The Department of Data-Driven Structural Dynamics is affiliated to the Institute of Structural Analysis and Design. Research and teaching activities in the fields of structural analysis, structural dynamics, materials, building envelopes and construction are bundled there in order to achieve a uniform basis for material-compatible design and construction. This includes fundamental theoretical work on calculation and design methods, material testing and simulation, material modeling, numerical simulations, safety theory, construction methods as well as applied research and development.

• Cost reduction in wind energy through improved service life and reliability calculations. Lifetime predictions for wind turbines are currently based on a large number of physics-based simulations, which means a high computational effort. This computational effort excludes the consideration of uncertainties and ensures that imprecise extrapolations are required. By replacing physics-based simulations with AI approaches, computing times can be reduced, uncertainties integrated and service life analyses made more precise.
• Measurement data-based assessment of bridges For the reliable use of bridge infrastructure, it is essential to assess the condition of many existing bridges in particular as they reach the end of their service life. Research is being conducted into how AI can be used to automatically detect and quantify changes in the condition of bridges based on the analysis of recorded measurement data. It is also being investigated how additional information can be obtained from this measurement data with the help of AI in order to refine the data as comprehensively as possible. An example of this is the automated determination of the position and size of the axle loads of trains.

Physics-informed machine learning, Machine learning with a small amount of training data (one-shot learning, semi-supervised learning), statistical machine learning, computer vision, Xplainable AI

The smooth transition between geometry and the properties of earth observation has become an integral part of modern photogrammetry and remote sensing. The importance of newly developed methods from the fields of machine learning, computer vision and data science, for example, has also grown considerably in recent years. The application of current methods to the various photogrammetric and remote sensing data is a central field of research. Data fusion, the processing of point clouds and semantic segmentation are a particular focus of the field.

The focus is on the evaluation of images and other sensor data of real scenes with reference to the earth's surface and the built environment.

• Detection/reconstruction of different environments, e.g. indoor, city, forest, etc.: Which AI models are suitable for specific remote sensing tasks?
  • Areas of application e.g. determination of forest parameters, scan-to-BIM. Deep learning is used here in various areas (e.g. detection of individual trees, fusion of different remote sensing data of different spatial resolution and sensor technology, e.g. hyperspectral data of medium resolution and high-resolution RGB data, determination of the health of vegetation, vector-based representation of indoor spaces)
  • Neural scene representations: How can multisensory data (e.g. LiDAR, multispectral, thermal) be fused with neural scene representation?
  • Determination of data quality: How can existing deep learning models be extended to enable the determination of the quality and evaluation of input images? In real data scenarios, data quality is now the limiting factor compared to the selected deep learning model

Object detection, semantic segmentation, development predictions, data acquisition, NeRF – Neural Radiance Fields, label quality, explainability, interpretability

In transport planning, methods and procedures are developed with interdisciplinary references and as part of interdisciplinary spatial planning. The focus is on the causes and effects of mobility and accessibility as well as integrating strategies for climate-resilient traffic management.

Traffic engineering looks at the fundamentals of traffic flow and the performance of transport facilities. This is supplemented by a comprehensive examination of modern traffic control technologies, traffic information systems and the collection, processing and evaluation of traffic data.

The focus is on disruptive mobility and the influence of artificial intelligence.

• Can intelligence be sustainable?
• How and why are mobility systems changing due to artificial intelligence?
• How and why are cities changing due to artificial intelligence?

scAInce lab: mobility of the future in virtual reality

We are researching how artificial intelligence and extreme events are changing our mobility systems and whether technological change is actually leading to sustainable mobility.

Disruptive mobility through new technologies and digitalisation

We are investigating how disruptive mobility influences our sustainability. We deal with the functional, digital and institutional networking of new transport systems, road transport technology, mobility services, information systems and artificial intelligence (AI).

I+I as well as E+E

The department researches solutions for material-specific increases in energy efficiency, the reuse of raw materials and the reduction of CO₂ emissions. The potential for future-oriented research lies increasingly in scientific work and technology applications in the micro and nano range. High-quality numerical modelling and experimental equipment can be used to find solutions for quantifying the service life behaviour of sustainable materials, for developing environmentally friendly and sustainable binders and for calculating the remaining service life of existing infrastructure structures.

