Universitat Politècnica de Catalunya
UPCommons. Portal del coneixement obert de la UPCNot a member yet
290727 research outputs found
Sort by
Estudi de prioritats de productes mèdics en una cadena de subministrament humanitària
No full textAquest treball aborda la millora del procés de revisió de la planificació de la demanda en la cadena de subministrament humanitària de Metges Sense Fronteres (MSF OCBA), on el volum d’articles amb el que es treballa (més de 1.000 articles per projecte i uns 44 projectes actius) i la manca d’un sistema de prioritats integrat dificulten el procés, generant ineficiències i riscos de ruptures d’estoc. L’objectiu principal ha estat desenvolupar una eina computacional de suport a través de Python, que a partir dels registres anuals dels productes mèdics, classifiqui els articles segons criteris quantitatius i qualitatius per així poder agilitzar la revisió trimestral de les ordres IMO (International Medical Order) a través d’estratègies de revisió diferenciades. La metodologia utilitza dades transaccionals d’Isystock de l’últim any (extractes mensuals agregats per article/projecte/missió) amb l’horitzó limitat a 12 mesos a causa de la volatilitat del context humanitari. El model desenvolupat es basa en dos eixos principals d’una matriu 3x3 i en dues alertes. Un eix correspon a una classificació ABC per valor econòmic (seguint la regla de Pareto per 80/15/5) i l’altre correspon a una classificació XYZ utilitzant el CV per caracteritzar la variabilitat de la demanda. Pel que fa a les alertes, s’ha incorporat un paràmetre qualitatiu (MSL) i el percentatge de ruptura d’estocs. L’eina genera les classificacions resultants, les quals s’han incorporat a les eines de planificació existents. A més, es disposa d’un quadre de comandament en Power BI per visualitzar els resultats. Els resultats mostren que aproximadament un 60% del catàleg està format per articles de baix valor econòmic (segment C). Dins d’aquest conjunt, un 37,40% formen part del segment CZ, aptes per a mecanismes de revisió automàtica, mentre que un 17,68% d’articles formen part del segment AY, AZ i BZ, aptes per a revisió de prioritat crítica. En conclusió, els resultats mostren que una segmentació flexible i contextualitzada permet reduir ineficiències associades a ruptures i malbaratament, distribuint així, els recursos crítics de manera informada
Disseny, desenvolupament i programació d'una llança meteorológica
No full textAquest treball se centra en el disseny i el desenvolupament d'un prototip de llança meteorològica per a la monitorització ambiental en camp, dins d'un projecte més ampli d'adquisició de dades amb drons. El projecte neix com una millora davant de solucions més aparatosses o dependents de connectivitat, resolent sobretot la necessitat de disposar d'un node compacte, autònom i fiable per a campanyes a l'exterior. L'objectiu principal és dissenyar i fabricar una PCB capaç de registrar variables ambientals de manera periòdica i desar les dades localment durant llargs períodes, incorporant una gestió energètica orientada a baix consum i una arquitectura modular que faciliti afegir o substituir sensors en el futur.This project focuses on the design and development of a meteorological probe prototype for in-field environmental monitoring, as part of a broader drone-based data acquisition initiative. It is conceived as an improvement over bulkier solutions or systems that depend on continuous connectivity, addressing the need for a compact, autonomous, and reliable node for outdoor campaigns. The main goal is to design and build a PCB capable of periodically logging environmental variables and storing the data locally for long periods, combining low-power energy management with a modular architecture that makes it easy to add or replace sensors in the future.Este trabajo se centra en el diseño y desarrollo de un prototipo de lanza meteorológica para la monitorización ambiental en campo, dentro de un proyecto más amplio de adquisición de datos con drones. El proyecto nace como una mejora frente a soluciones más aparatosas o dependientes de conectividad, resolviendo sobre todo la necesidad de contar con un nodo compacto, autónomo y fiable para campañas en exteriores. El objetivo principal es diseñar y fabricar una PCB capaz de registrar variables ambientales de forma periódica y guardar los datos localmente durante largos periodos, incorporando una gestión energética orientada a bajo consumo y una arquitectura modular que facilite añadir o sustituir sensores en el futuro
