Refine
Institute
- Fakultät Informatik (863)
- Fakultät Elektro- und Informationstechnik (701)
- THI Business School (548)
Version
- published (1191)
Die Technische Hochschule Ingolstadt (THI) verfügt über 20 Jahre Erfahrung mit kooperativen Promotionen. In diesem Zusammenhang wurde bereits im Jahr 2013 ein Graduiertenzentrum zur Unterstützung der Promovierenden und Betreuenden an der THI eingeführt. Im Jahr 2023 folgte der Start zweier Promotionszentren an der THI. Grundlage hierfür war das im Januar 2023 in Kraft getretene Bayerische Hochschulinnovationsgesetz (BayHIG), das die Möglichkeit zur Verleihung eines fachgebundenen und befristeten eigenständigen Promotionsrechts an bayerischen Hochschulen für angewandte Wissenschaften geschaffen hat. In einem Begutachtungsverfahren überzeugte die THI mit dem Promotionszentrum Ingenieurwissenschaften (PZ ING) und dem Promotionszentrum Künstliche Intelligenz / Informatik (PZ KI/INF) durch eine nachweisbare Forschungsstärke und die Einbettung der wissenschaftlichen Qualifizierung in eine grundständige akademische Lehre. In den Promotionszentren ING und KI/INF sind derzeit jeweils 23 forschungsstarke Professorinnen und Professoren tätig. Innerhalb der ersten zweieinhalb Jahre wurden 90 Doktorandinnen und Doktoranden in die beiden Promotionszentren der THI aufgenommen, drei von ihnen haben mittlerweile den kompletten Promotionsprozess durchlaufen und ihre Promotion erfolgreich abgeschlossen. Im Folgenden werden strukturelle und prozessuale Erfahrungen sowie vor allem Spannungsfelder und Aspekte der Qualitätssicherung in den Promotionszentren beleuchtet.
Marketing von Innovationen
(2009)
As traffic automation grows, research into communication between automated vehicles (AVs) and pedestrians is increasing. Since communication is particularly challenging for persons with intellectual disability (PID), we investigated which gestures are used during street crossings and whether they differ between PID and persons with no diagnosed disability (PnDD). We conducted a virtual reality study with N=70participants ( n=38PID, n=32PnDD), observing participants’ movement and gestures in a road-crossing scenario and having them rate their presence, user experience, and the AV’s behavior. PID descriptively gestured less often than PnDD (33.3% vs. 18.8% of trials without a gesture), though this difference was not statistically significant. The increased learning effect of using gestures improves crossing time for PID. We identified three primary gestures (Greet, Barrier, Stop) that both groups equally favored. Our work contributes to making bidirectional communication with AVs more inclusive by providing insights into gestures used by PnDD and PID and identifying indicators of subconscious learning effects when using gestures.
Indentation and crush tests are widely used as crash-relevant surrogates for mechanical intrusion in Lithium-Ion Batteries (LIBs), yet their interpretation still relies on macroscopic, binary pass/fail criteria established at the Beginning of Life (BoL). This neglects how calendar and cyclic degradation alter a cell’s mechanical, thermal, and structural state before abuse, leaving the literature on aged cells apparently contradictory. This review establishes an aging-aware interpretive framework that separates three levels commonly conflated: the boundary conditions governing deformation, the study-specific criteria and evidence strength behind a declared electrical event, and the post-trigger hazard progression captured within the available monitoring window. A descriptive synthesis of 22 normalized fresh-to-aged comparisons from 13 primary studies shows no universal aging-driven shift of the reported comparison event toward earlier or later force–displacement states, with force and displacement changing in the same direction in only about half of the cases. An apparent format-associated displacement pattern is substantially attenuated when the pouch-cell displacement ratios are normalized by the respective initial cell thicknesses: the pouch-cell median displacement ratio decreases from 1.07 to approximately 1.00, and the remaining variation differs across aging routes and states of charge, although their individual contributions cannot be separated in the available dataset. Many apparent contradictions likewise stem from non-equivalent mechanical boundary conditions and trigger definitions, and continuity of confinement—especially for pouch cells aged under stack pressure and tested in a relaxed state—is identified as a frequently overlooked interpretive variable, although no controlled aged-cell study has yet isolated its contribution. An earlier-reported trigger does not necessarily imply a more severe outcome. We distill these requirements into a tier-resolved reporting checklist, providing a more rigorous and reusable basis for abuse-test interpretation, model development, and safety qualification of batteries intended for long service life and second-life deployment.
