Digital twin — hit spec before the run
Caskent simulates distillation, maturation, energy, and line behavior so teams can select the safest policy before production is touched.
A living model of the distillery.
The twin combines equipment state, recipes, lab labels, sensory outcomes, warehouse conditions, and bottling constraints into a simulation workspace for operators and model policies.
Hit spec before the run.
Candidate cut points and operating envelopes are tested against expected congeners, methanol risk, hearts yield, and energy use.
- Compare heads, hearts, and tails windows before opening the valve.
- Predict congener movement against target product style.
- Reject policies that violate safety or quality limits.
What it simulates.
Distillation cut and congener profile
Heads, hearts, tails, ABV, volatile markers, methanol risk, and flavor targets.
Fermentation trajectory
Temperature, gravity, pH, yeast health, ester formation, and wash variability.
Cask maturation and angel's share
Extraction, oxidation, evaporation, warehouse position, and long-term inventory plans.
Energy, heat, and reflux
Steam demand, heat recovery, reflux strategy, and run-time tradeoffs.
Bottling-cell and robot paths
Fill, cork, label, inspection, rejection logic, and robotic movement constraints.
Simulate, select, commit.
- Step 1Thousands of trials
Simulate
Run candidate policies against site constraints, recipes, instrument history, and quality labels.
- Step 2Human approval
Select
Rank safe policies by spec fit, yield, energy, and operator preference.
- Step 3Bounded control
Commit
Publish the selected envelope to edge runtime with rollback and audit records.
Twin-tested autonomy policies.
Policies graduate only after passing simulated plant states, rare faults, recipe limits, and safety checks.
- Shadow mode compares model choices with operator actions.
- Assist mode proposes setpoints and requires approval.
- Graduated autonomy writes only inside signed envelopes.
Powered by simulation and optimization.
Hardware and simulation stack references are aspirational and selected per site architecture.
Omniverse and OVX simulation
Aspirational simulation workspace for plant layout, robot cells, sensor views, and operating scenarios.
cuOpt optimization
Optimization for scheduling, warehouse movement, cask selection, and bottling-cell routing where applicable.
Cosmos and Replicator rare-fault data
Synthetic data generation for low-frequency faults, edge cases, and inspection scenarios before production exposure.
Illustrative twin readouts.
Targets are validated per design partner and never treated as guaranteed outcomes.
The twin let us review the run before we changed the still. That made autonomy feel like an engineering process, not a leap of faith.
Twin inputs and outputs.
| Layer | Input | Output |
|---|---|---|
| Recipe | Mash bill, yeast, target style | Spec envelope |
| Process | Temperatures, flow, pressure, ABV | Cut and reflux policy |
| Quality | GC, spectral, sensory labels | Congener and risk forecast |
| Warehouse | Cask type, fill date, humidity | Maturation plan |
Digital twin questions.
Is the twin required for autonomy?
Yes for closed-loop actions. Shadow analytics can begin earlier, but control policies are tested in simulation first.
Does it need years of data?
No. Historical runs help, but live calibration, lab labels, and operator feedback can establish the initial site model.
Can it model maturation?
It forecasts cask behavior with uncertainty bands, then updates as tasting, loss, and warehouse data arrive.
Can operators override it?
Yes. Operators approve commit decisions and can pause or revert edge policies at any time.
Model your next run before it starts.
Bring one still, recipe, or maturation workflow into the Caskent design-partner program.