Vorträge

Electromagnetic Compatibility (EMC) and Signal Integrity (SI) are areas of electrical engineering that are vital to the functionality of all electronic devices and systems. EMC ensures compliance with internal and external electromagnetic interference. SI ensures error-free ransmission of data over wired interconnects; for example, on printed circuit boards. To address the highly complex, cross-disciplinary design for EMC and SI, decomposition, segmentation, and subsequent hierarchical, physics-based modeling are performed. Although an active field of research, machine learning (ML) has yet to become an inherent part of modeling. Given the limited availability of generalized datasets and highly complex tasks, the adoption of ML cannot be achieved using a brute-force, universal black-box approach. Instead, the modeling
hierarchy must be considered to introduce ML efficiently. Through analysis of EMC in electric vehicle powertrains and SI in high-speed interconnects we derive a general hierarchical modeling structure. Coupled with a review of ML methods and applications, we demonstrate how ML can be integrated into the established modeling workflow, demonstrated with hierarchical Gaussian process regression, neural networks, and a novel interface for circuit comprehension with large language models.

Zoomlink:
https://tuhh.zoom.us/j/81920578609?pwd=TjBmYldRdXVDT1VkamZmc1BOajREZz09

Vortrag (PDF, 147KB)

Radar technology enables the remote sensing of vital-sign parameters such as breathing and heart rate by measuring the temporal movements of the chest and micro-vibrations on the skin surface. The radar sensor compares the relative phase shift between the sinusoidal waveforms it transmits and receives upon scattering at the skin surface, and, thus, can resolve relative displacements orders of magnitude smaller than the wavelength. Typically, the resolution is in the single- to two-digit micrometer range. Operating without physical contact, the radar-based systems prove advantageous over state-of-the art medical instruments in terms of enabling continuous monitoring of neonates or patients after surgeries without introducing skin irritations. Although the fundamental feasibility has already been demonstrated by a multitude of proof-of-concept publications and clinical studies, it is still challenging to guarantee robustness of this method. This particularly holds true if the focus is on medically relevant parameters to be resolved in time domain such as the detection of the onset of individual heart beats. To understand the lack of robustness, a physically more accurate and, thus, mathematically more involved model of the interaction between the receive signal and the skin surface needs to be analyzed to account for nonnegligible wave propagation effects that previously have been ignored. According to this model, the radar receive signal effectively accords with the evaluation of an integral over the skin surface. This triggers the question of whether and how one could exploit the more accurate model to retrieve the skin surface motion more reliably and, thus, improve the robustness of radar-based vital-sign parameter estimation.

Zoomlink:
https://tuhh.zoom.us/j/81920578609?pwd=TjBmYldRdXVDT1VkamZmc1BOajREZz09

Differentialgleichungen (DGL) sind zentral für ingenieur- und naturwissenschaftliche Disziplinen. Es ist daher unverzichtbar, dass Studierende ein grundlegendes Verständnis dieser Gleichungen entwickeln. Dabei begegnen ihnen jedoch häufig Schwierigkeiten. Die Doktorarbeit untersucht, welche Schwierigkeiten Studierenden beim Verständnis der grundlegenden Konzepte von DGLs begegnen, entwickelt Lehrmaterialien, die das Verständnis fördern sollen und von Lehrenden unkompliziert eingesetzt werden können, und evaluiert die Wirksamkeit der Materialien.
Die Schwierigkeiten werden mit Hilfe eines Scoping Reviews und mehrerer Interviews mit Studierenden identifiziert. Übergreifende Themenbereiche werden mit qualitativen Forschungsmethoden ermittelt.
Als Lehrmaterialien werden sog. Tutorials entwickelt. In diesen werden Studierende in Gruppen in fachliche Dialoge eingebunden. Die Tutorials zielen darauf ab, Verständnisschwierigkeiten oder Fehlvorstellungen aufzudecken, Studierende damit zu konfrontieren und sie zu unterstützen, diese aufzulösen.
Um das Verständnis der Studierenden zu beurteilen werden wissenschaftliche Fragebögen zum Konzeptverständnis eingesetzt. Als Vortest wurde ein Test zum Grundverständnis im Bereich der Analysis übersetzt und weiterentwickelt und als Nachtest ein eigener Test entwickelt.
Im Vortrag erfahren Sie mehr über die drei Arbeitsbereiche der Promotion und sind eingeladen, sich intensiv in den anschließenden Ideenaustausch und die Diskussion einzubringen.

Zoomlink:
https://tuhh.zoom.us/j/81920578609?pwd=TjBmYldRdXVDT1VkamZmc1BOajREZz09

In this talk I will discuss recently proposed sequences of Gershgorin-type inclusion sets for the spectra and pseudospectra of finite matrices. In common with previous generalisations of the classical Gershgorin bound for the spectrum, our inclusion sets are based on a block decomposition of the matrix. In contrast to previous generalisations that treat the matrix as a perturbation of a block-diagonal submatrix, our arguments treat the matrix as a perturbation of a block-tridiagonal matrix, which can lead to sharp spectral bounds for particular matrix classes, e.g. for large Toeplitz matrices. Our inclusion sets take the form of unions of pseudospectra of square submatrices. In the Hermitian case these pseudospectra are unions of finite intervals centred on the real eigenvalues of these submatrices, and we show results obtained using open-source software written with ChatGPT.

This is joint work with Marko Lindner (Hamburg University of Technology).

Zoomlink:
https://tuhh.zoom.us/j/87285771127?pwd=bjlWT3AyQncwajZQN0l3dVd1WXJmZz09