Intel·ligència artificial per a la classificació varietal de vins mitjançant anàlisi de textos
No full textAquest Treball Final d’Estudis presenta un estudi comparatiu sobre l’aplicació de tècniques de Processament del Llenguatge Natural i Machine Learning a la classificació de textos, utilitzant com a cas d’estudi un conjunt de dades de ressenyes de vins. El treball se centra en la predicció de la varietat de raïm a partir de descripcions textuals, analitzant diferents estratègies de representació del llenguatge i models de classificació, des d’enfocaments tradicionals fins a models de llenguatge basats en arquitectures Transformer. Per assolir aquest objectiu, el treball es desenvolupa seguint dues aproximacions complementàries. En primer lloc, s’apliquen mètodes clàssics de representació textual basats en tokenització i vectorització mitjançant TF-IDF, combinats amb classificadors tradicionals com Multinomial Naive Bayes, amb la finalitat d’establir una línia base de rendiment. En segon lloc, s’estudia l’ús de models basats en Transformers, utilitzant tant embeddings preentrenats amb Sentence-BERT com models amb fine-tuning complet basats en RoBERTa, aprofitant arquitectures i pesos entrenats prèviament sobre grans corpus textuals. La metodologia emprada inclou la preparació i divisió del conjunt de dades, l’entrenament dels diferents models i l’avaluació mitjançant mètriques habituals en problemes de classificació multiclasse, com ara accuracy, precision, recall i f1-macro. Aquest enfocament permet analitzar de manera sistemàtica les diferències de rendiment, capacitat de captació semàntica i cost computacional entre els models clàssics i els basats en deep learning. Els resultats obtinguts mostren que els models basats en Transformers ofereixen, en general, un millor rendiment en la classificació de varietats de raïm, especialment en termes de comprensió semàntica de les descripcions textuals, tot i requerir un major cost computacional. Alhora, els models tradicionals presenten un comportament robust i eficient, constituint una alternativa vàlida en entorns amb recursos limitats. Finalment, el treball conclou que la combinació d’ambdós enfocaments proporciona una visió completa del problema i permet comprendre els avantatges i limitacions de cada model, establint una base per a futurs estudis i ampliacions en l’àmbit de la classificació de text
PyScript-powered photovoltaic simulator: integrating past and future meteorological data for predictive maintenance and performance forecasting
No full textThe rapid expansion of photovoltaic installations worldwide has created an urgent need for accessible simulation tools that can be used by facility managers, technicians and students to understand system performance and diagnose operational issues. However, current photovoltaic simulation software is desktop-based, proprietary and designed for expert users, which creates significant barriers to entry for non-specialists seeking to optimise solar installations. This final thesis aims to address this issue by developing and rigorously validating a web-based photovoltaic simulator that runs entirely within standard web browsers using PyScript and WebAssembly technologies. The simulator incorporates the PVUSA performance model, which is based on empirical data, and three physics-based degradation mechanisms: thermal stress via Arrhenius acceleration factors, humidity-induced corrosion through Hallberg-Peck modelling and Humboldt State University (HSU) particulate deposition model for soiling losses. These components combine to form a Transitory Deterioration Rate (TDR), which captures multi- year performance decline. A Flask backend integrates Open-Meteo API data in real-time, maintaining lightweight client-side execution. The system's modular design allows users to experiment with different configurations, enabling them to see how various factors, such as degradation, environmental conditions, and system losses, influence energy predictions. A 1009-day period of operational data from a residential installation in Lleida, Catalonia was used to validate the model, achieving an R² of 0.8716, a mean of 2.42 kWh, a median of 1.91 kWh and a 6.68% cumulative error across combined training and test periods. The simulator also provides 15-day energy forecasting, comprehensive validation metrics and full data export capabilities. This work contributes to renewable energy education and enables data-driven decision- making for small-scale system operators by democratising advanced photovoltaic analysis through accessible web technologies, eliminating installation requirements, licensing costs, and programming prerequisites