Enhancing Teleoperation Performance for Automated Vehicles Through Bird’s-Eye View Augmentation
(2026)
Uncertainty-Aware Diffusion Model for Multimodal Highway Trajectory Prediction via DDIM Sampling
(2026)
Energy- and Runtime-Efficient Trajectory Planning via SIMD Vectorization with Reduced Precision
(2026)
Long-Term Engagement with High-Level Driving Automation: Real-World Experiences from Early Adopters
(2026)
An essential objective of modern battery management systems is to ensure safe battery operation. The system shall issue a warning when it detects a fault condition—the earlier, the better. Reliable fault detection requires a sensitive but equally robust detection method. In battery systems, a conspicuous change in the cell-to-cell variation can be an adequate fault indicator. To detect anomalies like this, a recurring method in prior work is principal component analysis (PCA) or one of its variants. Its effectiveness depends heavily on input data preprocessing, though, a topic that is often disregarded. To raise awareness of proper preprocessing, we investigate common techniques in terms of their impact on PCA-based fault detection. We take thermal faults and internal short circuits in battery systems as prototypical examples for our investigation. Our results highlight the importance of selecting appropriate preprocessing for the problem at hand. Well-matched preprocessing shapes the input data such that the theoretical requirements of PCA are met to a greater extent, which enhances overall detection performance. Based on this understanding, we propose a combination of outlier-robust sample studentization and Pareto scaling that improves fault detection compared with alternatives from prior work in our exemplary application. The detection sensitivity and robustness are greatly enhanced, particularly for minor faults, without adding significant computational complexity.
AbstractAntimony containing Sn-3.9Ag-0.6Cu-3.0Sb (SAC 396 +) is a high reliability solder alloy but its microstructure and creep behavior is not sufficiently investigated yet. In this paper the correlation between microstructure, using SEM, TEM, and XRD, and compression creep properties was investigated. Furthermore, a comparison in creep performance between SAC 396 + and Sn-3.8Ag-0.7Cu (SAC 387) was drawn, and the long-term microstructure evolution until 1500 h at 125 °C was compared. For SAC 396+, microstructure investigations of pre-aged (24 h at 125 °C) bulk samples revealed that Sb is mainly dissolved in β-Sn (Sb2Sn23, tetragonal, I41/amd), whereas some fine SbSn precipitates (Sb0.49Sn0.51, rhombohedral, R-3 m) were found in β-Sn. Compression creep tests were performed at temperatures between 35 °C and 125 °C and at stresses between 7.5 MPa and 32.5 MPa. Stress exponents were in the range of 7.4 to 7.5 at 100 °C or below, and 4.8 at 125 °C. Activation energies were in the range between 84.7 kJ/mol and 99.6 kJ/mol, indicating dislocation creep, controlled by pipe diffusion or by lattice diffusion. Overall, Sb hardened SAC 396 + showed better creep performance at all tested temperatures, as well as after isothermal ageing at 125 °C for 24 h, 500 h, or 1500 h.
Although laissez-faire leadership is common in organizations and has been linked to detrimental effects on employees, little is known about its daily effects and how employees cope with this type of behavior. Drawing on the job demands-resources model and using a daily diary design, we examine the daily effects of laissez-faire leadership depending on followers’ coping styles, capturing both adaptive and maladaptive responses. Specifically, we argue that the negative effect of laissez-faire leadership on next-day performance via evening job satisfaction is mitigated on days that followers engage in job crafting, while it is amplified on days that followers engage in disengagement coping. We collected data twice a day over one working week in an experience sampling study with 127 employees (i.e., after work and before bedtime; 359 data points). Our findings revealed no direct effect of laissez-faire leadership. However, there was a positive indirect effect of laissez-faire leadership on next-day performance via evening job satisfaction on days when employees engaged in high levels of job crafting, supporting its hypothesized beneficial effect. Conversely, as hypothesized, a negative indirect effect was observed on days when disengagement coping was high. By identifying the mixed effects of laissez-faire leadership, this study offers a fresh perspective that challenges the dominant view of its uniformly negative consequences, revealing how its downstream effects vary within individuals depending on their daily coping strategies.
