Deep physics based structural health monitoring
PD Dr.-Ing. habil. Stefan Kollmannsberger, Chair of Computational Modeling and Simulation, Department of Civil and Environmental Engineering
Dr. rer. nat. Felix Dietrich, Chair of Scientific Computing in Computer Science, Department of Informatics
The final outcome of this project will be a data-driven, robust, and computationally efficient framework for the detection of defects in civil engineering structures. The framework will be able to generate three-dimensional digital models of investigated structures of civil engineering – much like a CT-scan, but using sound waves. We will advance computational methods in the field of artificial intelligence and data science and apply them to the problem of structural health monitoring.
Finding hidden structural defects is a crucial task in civil engineering. However, to date the detection of such flaws from sensor signals is everything between time consuming and impossible, since the associated inverse problems are difficult to solve and often too ill-posed for practical applications in civil engineering.
To address this problem, we will investigate the full range of available options in data science, ranging from data-driven supervised learning to physics informed unsupervised learning. We will develop data-driven surrogate models to aid or fully replace the conventional full order approaches. The ill-posedness of the problem will be addressed by learning the regularization from data using neural networks. All methods will be evaluated first on benchmark examples and, in the final phase of the project, tested on structures of practical interest in civil engineering.