Physics-Based vs Data-Driven Twins

A good twin blends first-principles physics with data-driven learning — each covers the other's weakness.

In brief

• Physics models extrapolate but are slow and imperfect.
• Data-driven models are fast but only trust-worthy where they've seen data.
• Hybrid twins use physics for structure, ML for speed and calibration.
• Uncertainty quantification says how far the twin can be trusted.

The detail

Pure physics is principled but too slow for real-time and carries modeling error; pure machine learning is fast but blind outside its training distribution. Kronos's twins are hybrid: physics provides the backbone and conservation laws, learned surrogates provide speed, and data assimilation calibrates the twin to the real machine — with uncertainty quantification flagging where the twin is guessing.