Die systematische Verwaltung und Wiederverwendung von Informationen und Wissen ist von zentraler Bedeutung für die industrielle Fertigungsplanung. Ein erheblicher Teil der relevanten Inhalte ist dabei über eine Vielzahl umfangreicher Dokumente verteilt. Dadurch sind Informationen trotz formaler Verfügbarkeit häufig nur mit hohem manuellen Aufwand auffindbar, insbesondere wenn sich Inhalte in unterschiedlichen Ablagestrukturen befinden. Aktuelle Entwicklungen der Künstlichen Intelligenz eröffnen neue Möglichkeiten für einen effizienten, zentralen Zugriff darauf. Für einen Einsatz in der Fertigungsplanung wird hierzu in der vorliegenden Arbeit ein Konzept erarbeitet, das die domänenspezifischen Rahmenbedingungen bei der Speicherung, Ausgabe und Prüfung der Inhalte berücksichtigt.
Dazu werden zunächst Anforderungen an ein solches Konzept abgeleitet, die die Grundlage für die Erarbeitung der Lösung bilden. Auf dieser Basis wird ein modularer Wissensspeicher konzipiert und prototypisch umgesetzt, der dokumentenbasierte Inhalte strukturiert erschließt und über einen retrievalgestützten Zugriff nutzerfreundlich bereitstellt. Der Ansatz kombiniert dabei eine dokumentenzentrierte Strukturierung mit inhaltlichen Verknüpfungen, um sowohl die Dokumentenstruktur als auch thematische Zusammenhänge über Dokumentgrenzen hinweg nutzbar zu machen. Ergänzend wird ein Verfahren zur Prüfung der generierten Antworten entwickelt, um unerwünschte Halluzinationen zu reduzieren und die Zuverlässigkeit der Antwortinhalte zu erhöhen. Die Evaluation erfolgt anhand öffentlicher sowie domänenspezifischer Daten aus der Fertigungsplanung und bewertet Eignung, Qualität und Nutzbarkeit des Konzepts.
Das Ergebnis ist ein geprüftes Konzept, das den Zugriff auf verteiltes dokumentenbasiertes Wissen vereinheitlicht und für unterschiedliche KI-Anwendungsfälle nutzbar macht. Gleichzeitig schafft es eine modulare Grundlage, um den Ansatz perspektivisch um weitere Datenquellen und Integrationen zu erweitern und damit die Breite industrieller Wissensbestände schrittweise abzudecken.
Small-scale biogas systems in developing regions are predominantly mono-digestion systems utilizing livestock manure as the primary feedstock. However, crop residues such as wheat straw offer significant potential for improving feedstock diversity and biogas production when applied in anaerobic co-digestion systems. Due to the recalcitrant nature of lignocellulosic biomass, pretreatment is required to enhance substrate degradability and methane production. In this study, the performance of integrated pretreatment and anaerobic co-digestion of cattle manure and wheat straw was evaluated using the modified ADM1_R3 model. Simulations were conducted under varying feedstock mixing ratios (0 to 100%wt wheat straw), pretreatment intensities (carbohydrate degradability levels of 50%, 67%, and 75%), organic loading rates (1 to 4 kgVSm−3day−1), and digester volumes (2, 4, and 6 m3). The results showed that increasing the wheat straw fractions improved biogas production, although with a slight reduction in methane. Pretreatment further enhanced overall process performance, with biogas production enhancement of between 40% and 56% across the different mixing ratios and degradability increase from 50 to 75%, while higher loading rates combined with higher pretreatment intensities increased the risk of process instability. The findings demonstrate the feasibility of this innovative approach of flexible small-scale co-digestion systems supported by appropriate pretreatment strategies. This study advances the application of anaerobic digestion modelling to small-scale biogas systems by providing an integrated framework for evaluating the effects of operational and design parameters on technical performance.