• Application of AI strategies for pattern recognition to further optimise CO2 reduction and support researchers. (DFG: SPP 2436 – Net-Zero Concrete)
• Application of AI strategies to predict uncertainties regarding the durability properties of environmentally friendly materials for floating offshore platforms for PV systems. (EU H2020: NatureSea-PV – Novel Eco-Cementitious Materials and Components for Durable, Competitive, and Bio-Inspired Offshore Floating PV Substructures)
• Application of AI strategies to predict uncertainties in deconstruction options with optimised CO2 reduction and reusability of materials.. (PtJ ERAMIN 2: RecycleBIM – Integrated Planning and Recording Circularity of Construction Materials through Digital Modelling)

General AI development

As part of basic research as well as research and development contracts, the department is involved in the development of engineering solutions in the fields of ethohydraulics and water continuity, water development and maintenance, flood and heavy rain protection, thermal energy utilisation of water bodies, sediment management at dams, structural hydraulics, hydropower utilisation and ecohydraulics. Methodologically, physical and numerical models as well as AI-supported analysis and simulation routines are used.

• Generative design (forward and inverse) of parametric structural systems. Modern generative deep learning algorithms such as cGAN or cVAE are being researched, which can be integrated into human planning teams as “assistants” for the design of structural systems with parametric representation and allow faster yet robust design iterations.
• Scientific Machine Learning Various approaches for combining “classical” numerical methods such as the finite element method or the phase-field method with deep learning approaches for calculation-oriented problems are being worked on. The applications range from material mechanics (e.g. fracture pattern of glass) to the calculation of large-scale structures (such as bridges). Research on safety theory and the explainability of scientific machine learning methods is also being investigated, such as the derivation of partial safety factors for predictive machine and deep learning models or the “grounding” of deep learning architectures with knowledge of structural analysis in text form
• Neural and implicit geometry representation in connection with computer-aided analysis: The subject of investigation here is how complex and composite geometries can be represented with neural scene representation and fed to numerical methods for solving differential equations.
• Determination of data quality: Development of metrics to determine the quality and evaluation of (structured and unstructured) data, which can be collected in real or synthetic form. High-quality and representative data is currently the most significant factor for the training and usability of machine and deep learning methods.

Generative AI, physics-informed artificial intelligence, explainable and interpretable AI, causal AI, semi- und self-supervised training, computer vision, LLMs

As part of basic research as well as research and development contracts, the department is involved in the development of engineering solutions in the fields of ethohydraulics and water continuity, water development and maintenance, flood and heavy rain protection, thermal energy utilisation of water bodies, sediment management at dams, structural hydraulics, hydropower utilisation and ecohydraulics. Methodologically, physical and numerical models as well as AI-supported analysis and simulation routines are used.

• Analysing measurement signals and data from laboratory and field operations
• Recognition, typification and prediction of behavioural patterns in live animal tests
• Recognising and tracking fish species and movements in ethohydraulic tests

Signal data processing, image processing, analysing large amounts of data

Our research and teaching focuses on computer-based methods for modelling and simulating engineering tasks. This includes the conception, development and application of innovative procedures, methods and demonstrators of information and communication technology for planning, construction, utilisation and dismantling as well as recycling management of buildings and their interactions with the environment in the sense of the digital transformation of engineering activities.

In the course of the digital transformation of the planning, operation, dismantling and circular economy processes of structures (buildings and infrastructure), a large amount of heterogeneous digital data is generated that can be used with AI methods to supplement traditional analytical and empirical methods and close methodological gaps. In the course of the digital transformation of the planning, operation, dismantling and circular economy processes of structures (buildings and infrastructure), a large amount of heterogeneous digital data is generated that can be used with AI methods to supplement traditional analytical and empirical methods and close methodological gaps.