Real-world impact of nirsevimab immunisation and maternal RSV vaccination against respiratory disease on emergency department attendances and admissions: a multinational retrospective analysis
No full textDr Damian Roland declares grants or contracts from Wellcome Trust (Funding for administrative support for the BronchStop study), Respiratory Syncytial Virus Consortium in Europe (RESCEU) (Funding to establish the BronchStart/Stop study), Imperial College London (Funding for administrative support for the BronchStop study), National Institute for Health Research (Funding for data collection for the study). Dr. Thomas Williams declares funding from the Wellcome Trust: funding for administrative support for the BronchStop study. Dr Antoni Soriano-Arandes declares consulting fees, honoraria for lectures and support for attending meetings from Sanofi, MSD, and Pfizer, and grants or contracts from la Marato tv3 (Funding addressed to my institution Vall d’Hebron Research Institute for a project about respiratory viruses). Clara Prats declares payments or honoraria for lectures by Societat Catalana de Pediatria (Presentation including results of the impact of nirsevimab in Catalonia), grants or contracts by Fundacion BBVA (Funding paid to the university), and Fundacio la Marato tv3. Aida Perramon-Malavez declares grants or contracts from Fundacio la Marato tv3 (Funding addressed to my institution Vall d’Hebron Research Institute for a project about respiratory viruses). Dr Jorgina Vila declares honoraria for presentation and support for attending meetings from Sanofi. Anna Creus-Costa declares support for attending meetings and/or travel by Sanofi and grant VHIR-AULA-2025-016 by Aula Vall d’Hebron. The rest of the authors have nothing to declareBackground. Nirsevimab, a long-acting monoclonal antibody against respiratory syncytial virus (RSV), was recently introduced in Catalonia (Spain, 2023–2024 season onwards) and Italy (2024–2025 season). The United Kingdom (UK) instead introduced maternal RSV vaccination (RSVpreF) in the 2024–2025 season. Our aim was to analyse emergency department (ED) attendances and admissions to hospital following RSVpreF and nirsevimab introduction, with hospitals in Iceland and Romania, where no intervention was introduced, used as comparators. Methods. Multi-national retrospective analysis of ED attendances and admissions for all diagnoses, respiratory diagnoses excluding bronchiolitis, and bronchiolitis from all hospitals in Catalonia (Spain), four UK hospitals (Bristol, Edinburgh, Glasgow, and Leicester), and one hospital in Italy (Rome), Romania (Bucharest), and Iceland (Reykjavík) from April 2018 to March 2025. Bronchiolitis diagnoses in the 2024–2025 season were compared to previous pre-intervention seasons (2018–2023, excluding the 2020–2021 COVID-19 year) by applying a generalised linear model in Poisson regression to obtain risk ratios (RR) and 95% confidence intervals (95% CI). Findings. In the 2024–2025 season, in Catalonia, there was a reduction in the RR for bronchiolitis ED attendances and admissions in infants of age <6 months (RR 0.45; 95% CI 0.43–0.47 and RR 0.40; 95% CI 0.37–0.43, respectively). This was not seen in Rome, where the RR for ED attendances with bronchiolitis in infants of age <6 months was 1.09 (95% CI 0.92–1.30) and the RR for admissions was 1.12 (95% CI 0.83–1.52). In the UK, for infants of age <6 months with bronchiolitis, there was a significant but modest reduction in 1 out of 4 hospitals for ED attendances (Leicester; RR 0.91, 95% CI 0.85–0.97) and in 2 out of 4 hospitals for admissions (Leicester; RR 0.80, 95% CI 0.69–0.94 and Edinburgh; RR 0.85, 95% CI 0.76–0.95). Interpretation. In Catalonia, there was a sustained reduction in bronchiolitis ED attendances and admissions for infants in the 2024–2025 season. However, no substantial reduction in bronchiolitis ED attendances or admissions was observed in Rome or the UK. These differences are likely to relate to the reduced uptake of RSV prevention products in these settings compared to Catalonia.Postprint (published version
