Fakultät Maschinenbau
Refine
Year of publication
Document Type
- Article (140)
- conference proceeding (article) (76)
- conference proceeding (presentation) (18)
- Part of a Book (11)
- Preprint (7)
- Book (5)
- conference proceeding (volume) (4)
- Doctoral Thesis (4)
- Other (4)
- Handout (3)
Language
- English (187)
- German (88)
- Multiple languages (3)
Publication reviewed
- begutachtet (239)
- nicht begutachtet (38)
Keywords
- Maschinelles Lernen (13)
- Fernwärme (12)
- Lastprognose (12)
- automated driving (12)
- deepDHC (11)
- KWK-Flex (10)
- advanced driver assistance systems (9)
- human factor (8)
- workload (8)
- KLEVERTEC / Vorlaufforschung (7)
Institute
- Fakultät Maschinenbau (278)
- IEAT - Institut für Energie- und Antriebstechnik (54)
- IFM - Institut für Fahrerassistenz und vernetzte Mobilität (25)
- KLEVERTEC (10)
- EPT – Institut für Effiziente Produktionstechnik (7)
- ECC-ProBell - Europäisches Kompetenzzentrum für Glocken (5)
- Fakultät Informatik (4)
- IFA – Institut für Innovative Fahrzeugantriebe (3)
- IMS - Institut für Maschinelles Sehen (2)
- Fakultät Elektrotechnik (1)
Using a Machine Learning Regression Approach to Predict the Aroma Partitioning in Dairy Matrices
(2024)
Aroma partitioning in food is a challenging area of research due to the contribution of several physical and chemical factors that affect the binding and release of aroma in food matrices. The partition coefficient measured by the Kmg value refers to the partition coefficient that describes how aroma compounds distribute themselves between matrices and a gas phase, such as between different components of a food matrix and air. This study introduces a regression approach to predict the Kmg value of aroma compounds of a wide range of physicochemical properties in dairy matrices representing products of different compositions and/or processing. The approach consists of data cleaning, grouping based on the temperature of Kmg analysis, pre-processing (log transformation and normalization), and, finally, the development and evaluation of prediction models with regression methods. We compared regression analysis with linear regression (LR) to five machine-learning-based regression algorithms: Random Forest Regressor (RFR), Gradient Boosting Regression (GBR), Extreme Gradient Boosting (XGBoost, XGB), Support Vector Regression (SVR), and Artificial Neural Network Regression (NNR). Explainable AI (XAI) was used to calculate feature importance and therefore identify the features that mainly contribute to the prediction. The top three features that were identified are log P, specific gravity, and molecular weight. For the prediction of the Kmg in dairy matrices, R2 scores of up to 0.99 were reached. For 37.0 °C, which resembles the temperature of the mouth, RFR delivered the best results, and, at lower temperatures of 7.0 °C, typical for a household fridge, XGB performed best. The results from the models work as a proof of concept and show the applicability of a data-driven approach with machine learning to predict the Kmg value of aroma compounds in different dairy matrices.
Effiziente Kühlung und Schmierung für Fräsprozesse: Frästechnologie und Hochdruckkühlschmierstoff
(2024)
Insbesondere bei der 5-achsigen Fräsbearbeitung existiert noch kein System, das die optimale bzw. minimale Hochdruckkühlschmierstoffmenge bei jeweils gegebenen Bearbeitungszuständen beschreibt.
Die Entwicklung, Umsetzung und Potentiale eines seriennah einsetzbaren, externen und vom Werkzeugmaschinenhersteller unabhängigen Systems zur optimierten Bereitstellung von Hochdruckkühlschmierstoff werden gezeigt.
Bisher existiert noch kein System, das die optimale bzw. minimale Kühlschmierstoffmenge bei jeweils gegebenen Bearbeitungssituationen unter besonderer Berücksichtigung der zu verarbeitenden Werkstoffe und deren signifikantem Verschleißverhalten beschreibt.
Die Entwicklung eines seriennahen unabhängigen Systems zur optimierten Bereitstellung von Kühlschmierstoffen mit angepassten Drücken für die jeweilige Werkstoffapplikationen wird gezeigt.