Experimental Performance Comparison of Four Different Hysteresis Models for Lithium-Ion Cells
(2026)
In this paper, four versions of hysteresis models parameterized by experimental data for nickel manganese cobalt (NMC) lithium-ion cells are compared. Starting from simple 1 to 5 RC models that only account for hysteresis by averaging open circuit voltage (OCV) curves measured for charge and discharge direction, three model extensions for calculating the hysteresis-biased OCV curve are investigated. Therefore, the nRC models are taken as base models and combined with the extended Preisach model, including a modified version (EPM and mEPM), which calculate the OCV in charge and discharge direction separately and thus require a complex parameterization profile. A one state model (OSM), which simply extends the nRC model by adding a hysteresis voltage source to the model structure, not only requires less parameterization efforts than the EPM and mEPM, but in combination with a simple 1RC model also shows the best model performance demonstrated by the lowest root mean square error (RMSE). The OSM models are compared to a second cell chemistry (lithium-ironphosphate), showing a decent performance, especially for medium-range state of charge values.
Impact of Hysteresis on Calendar Aging Mechanisms on Three‐Electrode Cells With Silicon‐Rich Anodes
(2026)
One promising strategy to increase the energy density of lithium‐ion batteries is to increase the silicon (Si) content in the anode. However, an increasing amount of Si is a challenge for lifetime and diagnosis, as it suffers from high volume expansion and strong voltage hysteresis. While cyclic aging dominates current research, calendar aging remains still comparatively underexplored. Therefore, we present calendar aging results of Si‐rich three‐electrode lab cells, focusing on voltage hysteresis, loss of active material, and electrode‐specific aging. We utilize OCV and float current analysis to assign electrode‐specific aging. Both methods have shown promising results for pure graphite cells and are here transferred to cells with 69wt% Si combined with NCA cathode. The tests are conducted at the target state‐of‐charge/voltage, approached from both charge and discharge directions. We found no loss of anode‐active material in the cells, but strong voltage hysteresis. As assigning the anode voltage time‐series data to the OCV curve was not possible, anode aging was calculated indirectly from cathode aging and loss of lithium inventory. Anode and cathode aging increased with electrode potential, full‐cell voltage, and SOC; however, no parameter exhibited a significantly strong correlation due to the limited accuracy of the reference electrode.
Feedback-Guided Knowledge Distillation for RGB-Only 3D Object Detection in Autonomous Driving
(2026)
Cooperative perception enabled by Vehicle-to-Everything (V2X) communication enhances autonomous driving safety by creating a unified environmental representation through shared sensory data. While recent works have advanced multi-agent fusion for improved perception, uncertainty quantification in such cooperative frameworks remains largely unexplored. This paper introduces Hyper-V2X, a hypernetwork-based framework for estimating both epistemic and aleatoric uncertainties in V2X-based perception. Specifically, we propose a partial weight generation scheme and V2X context embedding module that conditions a Bayesian hypernetwork on fused multi-agent features to generate weight distributions for stochastic Bird's-Eye-View (BEV) segmentation. Unlike existing deterministic BEV models, Hyper-V2X enables efficient uncertainty estimation with little computation overhead. Our approach is architecture-agnostic, and can be seamlessly integrating with modern cooperative backbones such as CoBEVT. Experiments on the OPV2V benchmark demonstrate that Hyper-V2X provides accurate, well-calibrated uncertainty estimates and improves overall perception reliability.
The automation of highly specialised software often relies on the use of equally specialised or proprietary Application Programming Interfaces (APIs). Over time, these APIs undergo version upgrades, making older automation scripts deprecated. In many cases, the huge amount of automation scripts developed makes it inviable to migrate them manually to the newest API version. To assist with this process, rule-based tools exist. However, they still require manual curation of migration rules, which is complex and time-consuming. On the other hand, AI-based tools like Large Language Models (LLMs) require substantial data for fine-tuning, which is often unavailable due to the proprietary or niche nature of APIs.
To overcome this problem, we propose LAMB, our LLM-Assisted Code Migration Bot. LAMB uses mapping tables generated programmatically from API documentation to curate migration rules at runtime based on input code. These rules are then used alongside the input code to build a few-shot prompt, which is then fed to an LLM to generate the output code in the new API version. LAMB is model-agnostic and is built as a standard Python package. Moreover, it integrates seamlessly with Jupyter AI notebooks. Early assessments indicate our approach is more effective than standard Retrieval Augmented Generation (RAG), confirming that this is a promising direction for development. Finally, by establishing an automated feedback loop with software developers, we aim to improve the generation of migration rules and, if possible, curate a gold-standard dataset for further refinement of LLMs.
