Search

Machine learning for design and analysis of steel frame halls The use of digital tools in design and construction processes increases efficiency, especially for recurring tasks and processes, such as the dimensioning of components or structures, which follow similar patterns. This paper develops an artificial intelligence (AI)‐based assistant for use in the pre‐design and design of steel frame halls as a prototypical co‐pilot. For this purpose, the necessary database of steel frame halls is first compiled from project archives of TRAGRAUM Partnerschaft Beratender Ingenieure mbB. This will be examined in the context of data analyses according to the decisive description parameters and similarities, so that finally suitable AI surrogate models for the prediction of the hall parameters are trained by means of regression analyses. Furthermore, an interface to the finite element program SOFiSTiK is implemented, which passes the predictions of the AI regarding the parameters of the steel frame halls to a parameterized input file and triggers the calculation and design. In the course of the calibration of the AI substitute models, in addition to the approximation qualities, possible safety margins to be provided are also discussed in the context of this article. This article shows that the prototypical co‐pilot for the preliminary design of steel frame halls could be successfully implemented. Furthermore the AI newly generated hall frames can be classified as stable with respect to the validation considerations. An outlook on future further developments of this approach as well as the transfer to other applications complete this paper. [ABSTRACT FROM AUTHOR]

Der digitale Wandel hat in den letzten Jahrzehnten die Geisteswissenschaften transformiert. Jede Art von Tätigkeit in Forschung und Lehre hat zumindest einen digitalen Anteil und sei es, dass dieser nur in der Nutzung digitaler Informationsinfrastrukturen für die Recherche besteht. In den historisch arbeitenden Wissenschaften mag der Beharrungswille größer sein als in anderen Disziplinen, aber auch hier haben sich in den vergangenen Jahren zwei Tendenzen herauskristallisiert: Erstens nimmt die Arbeit mit digitalen Surrogaten analoger Quellen zu – und wird noch mehr an Bedeutung gewinnen, je mehr die Fächer mit digital geborenen Quellen arbeiten. Zweitens werden die historischen Fächer selbst zu Produzenten von Forschungsdaten. Der Beitrag zeichnet diese Veränderungen nach und zeigt exemplarisch aktuelle Verfahren, Workflows und deren Voraussetzungen. Für die Gedächtniseinrichtungen argumentiert er, dass diese die sich abzeichnende Entwicklung nicht nur nicht ignorieren dürfen, sondern dass sie bereits jetzt ein wichtiger Teil des digitalen Umbruchs sind, der sich durch eine Konvergenz von Datenkompetenzen, aber auch von Verfahren und Modellierungen über die Grenzen der Fächer hinweg auszeichnet. Abschließend reflektiert der Beitrag, auf welche Weise Gedächtniseinrichtungen den Anforderungen dieser Veränderung und neuer Gruppen von Nutzenden gerecht werden können und welche Herausforderungen auch epistemischer Art in den bestehenden Sammlungen ruhen. The digital turn has shaped the humanities in recent decades. Every kind of activity in research and teaching now has at least a digital component, even if this only implies using digital information infrastructures for research. In the historical scholarship, the will to persevere time honed approaches may be greater than in other disciplines, yet even here two trends have emerged in recent years: firstly, scholarship based on datafied surrogates of analogue sources is increasing–and will become even more important the more scholars work with digitally born sources. Secondly, the historical disciplines themselves are becoming producers of research data. This paper traces the transformation, illustrating current techniques, workflows, and their prerequisites arguing that memory institutions should not only not ignore the development, but that they form already an important part of the digital transformation. For memory institutions the digital turn is characterised by a convergence of data competencies as well as procedures and models across the traditional boundaries of the disciplines. Finally, the article reflects on how memory institutions can meet the requirements of this transformation and new groups of users and what challenges, of an epistemic nature as well, are inherent to the existing collections. [ABSTRACT FROM AUTHOR]

KI/ML‐based Analysis and Interpretation of the IABSE‐Bridge Collapse Database Statistical analyses of bridge collapse data show that concrete bridges collapse significantly less frequently than bridges made of steel or wood. Since the main causes of bridge collapses worldwide are floods and associated fluvial processes, such as scouring, debris flows, etc. and impacts, it is reasonable to assume that the high dead load of concrete bridges leads to an overall more robust behavior in these events. This paper will examine whether the IABSE collapse database confirms this hypothesis and whether indications of further causes can be identified. For this purpose, the IABSE collapse database is examined using artificial intelligence and machine learning (AI/ML) methods. However, the AI/ML analysis does not confirm the previous thesis. The reasons for the rejection of the thesis, such as the representativeness of the data, are also discussed. An extension of the database for events with large numbers of collapses is recommended. [ABSTRACT FROM AUTHOR]

Explainable Domain‐Specific Artificial Intelligence for Bridge Engineering and Design The use of artificial intelligence (AI) is currently being discussed and evaluated across the board in research and practice in all industries. This paper first provides the necessary background knowledge on general AI and then elaborates the necessity of developing domain‐specific AI as well as explainable AI (XAI) for the calculation and design practice in civil engineering. The role of the civil engineering domain at the interface with AI in terms of data and modeling and design philosophy is discussed. The three examples of (i) computation of Kirchhoff slabs with physics‐informed neural networks, (ii) calibration of prior models for parametric bridge design from databases, and (iii) AI‐assisted conceptual design of bridges illustrate the application of domain‐specific machine and deep learning and XAI with special focus on concrete and bridge design. In particular, physics‐informed AI (piKI) represents an alternative to established numerical methods with special consideration of available simulation and experimental data and becomes explainable and comprehensible via the use of XAI. This paves the way for the use of piKI models in civil engineering as digital twins of a structure over its life cycles. [ABSTRACT FROM AUTHOR]

The Center for Data Analysis, Visualization, and Simulation (DAViS) at the University of Applied Science of the Grisons and the Swiss Center of Allergy and Asthma Research supports research on applications in topics like machine learning, big data, visualization, and simulation. The paper illustrates complex data analysis of life science data in a common research project. [ABSTRACT FROM AUTHOR]

This paper describes the development and validation of a phonetic-acoustic measurement procedure for a speech-based detection of sleepiness. The advantages of this automatic real-time approach are that obtaining speech data is unobtrusive and free from sensor application and calibration efforts. The chosen measurement process follows the speech-adapted steps of pattern recognition: (1) recording speech, (2) computation of 170 features describing prosody, articulation, and voice quality, (3) machine learning, and (4) evaluation. In a sleep deprivation study, a total of 380 simulated driver assistance samples ( n=32; 8:00 p.m.-4:00 a.m.) were recorded. One self and two observer assessments were used to obtain a Karolinska Sleepiness Scale (KSS) value, which served as an external validation reference. Features that proved to be especially sensitive to sleepiness are cepstral coefficients, formant bandwidth, intensity, and spectral measures. The best machine learning method, the support vector machine (SVM), achieved a significant validation correlation of r=0.46 in predicting sleepiness on unseen speakers. [ABSTRACT FROM AUTHOR]