335598 research outputs found
Sort by
Entwicklung eines Referenzmodells zum Reporting in wissenschaftlichen Einrichtungen anhand von DORA (ERRED)
No full textAls einer der ersten Unterzeichner der San Francisco Declaration on Research Assessment (DORA) ist das Karlsruher Institut für Technologie (KIT) ein aktiver Unterstützer der Reform der Forschungsbewertung. Ein Teil der Forschungsbewertung basiert traditionell auf bibliografischen Metadaten und bibliometrischen Indikatoren. Die Erhebung dieser Kennzahlen liegt meist im Tätigkeitsbereich der wissenschaftlichen Bibliotheken. Die KIT-Bibliothek bietet ebenfalls Beratung und Services im Bereich Bibliometrie und Reporting an. Ferner ist sie an Prozessen der Reform der Forschungsbewertung beteiligt und ein aktiver Fürsprecher für die DORA-Empfehlungen.
Das vom Bundesministerium für Forschung, Technologie und Raumfahrt (BMFTR) geförderte Projekt "Entwicklung eines Referenzmodells zum Reporting in wissenschaftlichen Einrichtungen anhand von DORA – ERRED" strebt die Entwicklung eines alternativen, übertragbaren Modells zur Forschungsbewertung an. Insbesondere umfasst dies die Entwicklung eines alternativen Sets an Indikatoren, das die Diversität im Forschungsalltag und im Output sowie verschiedene Karrierestufen berücksichtigt.
Im vorliegenden Sachbericht werden die Ergebnisse und die erfolgreich abgeschlossenen Arbeitspakete beschrieben. Darüber hinaus werden die Erfolgsfaktoren und Limitationen des Projekts dargestellt sowie mögliche Nachfolgeprojekte und Maßnahmen aufgezeigt
Towards Cost-Optimal Zero-Defect Manufacturing in Injection Molding: An Explainable and Transferable Machine Learning Framework
No full textIn the era of Industry 4.0, Zero-Defect Manufacturing is critical for injection molding but faces three major hurdles: severe class imbalance, the “black-box” nature of AI models, and the lack of scalability across machines. This study presents a comprehensive framework addressing these challenges. Using industrial datasets, we evaluated state-of-the-art supervised algorithms. Results show that CatBoost outperforms other architectures. Crucially, we demonstrate that maximizing accuracy is insufficient; instead, we introduce a cost-sensitive threshold optimization that minimizes economic risk, identifying an optimal classification threshold significantly lower than the standard. To enhance trust, SHAP analysis reveals that motor power and specific nozzle temperatures are the primary defect drivers. Finally, we validate a transfer learning approach using LightGBM, proving that models can be adapted to new datasets with minimal retraining. The implementation of cost-sensitive thresholding reduces total failure costs by over 75% compared to standard classification, while the transfer learning approach cuts the data requirements for new machine adaptation by more than half, providing a high-impact, scalable solution for sustainable smart manufacturing
DEMO Shutdown Dose Rate Assessment Inside the Vacuum Vessel
No full textShutdown dose rate (SDDR) assessments have been performed for the DEMO tokamak model, including the latest design and environmental configurations. The main objective of this study was to evaluate the shutdown radiation fields and establish dose rate limits to ensure safe personnel access to the Vacuum Vessel (VV) and nearby components. The simulations were based on the DEMO baseline model, further refined with the minor updates of the lower port, equatorial port limiter, and upper port assemblies. The computational approach employed the Monte Carlo particle transport code MCNP for neutron and photon transport calculations, coupled with the activation and decay code FISPACT-II to determine time-dependent decay gamma source terms. The mesh-coupled Rigorous Two-Step (R2Smesh) methodology developed in KIT was applied to achieve spatially resolved decay of photon source distributions and to compute corresponding SDDR 3D maps within the DEMO reactor configuration. The results provide a detailed characterization of the residual radiation environment inside the VV, offering insight into the accumulated activity, shielding performance of different materials, and potential access scenarios for maintenance operations in next-generation fusion devices
CAD to characterization: Machine learning on experimental data for additively manufactured helical distillation columns
