Live · plant telemetry

Your plant, running twice — once in steel, once in software.

DoTwin connects to the sensors you already have, learns how your equipment actually behaves, and tells you what breaks next — and what to change to stop it. Built for heavy industry, deployed on your own hardware.

Request a walkthrough See how twinning works
SAG MILL 01 · DIGITAL TWIN --:--:--
Health index
0.82
Remaining life
1840h
Mill power
2150kW
measured forecast alarm threshold
What makes it a twin

Two environments, kept in step.

A dashboard shows you what happened. A twin stays synchronized with the plant: data flows out from every unit on the flowsheet, and decisions flow back. The rate at which those two sides agree is the twinning rate — and it's the difference between reporting and operating.

Physical environment Feed Mill Cyclone Pump Physical processes Every unit on the flowsheet, running as it always has Operating state — vibration, current, torque, flow, density Virtual environment Virtual processes Feature engineering, forecasting, causal inference Digital state — the twin's belief about every unit One model registry, shared across the whole plant Data Synchronized at the twinning rate Information
Data out — raw tags from every unit, at the rate your sensors publish Information back — forecasts, alarms, setpoint recommendations
From sensor to decision

Five stages, one canvas.

Every stage is a node you drag onto a canvas and wire up. No notebooks to maintain, no scripts to hand off — the pipeline you draw is the pipeline that runs in production.

Ingest

Connect your tags

Stream from MQTT, or pull history from Postgres, MySQL and CSV. Your existing sensor tags, unchanged.

Engineer

Shape the features

Resample, roll, lag and split timestamps. Write formulas by clicking column names into place.

Train

Fit and tune

Pick a model, set a target, and search the hyperparameter space with grid or random search.

Evaluate

Check the numbers

Real ROC curves, confusion matrices and residuals from a real run — never a placeholder.

Operate

Watch it live

Register the model, bind it to a dashboard, and set alarm rules on the values that matter.

What you get

Four workspaces that share one model registry.

01 · Pipelines

Visual ML pipeline builder

Build a training graph node by node. The builder checks compatibility as you wire — a regression metric won't attach to a classifier, and a continuous target won't attach to a classification model.

02 · Twin

Live digital twin dashboard

Gauges, time series, flow-sheets and alarm logs, bound to live MQTT keys. Drop in your own plant SVG and pin readings onto the equipment they belong to.

03 · Causal

What-if analysis

Move a setpoint and see the predicted effect on throughput before you touch the plant. Driven by registered causal models, with confounders held constant.

04 · Federated

Federated learning

Train across sites without moving raw data off any of them. Pick the participating models, choose an aggregation strategy, and build a shared global model.

Deployment

Runs inside your fence line.

One Ubuntu server, one docker compose up. Process data never leaves the site network, and there's no per-seat licence metering how many engineers open the dashboard.

FastAPIMQTT / MosquittoRabbitMQ MinIOPostgreSQLMySQLDocker Compose
  • Air-gapped by default — nothing calls home, no external API keys required to run.
  • Every artifact lands in your own object storage: datasets, trained models, run results.
  • Training runs queue through RabbitMQ, so a long job never blocks the interface.
  • Sized for a single workstation-class server: 16 cores, 64 GB RAM, 1 TB NVMe.
  • Scales out by adding worker containers, not by rewriting pipelines.
Get started

Bring one asset. We'll twin it.

The fastest way to evaluate DoTwin is a single piece of equipment with a year of history behind it. Send us the tag list and we'll walk you through a working twin of it.

info@ai-twin.ir ai-twin.ir