Use of AI and FEM finite elements method in the design of large-size parts
No full textThis research assignment, conducted within the CASANDRA project, explores two key avenues for improving the manufacturing of large metal parts: the numerical modeling of welding processes and the use of Artificial Neural Networks for metal-sheet bending prediction. A bibliographic review provided insight into existing welding models and their limitations, particularly the need for experimentally calibrated parameters. The second part of the work focused on developing a simplified neural network based on theoretical bending equations to evaluate the potential of Machine Learning in forming operations. Although the results remained approximate, the study highlights the promise of AI-based methods for enhancing process accuracy and reducing operator dependency. Overall, the project emphasizes the value of combining simulation tools and data-driven approaches to support more efficient and modern industrial fabrication.Incomin
Exploiting brewery wastewater for the sustainable cultivation of Chlorella fusca: Biomass productivity, biochemical composition, and production cost
No full textBeer is the fifth most consumed beverage worldwide, and its production demands large volumes of water, generating 3–10 L of effluent per liter of beer produced. This wastewater often requires municipal management, representing high operational costs and environmental concerns, particularly for microbreweries with limited infrastructure. The use of microalgae in brewery effluent offers a sustainable alternative for nutrient recovery, biomass generation, and integration into circular bioeconomy strategies. This study evaluated the growth performance and biochemical composition of Chlorella fusca LEB 111 cultivated in brewery effluent under semicontinuous operation, aiming to reduce production costs while maintaining biomass quality. Cultures were grown in 1.8-L photobioreactors with effluent concentrations of 25%, 50%, 75%, and 100%, compared to a BG-11 medium control. The 100% effluent assay achieved a biomass concentration of 0.77 g L¿1 and productivity of 52.34 mg L¿1 day¿1, with a balanced biochemical profile of 39.96% proteins, 31.64% carbohydrates, and 28.89% lipids. Notably, the highest protein content was obtained in EA75 during the first cycle (46.20%), the maximum carbohydrate accumulation occurred in EA75 during the second cycle (35.59%), and the highest lipid content was recorded in EA50 during the third cycle (40.61%). Cost analysis revealed a substantial reduction in biomass production expenses, from over €84 kg¿1 in the control to €18 kg¿1 in the undiluted effluent assay. These results advance knowledge on the adaptive biochemical responses of C. fusca under nutrient-limited wastewater conditions, providing a framework for sustainable algal cultivation and biomass valorization from nutrient-limited industrial effluents.Postprint (published version
The centrifuge of the Pellet Launching System for JT-60SA: initial test results
No full textPart of the Broader Approach, JT-60SA was established as satellite experiment supporting ITER and paving the way for a future fusion power plant. Its research plan calls for a powerful Pellet Launching System (PLS), capable for both particle fuelling and ELM pacing. These different tasks are achieved simultaneously by combining a pellet source for each purpose on a centrifuge. This stop cylinder type centrifuge provides dedicated pellet launch slots, which can be filled by the pellet sources controlled by a master programmable logic control (MasterPLC) system. Previously, centrifuges like those used at ASDEX Upgrade and JET were built by modifying a large turbomolecular pump and adding an acceleration arm. The advanced novel system reported here takes an approach that separates the drive from the arm. A ferrofluidic rotary motion feedthrough, capable of speeds up to 166 Hz, transfers the rotary motion from a commercial servomotor operating in the atmosphere to a rotor blade in the high vacuum. This system is designed to launch pellets with a velocity of up to 600 m/s when the centrifuge is rotating at 120 Hz. The factory acceptance test has almost been passed. After passing the endurance test, the system will be delivered to IPP Garching, followed by a commissioning phase in the pellet lab.Postprint (published version