Data-Driven Nonlinear Modelling and Control of Structures

• Applicants: Professor Dr.-Ing. Dirk Abel; Professor Dr.-Ing. Okyay Altay
• Subject Area: Applied Mechanics, Statics and Dynamics
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 538131242

This research project aims to establish a multi-level modelling methodology for civil engineering structures, focusing on nonlinear dynamic response and its control. In the pursuit of sustainable and economical design, vibration control devices, including dissipators and tuned mass dampers, are employed to mitigate effects of dynamic loads. These elements exhibit high energy dissipation, offering dynamic resilience and protection against brittle failure. However, modelling of these control devices is challenging due to intricate nonlinear micro-scale physical processes. Moreover, modern structures incorporate (semi-)active devices capable of responding to changing structural and environmental conditions, such as degradation, natural disasters, and user demands. The control of these sophisticated devices necessitates highly adaptive strategies that consider the dynamic processes and account for uncertainties. In this interdisciplinary endeavour, we strive to devise an autonomous modelling approach for vibration control devices using artificial neural networks (ANNs). Our first objective is to create an active learning-based data generation method, guiding ANNs in identifying the most informative data for training through human expertise. Secondly, we aim to reduce human intervention by establishing a cyber-physical framework, directly coupling ANNs with testing apparatus, such as shaking tables and actuators. This approach extends classical performance assessment procedures into autonomous modelling. Our third goal centres on crafting a strategy for (semi-)active vibration control of civil engineering structures. Addressing model uncertainties and structural changes, we develop an adaptive modelling approach utilizing Gaussian Process Regression (GPR). This approach, integrated into a dual model predictive control algorithm, computes vibration control forces. GPR enables the model to replicate different structural states. Our fourth objective involves designing a state observer for the control algorithm. This observer estimates the displacement state of the structure using displacement responses from control devices and conventional acceleration measurements. This project’s overarching aim is to establish a methodological framework enabling structures to autonomously sense, analyse, and adapt their dynamic response. The theoretical insights of the project are validated numerically on benchmark buildings and experimentally both on a tuned liquid column damper and a nonlinear shear-frame structure. Validation involves real-time hybrid simulations and shaking table tests.

• DFG Programme: Research Grants
• Co-Investigator: Professorin Dr.-Ing. Heike Vallery