Diese Veröffentlichung beleuchtet die Herausforderungen und Lösungsansätze zur Bewältigung des Fachkräftemangels in der deutschen produzierenden Industrie. Der Mangel an qualifizierten Arbeitskräften verursacht erhebliche Kosten und verringert das Produktionspotenzial. Hauptursachen sind demografische Veränderungen und veränderte Wertvorstellungen der Beschäftigten. Die Attraktivität des Arbeitsumfelds soll durch die Sichtbarmachung individueller Beiträge und die Stärkung des Gemeinschaftsgefühls erhöht werden. Maßnahmen aus dem Umfeld von Large-Language-Models werden vorgestellt. Abschließend wird die Vision einer modernen, vollvernetzten und arbeitnehmerfreundlichen Produktionsstätte skizziert.
In an industrial context, AI-based methods are becoming increasingly important in the optical systems used for identification, inspection and classification. The reasons for this are that AI-based image processing algorithms are easy to use on the operator side and often achieve superior results. E.g. in complex classification tasks. In the sand cast industry, the complexity in optical inspection of cast parts is connected with strong variations in the local surface topography and in the global object geometry change. Despite the great potential of AI-based methods, application is often hindered by the immense effort involved in acquiring a suitable training dataset. This refers not only to the acquisition of the required number of images but also to the tedious labelling. In this work, we investigate the capabilities and limits of synthetic training data on an AI-based optical scanner used to identify and track cast parts. The optical scanner is capable of detecting and classifying a codification specifically designed for the casting industry. By reading the code, the scanner can deduce the specific number of the cast part. For synthetic image generation, we use physically based rendering, which has advantage of full control over all rendering parameters. This allows for both a systematic investigation of the importance of the parameters and, an automatic labelling process of the training datasets. Our results show that, in particular, a detailed geometric modelling of the local surface topography and global object geometry of the pins have a positive influence on the recognition rate of the neural network. With that accuracy rates up to 56 % are achieved using synthetic training datasets, only.
Large Language Models (LLMs) have become widely adopted recently. Research explores their use both as autonomous agents and as tools for software engineering. LLM-integrated applications, on the other hand, are software systems that leverage an LLM to perform tasks that would otherwise be impossible or require significant coding effort. While LLM-integrated application engineering is emerging as new discipline, its terminology, concepts and methods need to be established. This study provides a taxonomy for LLM-integrated applications, offering a framework for analyzing and describing these systems. It also demonstrates various ways to utilize LLMs in applications, as well as options for implementing such integrations.
Following established methods, we analyze a sample of recent LLM-integrated applications to identify relevant dimensions. We evaluate the taxonomy by applying it to additional cases. This review shows that applications integrate LLMs in numerous ways for various purposes. Frequently, they comprise multiple LLM integrations, which we term ``LLM components''. To gain a clear understanding of an application's architecture, we examine each LLM component separately. We identify thirteen dimensions along which to characterize an LLM component, including the LLM skills leveraged, the format of the output, and more. LLM-integrated applications are described as combinations of their LLM components. We suggest a concise representation using feature vectors for visualization.
The taxonomy is effective for describing LLM-integrated applications. It can contribute to theory building in the nascent field of LLM-integrated application engineering and aid in developing such systems. Researchers and practitioners explore numerous creative ways to leverage LLMs in applications. Though challenges persist, integrating LLMs may revolutionize the way software systems are built.
Clubfoot is a common congenital foot deformity that leads to constant pain and significant limitations if left untreated or not treated adequately. The most used method for treating clubfoot is the Ponseti method. It involves a correction phase where about five plaster casts are applied and changed weekly. This treatment lasting about 2 to 3 months, is the most chosen method due to its high success rate. However, treated babies often experience skin complications caused by stiff and tight casts. Previous research showed that viable solutions already exist including orthoses. In this research, a developed method known as VDI 2221 was applied and the printable orthosis using 3D printer was selected as an alternative to Ponseti method. Calculations and finite element method (FEM) analysis demonstrated that the orthosis made of PA6-CF provides sufficient stiffness and strength, assuming the weight force of the foot is 10 N. The selected design was developed based on requirements and functional analysis, effectively mitigating the disadvantages of the Ponseti method. The developed orthosis can be manufactured globally using the 3D printing process, with a manufacturing cost of around €150, excluding assembly costs. In summary, a new solution was proposed within the same treatment method, effectively eliminating skin complications, and enabling cost-effective manufacturability on a global scale.
