AI Enhanced System Modeling

XRG supports you in creating transparent and reliable AI solutions in the field of engineering. We combine the latest AI research with our many years of modeling experience: In our AI services and hybrid models, we keep a close eye on the underlying physics. Harness the potential of your measurement data for highly accurate system models and accelerate your simulations and optimization tasks by using AI-based surrogate models!
 

 

XRG Simulation combines the complexity of physics with stored data to ensure reliable simulation results. With our modular hybrid approach, we ensure consistency between the simulation scenario and the scope of training data at all times. We support you in implementing your requirements, from data analysis to the development of customised hybrid digital twins.

OPTIMISATION POTENTIAL FOR MANY APPLICATIONS AND INDUSTRIES

Hybrid and surrogate modeling can be used anywhere where measurement data from existing physical systems is available. This includes for example:

  • Manufacturing processes and machinery
  • Energy systems (power plants)
  • Mechanical systems
  • Transport systems of all kinds (vehicles, aircraft, ships)
  • Electromagnetic systems
  • Buildings and building automation
  • HVAC systems
  • Cooling and heating systems
  • Drive technology
  • Chemical processes
  • Fluid and hydraulic systems
  • Reactions and process engineering
  • Biotechnology, biology
  • Consumer goods (properties of products in general)
  • Natural processes (physical processes) of all kinds
  • See also here: Power plant, building model, brake hydraulics, vehicle cabin
  • …

HYBRID MODELS COMBINE PHYSICS AND DATA FOR EVEN MORE ACCURATE MODELS

We develop transparent, modular and adaptable hybrid models based on proven model libraries to meet your needs:

  • Use insights from your data directly in your simulation
  • Faster model development and calibration
  • Detailed component information already at the system level
  • Capture ‘hard-to-model’ properties in your system model using real measurement data
  • Ensure the predictive accuracy of your models for your entire simulation campaign by tracking the validity ranges of the training data
  • Avoid excessive retraining of AI parts through modular design and dimensionless input variables
  • Adjust the degree of hybridisation based on available data and simulation requirements: add physical equations to ensure training data validity
  • Coordinate data-driven AI equation-based model parts directly from Modelica with the XRG SMArtInt+ library
  • Make your AI model available in tools other than FMU

Surrogate models – AI-based surrogates for extensive simulation studies

We make information from large, detailed system models accessible for studies that require multiple simulation runs:

  • Accelerate your system models by replacing numerically expensive components with data-based surrogates.
  • Generate real-time capable surrogate models of your entire system.
  • Use generative AI for global model-based optimisation.
  • Calibrate your models using data with uncertainty information.
  • Automated generation of training data with DOE based on various sampling methods according to model requirements.
  • Learn model parameter dependencies with AI surrogates and perform real-time parameter sweeps and global optimisations
  • Create globally valid surrogates through AI-based model order reduction
  • Automated or semi-automated training of various architectures for different technical applications: from simple FFNN (Feed Forward Neural Networks) RNN (Recurrent Neural Networks) to NODE (Neural ODE), PeNODE (Physics Enhanced Neural ODE) and BNODE (Balanced Neural ODE)
  • Seamless Modelica integration and FMU export with the XRG library SMArtInt+