The increasing share of low-emission variable renewable energy sources increases the need for additional demand-side flexibility to better integrate and utilize variable renewable generation. Cold storage warehouses are promising candidates, as the thermal inertia of stored products allows cooling demand, and thus electricity demand for cooling, to be shifted in time. This paper proposes a linear optimization model for cold storage warehouses that explicitly accounts for pallet-level thermal behavior and temperature constraints of food products at warehouse scale. The approach is evaluated in a real-world case study for dairy products over one week using time-varying grid emission factors. Results show that the method achieves reductions in cooling-related CO2 emissions while strictly respecting pallet-level temperature limits. Overall, the proposed approach provides a promising basis for carbon-aware cooling optimization in cold storage warehouses.
Smartphone applications routinely collect and share personal data, yet users often struggle to understand these practices, particularly when third-party data sharing is involved. Existing mechanisms such as privacy policies and notices provide limited support for user understanding. To address this gap, we investigated how explainability concepts can enhance contextual privacy policies for mobile apps. We designed two interface prototypes integrating explanation strategies: contrastive explanations, which clarify data-sharing boundaries, and example-based explanations, which illustrate counterfactual scenarios. In an exploratory between-subjects user study ( N = 30), we evaluated their impact on comprehensibility, simplicity, cognitive load, and experience. Statistical analysis revealed no significant differences between the two types. Hence, the findings should be read as descriptive trends rather than confirmatory effects. These trends suggested example-based explanations better supported users mental models, while contrastive explanations better supported decision-making. Our findings contribute design recommendations on applying explanation strategies in privacy interfaces, offering guidance for developers and researchers seeking to improve user understanding and trust in digital systems.
Intelligent Transportation Systems (ITS) play a crucial role in modern mobility by leveraging advanced technologies to enhance human safety on the road, optimize traffic flow, reduce congestion, and minimize environmental impact. By integrating environment data perceived by both vehicles and roadside infrastructure units, ITS can enable dynamic management of transport networks, supporting efficient, safe, and sustainable travel for all road users. In recent years, significant advancements in multi-sensor perception have been achieved for vehicle-based platforms, driven by developments in autonomous driving and advanced driver assistance systems (ADAS). However, until recently, roadside infrastructure units primarily made use of RGB cameras for passive traffic monitoring. These cameras are highly vulnerable to low-light and adverse weather conditions, which significantly reduces their detection capabilities. Moreover, cameras are only able to grasp 2D information, failing to accurately represent the real-world 3D environment. But for ITS applications, the roadside infrastructure unit must detect various road users robustly and consistently in a 3D vector space under different light and weather conditions.
A promising solution to these limitations is to combine radar or lidar sensors with cameras. Compared to lidar, radar is more robust to varying lighting and weather conditions, lower in cost, and provides accurate, long-range depth and velocity data, making it an ideal complement to cameras. Fusing data from both sensors enables the use of their complementary strengths while compensating for their limitations. However, most commonly used decision-level fusion or late fusion (including object-level and track-level fusion) in roadside infrastructure setup processes radar and camera data independently to generate detections or tracks, which are then combined at a later stage. In adverse weather or low-light conditions, the performance of the camera degrades significantly, which diminishes its contribution to these fusion methods. Early-stage, intelligent fusion of camera and radar data using deep neural networks can improve overall detection quality, but this approach remains largely unexplored for roadside-mounted setups, especially with RGB cameras and 3D radar. This research gap is significant, and at the onset of this study, no public dataset existed for this sensor combination in a roadside installation, further motivating the need to develop new datasets and methodologies in this field.