Algorithms for causal discovery have recently undergone rapid advances and increasingly draw on flexible nonparametric methods to process complex data. With these advances comes a need for adequate empirical validation of the causal relationships learned by different algorithms. However, for most real and complex data sources true causal relations remain unknown. This issue is further compounded by privacy concerns surrounding the release of suitable high-quality data. To tackle these challenges, we introduce causalAssembly, a semisynthetic data generator designed to facilitate the benchmarking of causal discovery methods. The tool is built using a complex real-world dataset comprised of measurements collected along an assembly line in a manufacturing setting. For these measurements, we establish a partial set of ground truth causal relationships through a detailed study of the physics underlying the processes carried out in the assembly line. The partial ground truth is sufficiently informative to allow for estimation of a full causal graph by mere nonparametric regression. To overcome potential confounding and privacy concerns, we use distributional random forests to estimate and represent conditional distributions implied by the ground truth causal graph. These conditionals are combined into a joint distribution that strictly adheres to a causal model over the observed variables. Sampling from this distribution, causalAssembly generates data that are guaranteed to be Markovian with respect to the ground truth. Using our tool, we showcase how to benchmark several well-known causal discovery algorithms.
Energy storage is a crucial flexibility measure to temporally decouple power generation from power demand and is touted as the missing link in realizing a decarbonized energy system based on renewable energy. Energy storage capacity buildup at all levels of the global energy system is expected to accelerate the decarbonization process. To this end, a coherent mathematical framework to ascertain the carbon footprint of localized energy systems with energy storage is indispensable. This article presents an open-source energy system simulation program — Energy System Network (ESN). A variety of energy system configurations can be simulated with the Python program, which incorporates key energy system components such as generation, grid, storage, and loads. ESN features an integrated bottom-up approach that combines energy system modeling with streamlined life cycle assessment techniques to quantify the carbon footprint of all components in a localized energy system. The lifecycle phases of each component, including production, operation, and end-of-life treatment, can be considered. Carbon footprint values are obtained for two demonstrative case studies with lithium-ion battery applications: energy arbitrage and home energy systems. The metric Levelized Emissions of Energy Supply (LEES) has been used to evaluate the carbon footprint of each application. An unconventional energy arbitrage strategy designed to exploit the grid carbon intensity spreads instead of the energy price spreads manages to achieve a LEES value about 17% lower than the conventional variant. The influence of rooftop solar generation, battery energy storage system, and the energy management strategy on the LEES values for a home energy system is explored. A maximum LEES reduction of over 37% vis-á-vis the base scenario was observed with optimal energy management for the solar generation and the battery system. The open-source availability of ESN can contribute to transparency, comparability, and reproducibility in carbon footprint assessments of localized energy systems with energy storage.
Dispatch of a grid energy storage system for arbitrage is typically formulated into a rolling-horizon optimization problem that includes a battery aging model within the cost function. Quantifying degradation as a depreciation cost in the objective can increase overall profits by extending lifetime. However, depreciation is just a proxy metric for battery aging; it is used because simulating the entire system life is challenging due to computational complexity and the absence of decades of future data. In cases where the depreciation cost does not match the loss of possible future revenue, different optimal usage profiles result and this reduces overall profit significantly compared to the best case (e.g., by 30–50%). Representing battery degradation perfectly within the rollinghorizon optimization does not resolve this—in addition, the economic cost of degradation throughout life should be carefully considered. For energy arbitrage, optimal economic dispatch requires a trade-off between overuse, leading to high return rate but short lifetime, vs. underuse, leading to a long but not profitable life. We reveal the intuition behind selecting representative costs for the objective function, and propose a simple moving average filter method to estimate degradation cost. Results show that this better captures peak revenue, assuming reliable price forecasts are available.
