Physicochemical blocks — material and heat balances, Gibbs reactor, thermohydraulics — plus industrial AI. No generative models, no black boxes, with metrological confirmation.
Temperatures, flows, pressures, modes; laboratory; met station
Material and heat balances, Gibbs reactor, thermohydraulics
Correcting blocks trained on mode archives and measurements
Mass emissions by component in real time
Checksums, self-tests, unreliability flags
Reading export, reporting, DSS integration
No new sampling systems: we use what production already measures. If signals are missing — minimal sensor upgrades.
| Parameter | Source | Role in the model |
|---|---|---|
| Gas and equipment temperatures | SCADA | Heat balance, thermohydraulics |
| Gas, fuel and feedstock flow rates | SCADA / inline flow meters | Material balance |
| Pressures and drafts | SCADA | Gas-path hydraulics |
| Feedstock and fuel composition | Laboratory | Gibbs reactor, stoichiometry |
| Equipment operating modes | SCADA | AI correcting blocks |
| Weather parameters | Met station | Dispersion model, DSS |
| Periodic measurements | Laboratory / mobile complex | Validation and protocol-based retraining |
The material balance closes the flows of feedstock, fuel and products; the heat balance closes the process energy. The Gibbs reactor computes the equilibrium gas-phase composition via Gibbs energy minimization; thermohydraulics describes the path from unit to stack.
On top of the physics, correcting blocks of industrial AI are trained on mode archives and instrumental measurements. This is not an LLM: the model does not learn in production; retraining follows a protocol with subsequent validation by the mobile complex.
This architecture yields explainable results: every computed value decomposes into contributions of physical blocks and corrections.
The computing core runs in an isolated environment: model and input-data checksums are verified before every calculation cycle, self-tests confirm integrity, and any interference is logged.
Data that fails checks is flagged as unreliable — the system fundamentally does not fabricate values. This is a key requirement for using results in supervision.
The platform is the Russian Astra Linux OS; the software is in the register of domestic software and suits critical infrastructure.
For comparison: a typical CEMS outfitting project takes 12–18 months.
2–4 weeks. Audit of sources, SCADA instrumentation, laboratory data. Applicability conclusion and exact estimate.
Assembly of the digital twin from physicochemical blocks; training of correcting blocks on the archive.
The mobile instrumental complex confirms 98–99% accuracy on real operating modes.
Registration in ARSHIN as an alternative-type measuring system, approval with Rosprirodnadzor, data delivery launch.
| SCADA | CEMS | Axioma (PEMS) | |
|---|---|---|---|
| Principle | shows process parameters | physical gas measurement | emission calculation from a process model |
| Output | temperatures, flows, pressures | concentrations at the sampling point | mass emissions by component, continuously |
| Accuracy | does not measure emissions | baseline | 98–99% relative to CEMS |
| Availability | high | 85–95% (verifications, repairs) | 99% |
| Cost of ownership | already on site | RUB 400–430 mn / 15 yrs / source | CAPEX ÷2, OPEX ÷5 vs CEMS |
| Role | data source | reference for validation | alternative-type measuring system |
A demo on your own data, or a paid preliminary survey. We reply within one business day.