To achieve these objectives, the research begins with the development of a modular, portable, and experimental multi-sensor roadside infrastructure unit, encompassing mechanical and electrical design as well as a ROS-based (Robot Operating System) data collection framework. Innovative methods for multi-sensor calibration and time synchronization are developed to ensure accurate data acquisition. Several data collection campaigns are then carried out at two locations in Ingolstadt under various lighting and weather conditions to gather data with relevant road users. To reduce manual labeling effort when preparing the dataset for early fusion algorithm training, a semi-automatic annotation methodology is developed for both camera images and radar point cloud data on a frame-by-frame basis. Subsequently, a deep learning model called Infra-3DRC-FusionNet is designed and implemented to jointly fuse synchronized and calibrated RGB images with the 3D point cloud, enabling detailed,
frame-wise detection of road users, such as pedestrians, bicycles, motorcycles, cars, and buses, in urban environments. The results produced by the fusion model, which include 2D image bounding boxes, object categories, confidence scores, and associated 3D radar points for each detected road user, are further processed using a 3D extended object generation technique to create 3D oriented bounding boxes in the ground plane relative to the infrastructure unit. Various experiments are conducted to train the Infra-3DRC-FusionNet model by modifying specific parameters and sub-modules to both maximize performance and analyze the impact of different settings on the model’s output. The best-performing model is then evaluated under varying lighting and weather conditions using 3D ground truth data from a lidar sensor. The evaluation focuses on position and classification accuracy on the ground plane, early detection performance when a road user enters the sensor field-of-view, and detection consistency across frames while the user remains in view. The model is also benchmarked against four state-of-the-art object-level fusion methods using the same evaluation metrics. Results show that early fusion of radar and camera data in a roadside-mounted setup significantly outperforms late fusion approaches, especially under low-light and adverse weather conditions. This work not only enhances the perception capabilities of smart roadside infrastructure but also contributes to the research community by providing publicly available methodologies and datasets to support further development in intelligent transportation systems (ITS).
Li-Ion batteries still prove to be difficult to extinguish throughout many experimental studies. The effectivity of different extinguishing agents has been analyzed in this study. Thermal runaway has been initiated in modules via external heat and different extinguishing agents have been applied on the modules to compare the effectiveness of their extinguishing properties. The study provides a reference for future choice of extinguishing agent in the event of a li-ion battery fire.
The present work addresses the parallel pack formation in multilevel inverters. In addition to examining the influence of the quantity of energy storage units within the parallel pack on energy losses, detailed investigation is also conducted into the impact of the resistance ratio between the switch and the energy storage unit paths. Besides the consideration of the potential savings in energy losses, which can be more than 15 % compared to a pure series mode, the problem of maximum current load on the outermost energy storage units in a parallel pack is explained in detail. Using the Helmholtz superposition principle, it is shown that, depending on the resistance ratio of the energy storage unit paths to the switch paths, a worst-case current divergence up to 100 % can occur between the mid and outermost storage units. Therefore, this paper additionally presents a parallel pack formation strategy that addresses this edge problem achieving a more balanced percentage of the edge storage units. This measure facilitates reducing the observed state-of-charge drift in alternative formation strategies to below 1 %.
These results provide valuable design guidelines for multilevel inverter systems, in particular regarding the optimal formation of parallel packs under dynamic operating conditions. The application of the developed radical algorithm enables improved energy efficiency as well as a more balanced utilization of the inherent energy storage units, which is of direct relevance for enhancing system reliability and lifetime in electric vehicle and stationary energy storage applications.
Das Video ist an der Technischen Hochschule Ingolstadt als Beitrag für den Best-Practice-Slam Informations- und Medienkompetenz, der von der Kommission Informationskompetenz von dbv und VDB am 12.03.2026 veranstaltet wurde, entstanden. Bei dem Wettbewerb hat der Beitrag den ersten Platz belegt.
Er zeigt auf unterhaltsame Weise den strategischen Rechercheweg, welchen die Teaching Library der THI in ihren Schulungen vermittelt. Ausgangspunkt ist die Lebenswelt der Studierenden bzw. Schülerinnen und Schüler. Schrittweise wird der Weg von der einfachen Suche zur komplexen Rechercheanfrage mit bibliothekarischen Mitteln gezeigt.
Driving simulators are indispensable tools in modern automotive research and development. However, the transferability of findings to real-world driving, and thus, the validity of simulator-based results, cannot be assumed without empirical validation.
In this study, we examined physiological (Galvanic Skin Response-based measures, Electrocardiogram-based measures, salivary cortisol) and cognitive (NASA Task Load Index, Short Stress State Questionnaire, single-item ratings) stress indicators by comparing a real-world driving circuit with seven distinct sections to a medium-fidelity driving simulator, applying a Bayesian analytical approach. The results present a mixed picture, with both absolute and relative validity observed for certain physiological and cognitive stress indicators. Overall, our findings suggest that stress responses in the simulator and real-world driving are comparable, although the simulator was subjectively perceived as more stressful.