The electrification of transportation modes such as cars, buses, and boats offers the potential of providing vehicle-to-X services during idle times. Pools of vehicles can provide balancing power, trade on the electricity market, or be used for load peak shaving. In this work, the usage patterns of electric cars, electric buses, and electric boats are investigated, and the provision of vehicle-to-X with these vehicles is simulated using an open-source simulation tool. A data analysis and a vehicle usage pattern assessment show that especially private electric cars behave predictably at night. It also reveals that the vehicle-to-X availability varies over the week for all vehicle types and is highest at night for cars and buses. During the day on weekdays, private cars are available for vehicle-to-X 30 to 70% of the time, the analyzed buses 5 to 50% of the time, and the availability of the boats depends on their primary use as ferries or private boats. If the three transportation modes provide vehicle-to-X during idle times, the equivalent full cycles that the lithium-ion batteries complete increase at different rates depending on the vehicle pool size, while the mean charging rates decrease. Furthermore, an exemplary aging analysis shows that the additional load of vehicle-to-X provision slightly increases the capacity loss of the car batteries compared to a paused unidirectional charging strategy.
Die Veröffentlichung beleuchtet die Herausforderungen und Lösungsansätze zur Bekämpfung des Fachkräftemangels innerhalb der produzierenden Industrie am Standort Deutschland. Der Mangel an qualifizierten Arbeitskräften führt zu erheblichen Kosten und reduziert das Produktionspotenzial. Hauptursachen sind der demographische Wandel und veränderte Wertvorstellungen der Beschäftigten. Es gillt, das Arbeitsumfeld Produktion attraktiver zu gestalten, indem individuelle Beiträge sichtbarer gemacht und das Gemeinschaftsgefühl gestärkt werden. Zudem soll die wahrgenommene Komplexität reduziert und die Autonomie des Fertigungspersonals erhöht werden. Beispiele wie ein digitales Ampelsystem und die Visualisierung individueller Beiträge verdeutlichen diese Ansätze. Die Zukunft der Produktion wird durch sieben Thesen skizziert, die die Bedeutung einer partizipativen Planung, die Veränderung der Anforderungen und die Notwendigkeit eines Kommunikationsraums betonen. Die Präsentation endet mit der Vision einer modernen, vollvernetzten und arbeitnehmerfreundlichen Produktionsstätte.
Driving simulators are used to test under reproducible conditions, however, they must be validated for each application. This guarantees that the gathered data on the simulator is representative of real vehicle data. This paper examines and compares objective data from 20 drivers that are recorded on a six degrees of freedom (DOF) high dynamic driving simulator and a passenger vehicle in the compact class on a proving ground. The purpose of this study is to investigate the comparability of the behavior of the subjects in their driving task on the driving simulator compared to the real driving test. The driving maneuvers include the 18 m slalom and an ISO double lane change (ISO 3888-2). The real car’s measurement setup is composed of an inertial measurement unit and access to the chassis CAN messages. The driving simulator is equipped with the same real electrical power steering as the test vehicle. Furthermore, a fully validated vehicle model is used in the simulation. Objective key performance indicators such as maximum steering wheel angle, steering wheel torque, lateral acceleration, yaw rate, and yaw gain deviate from around -18% to 10% in the slalom, with the majority of parameters not showing significant differences. Bigger differences are found for the double lane change. Overall, the results demonstrate a satisfactory degree of correlation between the driver behavior on the driving simulator and the real vehicle, even up to achieving absolute validity.
Comfort evaluation on a dynamic driving simulator with advanced tire, road and vehicle models
(2024)
The topics of automated driving and digitization are becoming increasingly im-portant and will shape the future of mobility. The potential of this technology is enormous. Concurrently, manufacturers want to continue to differentiate them-selves in driving characteristics typical of their brands. Rapid developments re-garding technological changes as well as legal regulations combined with short development times present new challenges for the entire automotive industry. In this context, virtualization and front-loading methods play a major role within the vehicle development. There has been a clear trend of pushing virtual devel-opment via simulation to reduce the number of necessary prototypes. Since however, both engineers and management still rely heavily on the crucial in-sights gained by real road tests, subjective closed-loop assessment must remain a part of this virtual process. Driving simulators have the potential to bridge these gaps, allowing engineers and test drivers to subjectively experience and assess new systems in an early virtual phase of development.
