SDU
Structural Dynamics Unit

From structural analysis to service life prediction

Extending the service life of civil engineering structures is a critical issue in the context of sustainable construction. Simplified calculation methods and assumptions can lead to large deviations between predicted and actual conditions, especially for dynamically excited structures such as bridges or wind turbines. One way to overcome this challenge and achieve realistic analyses and predictions is to use data-driven methods. The Structural Dynamics Unit is investigating how vibration measurement data can be used for computational models of the future. The focus is on bridges and wind turbines as well as their interaction with pedestrians, vehicles and the air.

At the intersection of Scientific Machine Learning (SciML), structural mechanics, and computational design, the Structural Analysis Intelligence research group is pioneering the next generation of engineering workflows. We bridge the gap between rigorous mechanical principles and advanced artificial intelligence to transform how structures are analyzed, designed, and maintained.

Our mission is twofold: to research advanced foundational AI for safety-critical structural engineering and to create intelligent agentic and classical AI systems that help engineers make faster, smarter, and more sustainable decisions from the earliest design stages through the full lifecycle of a structure.

Prof. Dr.-Ing. Michael Kraus,
Group Head Structural Analysis Intelligence

By embedding domain-Knowledge from structural engineering into modern artificial intelligence, we are unlocking a new paradigm of human-AI collaboration for a sustainable yet safe and reliable built environment.

Current projects

Partner:    Professorship Structural Mechanics, TU Darmstadt, 

Project Brief:

Estimating cost for complex bills of quantities (BoQs) in the construction industry has traditionally been a tedious, manual process requiring highly specialized labor. Today, companies face a difficult bottleneck: while market demand is high, a severe shortage of skilled personnel combined with intense pricing pressure has led to a massive influx of bid requests. Because only about 5% of these proposals successfully convert into actual contracts, firms waste significant resources on unsuccessful bids.

To solve this problem, this project aims to develop an AI-driven application that autonomously analyzes bills of quantities. The system will automatically source and aggregate data on products, personnel, and machinery from internal ERP systems as well as external suppliers like Schüco. By automating this workflow, the application is designed to cut estimation times by 85% and deliver 95% accurate, reliable cost estimates, even when dealing with early-stage project data with a low level of detail (starting at LOD 300).

Partner: Professorship Structural Mechanics, TU Darmstadt

Project Brief: The Dec KI project is developing an AI assisted design system for resource efficient solid building slabs. The project focuses on the early planning phase, where critical decisions regarding structural behavior, material utilization, economic viability, and ecological impact are made. By providing a conversational assistant system, the project aims to help designers select material efficient and climate conscious slab systems, thereby significantly improving the quality of early stage design decisions.

The system integrates four core methodological components:

1.A generative model utilizing Conditional Variational Autoencoders (CVAE) to learn the complex relationships between design parameters and technical, economic, or ecological performance.

2.Structural analysis tools, specifically connecting Dlubal RFEM and SoFiSTiK as agent tools via MCP or APIs.

3.A retrieval module that accesses standards, technical literature, and material data contextually to ensure all design recommendations are structurally and regulatory sound.

4.An AI agent architecture based on the Reason and Act principle, which manages the conversational interaction with human experts, breaks down complex planning queries into actionable steps, and coordinates the various tools, MCPs, and APIs.

Through this approach, the system not only generates design options but also provides transparent reasoning and verified technical backing for its decisions. This effectively bridges a major research gap, as early stage slab design currently lacks a knowledge based, dialogue driven assistant capable of combining generative modeling with verified engineering data. Ultimately, the project aims to accelerate the design process, maximize material efficiency, and reduce carbon emissions in structural engineering.

This development is a collaborative effort between TU Darmstadt, ETH Zürich, and industry partners from the structural engineering sector. The final outputs will be delivered as a prototypical, expandable system ready for use in research and professional practice.

Partner:  Institute of Steel Construction, RWTH Aachen University; Institute of Structural Design (KE), University of Stuttgart; Professorship Structural Mechanics, TU Darmstadt

Project Brief:

The AI-FACT research project is developing an innovative, data driven methodology to accurately predict the fatigue life of welded structural steel details, allowing engineers to efficiently unlock the structural reserves left untapped by DIN EN 1993-1-9.

Because historical test databases often suffer from significant documentation gaps, which introduce epistemic uncertainties, the project combines machine learning with domain specific engineering expertise. The core approach utilizes a Mixture of Experts architecture. Through a trainable gating network, this framework links purely data driven machine learning models with a physical, structural mechanics two phase model that accounts for crack initiation and propagation. To strengthen the dataset, synthetic data generated via Monte Carlo simulations is integrated into the system. This comprehensive approach reduces uncertainties, ensures robust model validation, and leverages explainable AI to build the trust necessary for regulatory approval, ultimately boosting competitiveness for small and medium enterprises.

Partner: Professorship Structural Mechanics, TU Darmstadt; ZM-I München GmbH

Project Brief: RABiT introduces a pioneering intelligent assistant system for construction planning designed to seamlessly extract and transform heterogeneous data, including domain expertise and existing building records. Specially developed RAG agents transform multimodal information — ranging from textbook and regulatory standards to company specific project data — into a unified, semantically enriched graph database. This database operates via RAG agents working in multidirectional cooperation with industry professionals to support planning and assessment processes as a SaaS solution.

In a 12 month feasibility study conducted by the ZM-I Group in cooperation with TU Darmstadt, the team is developing a prototype for structural engineering that can later be expanded to bridge construction. RABiT directly addresses critical industry challenges such as the skilled labor shortage and the backlog of building renovations. It achieves this by boosting efficiency, shortening engineering and planning times, increasing data transparency, and optimizing sustainability across the construction sector.

Head

  Name Contact
Prof. Dr.-Ing. Michael Kraus
Chair of Structural Analysis
+49 6151 16-23037
L5|06 620

Team

  Name Contact
Jing Han M.Sc.
Structural Intelligence Unit
+49 6151 16-23011
L5|06 642
Isamu Lautenschläger M.Sc.
Structural Intelligence Unit
+49 6151 16-23011
L5|06 642
Marcel König M.Sc.
+49 6151 16-23012
L5|06 642