Disrupt with AI
Numerical simulation is a common tool to tackle engineering challenges in a safe and efficient way. By merging physics and IT, we manage to achieve approvements that each discipline cannot accomplish by itself. This is our core competence we would like to share - This is Digital Physics AI.
​
By utilizing machine learning and numerical simulations of digital twins, we have been supporting our customers for years in solving their engineering challenges. We are at home in the field of mechanical (Finite Element Method) and fluid (Computational Fluid Dynamics) simulations.
​
We support in problem definition, model building and execution of simulations. Afterwards we guide through data analysis and develop an individual AI model for your specific task. Realization of every project is carried out in steps and constant dialog with our customer. We do this, to make sure that we generate results faster and more efficient than with traditional approaches.
​
01
Problem definition
In the first step simulation models, experimental data or even just a problem description can serve as a starting point. Our team assists in creating a solid initial dataset.
We proceed in the following steps:
-
defining the problem
-
building the model
-
executing simulations (CFD or FEM)
-
creating the data matrix
02
Machine Learning
Creating a data set from your experiments or simulations is just the beginning. The created data sets the foundation for the machine learning. We train an individual AI model to accelerate predictions and increase precision. Not every combination of parameters has to be tested or simulated, since our AI model can describe a system faster and therefore increase efficiency.
Our approach for this:
-
prepare training data
-
choose AI configuration
-
execute machine learning
03
Increase precision
Finding global or local extrema with numerical simulations comes at increased computational cost. Thanks to our AI model, less simulation runs are needed to find them. After the evaluation of our AI based prediction with respect to calculated or measured reference values, the AI model precision gets optimized. Through this approach we are also able to identify gaps in data sets and close them for further predictions.
Our steps in short:
-
identifying and closing gaps
-
quantifying relevant factors
-
increase accuracy of prediction
04
Prediction
Our validated and optimized AI model can now serve two purposes. It cannot only be used to predict results, it can also predict input parameters for a desired output. Of course, we also process the results and give you a profound interpretation of them. We deliver either optimized parameters, or show you correlations with repsect to your specific problem.
We do:
-
execute AI prediction
-
process and interpret predictions
-
show correlations