Kempten University of Applied Sciences is working with research and technol-ogy partners to research and further develop their dynamic driving simulator. With the goal to develop use-case specific methods for virtual vehicle develop-ment, the simulator’s novel motion platform is used specifically for research projects in areas requiring high dynamic performance such as vehicle dynamics and ride. This paper describes the methods and solutions developed in an R&D project investigating the simulator’s capabilities for ride comfort evaluation, such as primary & secondary ride. With the goal to enable experienced test drivers to perform a subjective ride evaluation in a very early development phase, the simulator’s real-time environment was extended with the highly so-phisticated tire model FTire. This paper provides an overview of the system’s performance regarding subjective ride assessment. It presents a brief insight into the detailed road modelling and describes the measures taken to ensure real-time capability of the individual model interfaces. Objective performance evaluation shows the benefit of this work for comfort evaluation in early phases of virtual development.
Data Mining und Knowledge Discovery in Databases (KDD) sind Forschungs- und Anwendungsgebiete, die sich mit der Extraktion von nutzbarem Wissen aus Daten befassen. Dazu werden unter anderem Methoden des maschinellen Lernens eingesetzt. Die Induktive Logikprogrammierung ist ein Teilgebiet des Maschinellen Lernens, dessen Gegenstand das Lernen aus multi-relational und prädikatenlogisch repräsentierten Daten ist, während andere Lernverfahren üblicherweise Daten voraussetzen, die in Form einer einzelnen Attribut-Wert-Tabelle vorliegen. Der Einsatz von ILP-Methoden für KDD und Data Mining wird auch als Relationales Data Mining bezeichnet. Ein Anwendungsgebiet von KDD und Data Mining ist die Teilgruppenanalyse. Dabei wird eine Population von Fällen, die in einer Datenbank repräsentiert sind, nach besonders interessanten Teilgruppen der Population durchsucht, indem mögliche Teilgruppen generiert und mithilfe geeigneter Interessantheitsfunktionen bewertet werden. Die vorliegende Arbeit hat sich zum Ziel gesetzt, Methoden zur sicheren Beschränkung des Suchraums bei der Suche nach interessanten Teilgruppen in multi-relationalen Daten zu erarbeiten und zu evaluieren. Dazu wird ein Verfahren zur Suche nach interessanten Teilgruppen in multi-relationalen Datenbanken entwickelt, das verschiedene Methoden zur Suchraumbeschränkung integriert. Die verschiedenen Methoden zur Suchraumbeschränkung werden in Experimenten evaluiert und das entwickelte Verfahren zur Bearbeitung eines echten Data Mining-Problems eingesetzt. Die Arbeit bietet im einzelnen: (1) eine Formalisierung der Teilgruppenanalyse im Rahmen der ILP, (2) Optimumschätzfunktionen zu ausgewählten Interessantheitsfunktionen, (3) eine Erweiterung des bekannten Apriori-Suchverfahrens zur Warenkorbanalyse, die es erlaubt, die von Apriori durchsuchte Hypothesensprache einzuschränken, (4) einen ILP-Sprachbias für die Teilgruppenanalyse, der die Anwendung der Teilmengenbedingung des Apriori-Suchverfahrens zur Beschränkung eines ILP-Suchraums erlaubt, (5) einen SQL-Sprachbias für die Teilgruppenanalyse in multi-relationalen Datenbanken, (6) einen Ansatz zur Integration der Suchraumbeschränkung anhand von Taxonomien in einen Apriori-artigen Suchalgorithmus, (7) eine Methode zur Behandlung diskretisierter numerischer Attribute, die die Suchraumbeschränkung anhand von Allgemeinerbeziehungen zwischen Intervallen vereinheitlicht mit der Suchraumbeschränkung anhand von Taxonomien, (8) Experimente zur Wirksamkeit der verschiedenen Möglichkeiten zur Suchraumbeschränkung, (9) die Anwendung der entwickelten Ansätze auf ein echtes Data Mining-Problem mit Bank-Daten und ausführliche Vergleiche mit verwandten Arbeiten. Die Experimente wurden mit einer prototypischen Implementation der in dieser Arbeit entwickelten Ansätze durchgeführt. Dabei haben sich Teilmengenbedingung und Optimumschätzfunktionen als wirkungsvolle und zuverlässige Methoden zur Beschränkung des Suchraums erwiesen, während der Beitrag der Taxonomien zur Suchraumbeschränkung zwischen verschiedenen Anwendungen stark schwankte und in einigen Fällen nur gering war. Ein wichtiges Ergebnis der Versuche ist, daß die Teilmengenbedingung, die bisher nur zur Suchraumbeschränkung in ein-relationalen Datenbanken eingesetzt werden konnte, für multi-relationale Datenbanken und ILP-Sprachen genauso wirkungsvoll sein kann wie für ein-relationale Datenbanken.
