Fraunhofer Institute for Mechanics of Materials IWM
Fraunhofer Institute for Mechanics of Materials IWM

Virtual development and testing using AI and accelerated simulation

Do you need fast and reliable answers for your next development decision? That’s exactly what we deliver: Using AI-based surrogate models, data-efficient prediction methods, and autonomous development assistants, we drastically shorten simulation and testing cycles without compromising physical robustness. We handle the entire process, from selecting suitable data to training and validating the models, all the way through to integrating them into your development processes. In doing so, we solve the specific problems that are holding you back: parameter spaces that are too large, insufficient test data, excessively long computation times, and overly expensive testing programs. The result isn’t academic prototypes, but operational tools that enable you to arrive more quickly at robust materials, processes, and component designs.

AI-based surrogate models 

We develop AI-based surrogate models that are trained using validated simulation and experimental data to replace time-consuming simulations and experiments in your design processes. This drastically reduces the number of simulation runs required, enables faster variant studies, and lowers the costs associated with finding optimal material and process designs.

R&D services:

  • Training of surrogate models based on your FEM or multiscale simulation data
  • Grey-box models that combine physical submodels with data-driven components
  • Uncertainty quantification and specification of the validity range of the surrogate models
  • Multi-objective optimization and variant studies based on the trained model
  • Integration of the models into existing simulation and development environments
  • Mehrzieloptimierung und Variantenstudien auf Basis des trainierten Modells
  • Integration der Modelle in bestehende Simulations- und Entwicklungsumgebungen

Research projects:

 

AI-supported interatomic potentials for multiscale simulation 

For manufacturers of optical components, electronics, and alloys: Component damage often results from atomic processes over time and temperature that, using classical methods, can only be calculated on tiny scales. Our AI-based interatomic potentials achieve quantum-mechanical accuracy at significantly larger time and length scales, thereby supporting your material design and service life prediction in the analysis of damage mechanisms, surface impurities, hydrogen/oxygen effects, and interfacial effects.

R&D services:

  • Training of AI-based interatomic potentials based on quantum mechanical reference data
  • Analysis of atomic damage mechanisms over larger time and length scales
  • Investigation of interfacial effects in alloys and composite systems
  • Evaluation of the effects of hydrogen, oxygen, and impurities on material properties
  • Coupling of potentials with multiscale simulations for material and service life evaluation

Research project:

  • Cross-scale, microstructure-dependent evaluation of intergranular cracking during high-temperature fatigue with dwell times in polycrystalline superalloys
    MMMCrack

 

AI-based translation of simple tests into high-quality material parameters

We use trained AI models to derive high-quality material parameters from simple, rapid test procedures, thereby replacing costly long-term tests. This allows you to reduce testing costs and shorten development times by obtaining reliable input data for design and simulation at an early stage.

R&D services:

  • Development of conversion models from short-term tests to long-term material properties
  • Combination of internal test data with open reference databases
  • Prioritization of critical material candidates prior to extensive testing programs
  • Validation of predictions using selected long-term tests
  • Integration of the derived material properties into your design and simulation processes

Research projects:

Data-efficient property prediction

We predict material properties based on a small number of sample data sets, internal databases, and physics-informed, data-sparse AI approaches—including by leveraging correlations between related material properties. This allows you to rule out unsuitable candidate materials more quickly, reduce trial-and-error, and obtain early estimates of contamination tolerances and service life.

R&D services:

  • Development of physics-informed prediction models for small, heterogeneous datasets
  • Utilization of correlations between related material properties
  • Estimation of impurity tolerances and critical property limits
  • Prioritization of material candidates before costly test series
  • Integration of internal databases to expand the predictive basis

Research project:

  • Tailored Materials Properties through Microstructure Optimization: Machine Learning Methods for Modeling and Inverting Structure-Property Relationships and Their Application to Sheet Metal Materials
    DFG-Projekt ML4MicroOpt

AI-supported material design

We optimize chemical compositions or microstructures for defined property profiles using AI and physics-based models, even for complex material systems such as metals, composites, and ceramics. This helps you shorten development times and reduce costs by minimizing the number of development iterations.

R&D services:

  • Optimization of chemical compositions for defined property profiles
  • Microstructure optimization for metals, composite materials, and ceramics
  • Coupling of AI models with physics-based material models
  • Identification of trade-offs between competing properties
  • Prioritization of promising material candidates prior to experimental validation

Research project:

  • Tailored Materials Properties through Microstructure Optimization: Machine Learning Methods for Modeling and Inverting Structure-Property Relationships and Their Application to Sheet Metal Materials
    DFG-Projekt ML4MicroOpt

AI-supported process design

We use hybrid and physics-based modeling to analyze the complex relationships between process parameters, microstructure, and material properties for process design in the forming of metallic materials. This reduces the number of trial runs and scrap, resulting in significant time and cost savings.

R&D services:

  • Modeling of process-structure-property relationships for forming processes
  • Identification of robust process windows based on hybrid models
  • Reduction of experimental trial runs through simulation-based preliminary evaluation
  • Identification of critical process parameters with the greatest influence on properties
  • Integration of optimized process windows into your production planning

Research project:

  • Tailored Materials Properties through Microstructure Optimization: Machine Learning Methods for Modeling and Inverting Structure-Property Relationships and Their Application to Sheet Metal Materials
    DFG-Projekt ML4MicroOpt

Research topic: Autonomous development assistance (self-driving labs)

In our research, we take an approach based on self-governing development frameworks in which AI agents independently plan, execute, and evaluate simulation and test campaigns—by integrating simulation tools, test benches, databases, and literature, and always with the approval of the responsible engineer. Looking ahead, this enables significantly faster qualification cycles, the systematic exploration of design spaces that cannot be covered manually, early detection of dead ends, and a traceable decision history.

R&D services:

  • Design of autonomous agents for simulation and test campaigns
  • Integration of test benches, simulation tools, databases, and literature
  • Definition of approval points for the responsible engineer
  • Exploration of large design spaces with automated planning and evaluation
  • Documentation of a traceable, auditable decision history