The engine that keeps your process in sync.
At the core of Twynvex is a physics-based simulation engine that mirrors your reactor train in real time. Every 15 seconds, live sensor data is assimilated into a running process model — encoding heat and mass balance equations specific to your unit operations.
Request Pilot Access
How the physics model works
Traditional digital twins use statistical models trained on historical batch data. These work well in normal operating envelopes but fail silently when conditions shift beyond the training distribution — exactly when you need the model most.
Twynvex takes a different approach. We encode the actual governing equations for your process: enthalpy balances across heat exchangers, component mass balances for each species in a distillation column, kinetic rate expressions for your reactor. The model is derived from first principles, not inferred from data alone.
This means when feed composition shifts by 3%, the model responds the way your process does — not by extrapolating a regression it's never seen before.
The 15-second sync cycle
Every 15 seconds, the twin receives a batch of sensor readings from your plant. A Kalman filter state estimator reconciles the measured values against the model's predicted state, correcting for sensor noise and small model-plant mismatch. The corrected state becomes the initial condition for the next prediction horizon.
This continuous reconciliation loop keeps the twin aligned to your physical process — not drifting on its own dynamics.
Engine specifications.
| Parameter | Specification |
|---|---|
| Update frequency | 15-second twin synchronization cycle (configurable 5–60 sec) |
| Model fidelity | First-principles physics: heat and mass balance per unit operation |
| State estimation | Extended Kalman filter with tunable process/measurement noise covariance |
| ODE solver | Adaptive Runge-Kutta (RK45) with automatic step-size control |
| Supported unit operations | Distillation columns, reactors (CSTR, PFR), heat exchangers, evaporators, fermenters, lyophilizers |
| Prediction horizon | 2–6 hours (configurable based on process dynamics) |
| Alert latency | <200 ms from deviation detection to dashboard notification |
| Sensor handling | Noise rejection, missing-value imputation, drift compensation |
See the engine configured for your process.
We model your specific unit operations — not a generic template. The first conversation is a technical scoping discussion with a process engineer.