VBA bietet das Potenzial, effektive Digitalisierungslösungen mit geringem Aufwand zu realisieren. “VBA für Office-Automatisierung und Digitalisierung" zeigt mit vielen Codebeispielen die Automatisierung von Excel, Word, Outlook, PowerPoint, SAP ERP und SOLIDWORKS und das Zusammenwirken dieser Systeme. Auch Webservices und Rest APIs werden mit VBA angesprochen und erschließen interessante Möglichkeiten bis hin zu KI. Das Buch erläutert wichtige Konzepte und gibt viele Tipps, um VBA-Anwendungen mit einfachen Mitteln unternehmenstauglich und administrierbar zu gestalten.
„Hybride Montage“ als Antwort auf Modell Mix und Variantenvielfalt im produzierenden Mittelstand
(2024)
Die "Hybride Montage" als Kombination aus konventioneller Fließfertigung und innovativer Matrixproduktion offeriert vielfältige Potentiale, dem zunehmenden Maß an Variantenvielfalt in der produzierenden Industrie Rechnung zu tragen. Im Zeitalter des Industrial Metsverse und der damit verbundenen "Servitisierung" und des "Manufacturing as a Service" - d.h. der Güterproduktion als Dienstleistung - ermöglicht die Hybride Montage außerdem eine flexiblere Arbeitsplanung.
Die produzierende Industrie des DACH-Raumes erlebt im Jahr 2023 vielfältige Herausforderungen. Gerade scheint die Covid-Pandemie überwunden, welche die Vulnerabilität globaler Lieferketten schonungslos offenbart hat, treten neue Herausforderungen zutage. Veränderungen der gesetzlichen Anforderungen (wie EU Data Act), steigende Ansprüche an ökologische Nachhaltigkeit (z.B. Kreislaufwirtschaft) oder veränderte Kundenbedürfnisse (wie insbesondere Servitisierung) führen – bei einem konstant hohen Niveau an Variantenvielfalt – zu erheblichen technologischen Herausforderungen. Diese werden darüber vielerorts flankiert und im negativen Sinne überlagert durch einen erheblichen Mangel an Arbeits- und Fachkräften in der Produktion. Das Produktionssystem der Zukunft wird unserer Einschätzung nach daher nicht nur unternehmensübergreifenden Datenaustausch ermöglichen, Kreislaufwirtschaft befähigen und verstärkt kundenzentriert ausgerichtet sein. Es wird Arbeits- und Fachkräfte in der Produktion auf vielfältige Art und Weise „begeistern“. Auf diese Weise wird Fluktuation reduziert, generisches Wissen im Unternehmen gehalten und so die Grundlage für nachhaltigen Geschäftserfolg und technologischen Fortschritt sowie die langfristige Sicherung von attraktiven Arbeitsplätzen in der produzierenden Industrie geschaffen.
Nach unserer Überzeugung setzen sich in der produzierenden Industrie des DACH-Raumes diesbezüglich zwei Erkenntnisse durch:
1. Der „kritische Wettbewerb“ stammt selten aus Europa
2. Für die Bewerkstelligung dieser Herausforderungen sind authentische Impulse von außen – insbesondere von anderen, vergleichbaren Unternehmen – ein zentraler Erfolgsfaktor.
In dieser Gemengelage haben wir mit dem SUMMIT ALLGÄU eine Managementkonferenz „von der Industrie für die Industrie“ ins Leben gerufen. Im Zentrum des Veranstaltungskonzeptes stehen hierbei keine wissenschaftlichen Fachvorträge, sondern authentische Erfahrungsberichte hochkarätiger Referenten aus der industriellen Praxis. Beim ersten SUMMIT ALLGÄU Produktion am 23. und 24. Oktober 2023 in Marktoberdorf standen inhaltlich insbesondere die Themenkomplexe „Transformation & Nachhaltigkeit“, „Faktor Mensch in der Produktion“ sowie „Variantenvielfalt“ im Fokus. Auf überfachlicher Ebene wurden vor allem der persönliche Austausch und das Networking zwischen den zahlreichen Teilnehmern, Referenten und Ausstellern fokussiert.