These results provide valuable insights into the validity of simulators for stress research and underscore the need to consider individual differences, experimental conditions, and methodological approaches in future studies.
Diese Arbeit untersucht die Biomethanbereitstellungskosten entlang der gesamten Prozesskette bis zum Einsatz in hochflexiblen BHKW unter dem EEG 2023. Hierbei wird die Biomethanprozesskette techno-ökonomisch modelliert und gemäß der Annuitätenmethode (VDI 2067) bewertet. Die ermittelten Bereitstellungskosten einschließlich Netzanschlusskosten am Gasentnahmepunkt liegen zwischen 9,8 und 13,4 ct/kWhHs. Ergänzend wird der Einfluss eines Doppelmembrangasspeichers am Gasentnahmepunkt auf die Kosteneffizienz analysiert. Der Einsatz eines solchen Gasspeichers ermöglicht die Entkopplung des maximalen BHKW-Gasbedarfs vom maximalen Netzbezug, wodurch die maximale Gasbezugsleistung aus dem Netz begrenzt und leistungsabhängige Netzkosten gesenkt werden können. Den Kosteneinsparungen durch reduzierte Netzentgelte stehen zusätzliche Speicherkosten sowie strommarktseitige Mindererlöse infolge speicherbedingter betrieblicher Restriktionen gegenüber; letztere werden mittels MILP-basierter Fahrplanoptimierung quantifiziert und in die Kostenbewertung integriert. In den meisten Szenarien ist der Speichereinsatz kosteneffizient und senkt die Bereitstellungskosten je nach Netzentgeltstruktur um bis zu 1,7 ct/kWhHs. Diese Kostenreduktion hängt u. a. von Netznutzungsentgelten, Gasspeicherkosten und Strommarktpreisen ab. Darüber hinaus kann der Gasspeicher die Anzahl ökonomisch darstellbarer Biomethan-BHKW-Standorte erhöhen, weshalb diese Arbeit für die operative Umsetzung eine hohe praktische Relevanz entfalten kann.
This paper evaluates technical aspects of the Directives on periodic roadworthiness testing (2014/45/EU) and roadside inspection of commercial vehicles (2014/47/EU). Key findings focus on harmonising minimum content for Periodic Technical Inspection and Roadside Inspection requirements across Member States and establishing a cross-border data-exchange platform for registration, driver licence and inspection data. For remote sensing, comparative campaigns across Member States are recommended, noting its benefits while zero-emission vehicle uptake remains low. This document was provided by Policy Department B at the request of the TRAN Committee.
This study addresses limitations of traditional benchmarking methods for Retrieval-Augmented Generation (RAG) systems by proposing an evaluation framework for RAG-enhanced Large Language Models (LLMs). The framework structures evaluation dimensions and metrics, identifies suitable datasets and question types, and provides guidance for applying the framework in practice. A systematic literature review (SLR) was conducted, synthesizing evidence from 12 studies focused on the evaluation of RAG systems. The review employs a concept matrix to classify evaluative approaches and maps metrics to dimensions, evaluator types, and pipeline stages. In addition, we systematize dataset and question-type requirements that enable the proposed measurements and derive implementable evaluation guidance. The findings reveal substantial variation in evaluation practices and underscore the need for a multidimensional view. The framework integrates context relevance, faithfulness, answer relevance, correctness, and citation quality with corresponding metrics and links them to dataset prerequisites. It further outlines how the framework can be adapted to different RAG pipeline configurations, supporting use in real-world evaluation settings. The framework supports more systematic and transparent RAG evaluation design by consolidating dimensions, metrics, evaluators, and dataset requirements into a coherent structure. It offers actionable recommendations for selecting and operationalizing metrics and for integrating evaluation into RAG pipelines, thereby supporting the assessment and deployment of RAG-enhanced LLMs in dynamic environments.
Automotive software is becoming increasingly pivotal, enabling connected vehicles, advanced driver assistance, and the transition toward highly and fully automated driving. This workshop provides a forum to discuss current challenges and solution approaches in automotive software engineering. Participants will exchange suitable methods, techniques, and tools to cope with rising functional complexity and stricter demands on reliability, functional safety, IT security, and data protection.