No full textTraditional workflows in chemical engineering, from design to characterization, have served the field well. Emerging technologies, such as machine learning, now offer opportunities for faster design iterations and better integration of the various development stages. This study demonstrates how machine learning (ML) can reduce this gap by directly integrating experimental data into predictive models. Focusing on a Additive Manufactured helical micro-distillation column design for separating liquid mixtures, several ML models were trained to predict the number of theoretical stages based on geometric parameters and operating conditions. 11 variants were designed, built, and tested to collect 197 experimental data sets to assess the feasibility of five predictive ML models from experimental data. Among the algorithms tested, the Gradient Boosting Regressor achieved the best performance with a coefficient of determination = 0,7934. The work highlights the behavior of the model across different regimes, identifies key sources of error, and proposes a hybrid experimental–ML workflow for rapid screening of distillation designs. This approach accelerates process development and reduces the need for extensive experimentation, especially in time-consuming tasks such as distillation
Qualification of the European gyrotrons and power supplies of the Electron Cyclotron Heating and Current Drive system of ITER
No full text
Meteoric water δO across the Dinarides: Role of topography, air-mass mixing, and precipitation seasonality
No full textStudy region: Streams in the Dinarides ranging from coastal Croatia across the high-elevation
basins of Bosnia and Herzegovina to the lee of the Dinarides. Study focus: The topographic evolution of the Dinarides is poorly-constrained and its controlling geodynamic mechanisms remain unclear. The oxygen-isotope composition (δO) of authigenic minerals is a common paleo-altimeter for reconstructing past topography, proper interpretation requires thorough constraints on mechanisms modifying modern meteoric-water δO. To
constrain modern δO patterns across the Dinarides, we collected new stream samples and integrated them with published water stable isotope data. New hydrological insights for the region: Meteoric-water data show δO is higher at the coast (~-6 ‰) and lower at the peak (~-11 ‰). We use moisture trajectory models to show isotopic patterns across the Dinarides reflect two distinct moisture sources. The dominant source of moisture on the windward side originates from the Mediterranean and the leeward side has a continental source. This difference in moisture sources is reflected in d-excess values, which are high along the windward margin—reflective of Mediterranean moisture—and low in the lee,
reflective of summertime, continental-sourced moisture. We interpret orographic rainout as the primary-driver of modern precipitation and surface water δO with secondary influences from moisture sources and precipitation seasonality. Our findings have implications for understanding the climatic processes that deliver moisture as well as our understanding of the past topography of the Dinarides
A Multilevel Design for Scalable Pareto-Optimal Public Transit Queries
No full textThis thesis investigates routing algorithms for public transit networks. Especially in the context of countrywide or continental-sized transit networks, current state-of-the-art algorithms reach their limits: Fast query times can usually only be achieved through time-consuming preprocessing and high memory overhead. Until now, no algorithm has been able to efficiently compute Pareto-optimal journeys with respect to minimal arrival time and minimal number of transfers with both low memory consumption and minimal preprocessing effort on large networks.
We present T-REX (Transfer-Ranked EXploration), a new algorithm that narrows this gap by combining the state-of-the-art Trip-Based Public Transit Routing Algorithm with multilevel partitioning techniques. The underlying idea is to partition stops into multilevel cells and identify relevant trans-
fers for long-distance travel during a short precomputation phase. T-REX then uses this precomputed information to restrict the search space during a query, effectively pruning unnecessary local transfers. Unlike previous partition-based approaches, T-REX has two key advantages. Firstly, the query phase operates directly on the existing graph structure, without the need to build any additional data structures. Secondly, as T-REX focuses on transfers rather than lines or trips,
it can directly determine which events are worth exploring next, avoiding the overhead of computing the next reachable events. In this work, T-REX is described in detail, compared with existing methods, and evaluated in extensive experiments on real public transit data.
The algorithm is able to handle even large networks such as the whole of Europe with preprocessing times of a few minutes. At the same time, it allows for query times with an average runtime of around 18 ms, compared to 222 ms achieved by previous state-of-the-art algorithms, and requires only 515 MB additional memory compared to TB. These properties make T-REX particularly suitable for use in interactive real-time applications, e.g., in mobile navigation services