The central research questions with the corresponding research fields are:

• Which AI methods are best used for which engineering problems in construction and the environment? To this end, different machine learning methods are being researched in the context of “sub-symbolic” AI methods (i.e. machine learning to develop “intelligent” behaviours based on patterns and statistical analysis) in the sense of data-driven modelling of engineering systems.
  • Research projects:
    • DABKO (digitally optimised packaging planning of activated concrete structures in Konrad containers during the dismantling of nuclear facilities, AI: reinforcement learning)
    • KISStra (AI-based indicator-based safety management for the road traffic infrastructure, AI: Convolutional Neural Networks)
    • ZEKISS (condition assessment of railway bridges and vehicles with AI methods for the evaluation of sensor data and structural dynamic models, AI: Fully Connected Neural Networks, in cooperation with the Institute ISM+D at FB13)
    • INSITU (digital crime scene documentation, AI: computer vision)
• How can AI methods be further developed in hybrid form so that the explainability and comprehensibility of the results and findings generated with AI can be validated in the sense of an engineering-oriented validation? WoDa:Knowledge-orientated database system, To this end, hybrid explainable AI methods are being researched in relation to engineering problems. To this end, “symbolic” AI methods (i.e. the use of formal logic and rule-based systems to represent knowledge and make decisions) in combination with sub-symbolic AI methods are being developed in hybrid form to create explainable AI.
  • Research project:
  • WoDa: Knowledge-orientated database system

AI methods in sensor technology (need for cooperation with the IT and mechanical engineering departments); computing with AI hardware (need for cooperation with the IT and computer science departments); hybrid and explainable AI (need for cooperation with the computer science department); AI-suitable knowledge modelling (need for cooperation with the computer science department)

The Institute of Geotechnics is fundamentally concerned with both basic research and application-orientated research into engineering tasks. The fields of research range from soil-structure interaction in offshore wind turbines or large pile groups to the further development of numerical methods and applications of AI through to questions of the influence of climate change on geotechnical structures. We have high-performance research laboratories for soil mechanics tests as well as medium and large-scale geotechnical modelling tests and use the latest numerical methods, some of which we have developed ourselves, to solve geotechnical problems.

Due to the extensive data processing capabilities of AI systems, artificial intelligence in geotechnical engineering is expected to enable profound changes in analysis and forecasting, which can lead to improved design precision and risk assessment, for example. Geotechnical engineering also uses numerous ground improvement and construction methods in which construction machinery interacts with the subsoil and already collects extensive machine data.

• Intelligent soil compaction In an industrial project, an AI was developed for a manufacturer of construction machinery for soil compaction, which is able to determine the degree of compaction of the soil during the compaction process and indicate to the machine operator via a traffic light system when sufficient compaction has been achieved. The diversity of natural soils, the multitude of physical soil properties and the parameters influencing the compaction process and success in combination with a wide range of machine influencing parameters make this task very complex. Extensive field data was recorded and geotechnically interpreted on the basis of numerous test compactions and test fields, and a training database for the ML algorithm was created. The applicability of the system was demonstrated by building a prototype, which is now being prepared for series production.
• Automatic calibration of advanced material models Advanced material models for describing the stress-strain behaviour of soils are essential for the numerical investigation of complex geotechnical issues, such as earthquake safety and the long-term behaviour of offshore wind turbines. However, determining the model parameters is complex, time-consuming and requires a high level of expertise. In response to these challenges, we have developed software to automatically calibrate such models. Using heuristic optimisation algorithms, it determines a parameter set on the basis of laboratory tests that is superior to classical calibration in terms of robustness and predictive ability. A central challenge lies in the highly non-linear and discontinuous nature of the optimisation problem. We have developed a new optimisation algorithm for this purpose. In this field, we are therefore conducting research both with AI and on AI.
• Outlook The next steps are concerned with enabling automatic calibration of the advanced material models on the basis of field tests such as pressure sounding or a combination of field tests and laboratory tests. Another future field of research at the IfG is concerned with the complete or partial substitution of these material models by AI systems. Particular challenges here are the highly non-linear, path-, stress- and density-dependent material behaviour of soils.

Optimisation algorithms, physically informed optimisation, analysis of large amounts of data, pattern recognition