Automated data-driven damage detection
The overall objective of the FOR 3022 is to gain a thorough understanding of an integrated structural
health monitoring (SHM) system in laminates with layers of large impedance difference using guided
ultrasonic waves (GUW) under real-world conditions. In this subproject we focus on automated damage
detection and merge the expertise from mathematics and computer science. As the basis for the automated methods serve mathematical
models built upon physical principles, mathematical tools to make the models computationally tractable and machine learning methods. Consequently, WG Lorenz (working on physics-informed
neural networks (PINNs)) and WG Gräßle (working on model order reduction and data assimilation) join WG Bosse (working on machine learning method) in this project.

Time period: 01.10.2023 - 30.09.2026
Leadership: Prof. Dr. Dirk Lorenz

Training Data Driven Experts in Optimization
The main goal of TraDE-OPT is the derivation and analysis of efficient optimization algorithms for solving data-driven problems with a wide range of applications, e.g. in social sciences, economics or healthcare. Data is now produced by numerous different sensors in industry, in vehicles, scanners, on the internet or by mobile devices and the production of data has virtually exploded. One of the new challenges is to extract interpretable and useful information from this data. A central tool for this is (especially convex) optimization.

Project webseite

Time period: 01.06.2020 - 01.12.2024
Leadership: Prof. Dr. Dirk Lorenz

Mathematics for Machine Learning for Graph-Based Data with Integrated Domain Knowledge
The aim of this project is to further develop and analyze deep neural networks for industrial problems that allow existing domain knowledge to be incorporated into the architecture of the networks. Such a hybrid approach can capitalize on the complementary respective strengths of end-to-end learning approaches and "a priori models/rules". This approach promises substantially more efficient solutions for many fields of application. For example, significantly less data is required or the predictions of the ML model are consistent with existing knowledge.

Time period: 01.04.2020 - 31.12.2023
Leadership: Prof. Dr. Dirk Lorenz