affective expression that appear on the faces of people. However, can machine learning and deep learning techniques offer to be able to recognize these facial expressions sufficiently efficiently? In this project, we will apply these mechanisms offered by these two technologies to recognize feelings or emotions in people. To do this, we will first process and obtain the information we want from a database. This database is a series of images of people showing different feelings. Once obtained these data we will see that although the two techniques are promising, in Deep learning will obtain better results in precision and more robustness in the code.