Classical AI requires enormous amounts of data. We take a different approach: by directly integrating physical laws (such as conservation laws) into data-driven models, we drastically reduce the amount of data required while simultaneously increasing precision and robustness.
Advantages
- Less data, more precision: Efficient modeling through physics-informed prior knowledge.
- Reconstruction of inaccessible quantities: Determination of physical fields and parameters without direct measurability by leveraging physical laws.
- Increased robustness & generalizability: Reliable predictions despite incomplete or noisy data, as well as robust extrapolation to new operating conditions.
Core Applications: Flows & Combustion
In complex thermo-fluid dynamic processes, we reconstruct:
- Process variables: Velocities, pressure, temperature, and local compositions from experimental measurement data.
- System properties: Robust identification of species information and determination of flame transfer functions for characterizing dynamic system behavior.
