Digital twins and process optimization
Mathematical models of process units and stages: regime forecasting, setpoint recommendations, optimization of energy consumption and reagent use. The effect is a 3–8% gain in process-stage efficiency with no capital expenditure. We work in operator-advisor mode — the model recommends, the operator decides — with read-only access to the historian and no intervention in the process control loop.
What problems it solves
Pain
Regime quality varies between shifts by 2–5% in target-product yield at the same feed rate
Solution
Real-time quality soft sensors plus a recommender model of optimal setpoints built on the historical “best shifts”
All shifts brought up to the best shift's regime: +0.5–1.5% yield
Pain
Energy and reagent overconsumption: regimes are chosen by procedure and experience rather than by the actual state of the process
Solution
A hybrid of a physical model and ML: a digital twin of the unit forecasts the process response to setpoint changes (what-if analysis)
−3–7% energy and reagent consumption
Pain
Full APC systems cost $1–3 million per unit and require months of step tests
Solution
Operator advisor (open-loop): recommendations on a separate screen, read-only access to the historian — no closed loop and no shutdowns
~80% of the APC effect at a fraction of the price, with no intervention in the control system
Pain
Waiting 4–8 hours for a laboratory analysis: the operator runs the unit “by yesterday's sample”
Solution
Soft sensors: real-time prediction of the quality parameter from process parameters
Control by actual values, not by a lagging laboratory
Pain
Heat-exchanger fouling: cleanings by the calendar, 1–3% of fuel consumption lost
Solution
Heat-transfer coefficient calculation + ML forecast of degradation + cleaning-schedule optimization
−10–20% fuel costs in feed-preheat sections
What data we work with
- Process control system historian (PI, PHD) — 2–3 years of tags
- Laboratory analyses (LIMS)
- Process procedures and regime charts
- Feedstock quality data
What the customer gets
- Shift-variability audit on a historian export — potential quantified in money
- Soft sensors for key quality parameters
- Operator advisor screen with setpoint recommendations
- A/B assessment of the effect across shifts over the pilot period
What a digital twin is in industry
It is not a 3D visualization of the plant but a working model of the process: a hybrid of physical equations and machine learning that answers in real time “what happens if we change this setpoint” and “which regime is optimal for the current feedstock”. On a typical process unit, regime quality varies between shifts by 2–5% — the twin brings every shift up to the level of the best one.
How the approach has been proven in Kazakhstan
Industry benchmarks are public. The “digital advisor” on a jigging machine at Donskoy GOK (ERG) cut chromium losses by 2.4% — about 123 million tenge (KZT) a year from a single machine. Samruk-Kazyna Ondeu in Stepnogorsk deployed a digital twin of its sulfuric acid plant with a 5–8% cost reduction. The combined effect of AI solutions across the ERG group in 2025 was 55.7 billion tenge. This is the same “advisor” architecture we build for customers without in-house IT teams.
How the solution works
- Data: 2–3 years of process control system historian data (PI, PHD, WinCC), laboratory analyses (LIMS), feedstock quality
- Models: quality soft sensors (gradient boosting/neural networks), a process-response model, a setpoint optimizer (Bayesian optimization / offline RL on historical “best regimes”)
- Operating mode: advisor — read-only connection to the historian, recommendations on a separate operator screen and in a report to the unit manager. A closed loop only as phase 2, at the customer's decision
- Deployment: on-premise. A read-only connection removes 90% of information-security and metrology objections
Why an advisor rather than full APC
Classical APC (Honeywell Profit Controller, AspenTech DMC3) costs $1–3 million per unit and requires months of step tests that disturb the process. An advisor built on historical data delivers 70–80% of the APC effect without step tests or shutdowns, in months rather than years — and applies to units where full APC is not economically justified.
Related
November 22, 2026Digest
Production data gap slows physical AI on the shop floor
Robotics and physical AI attracted almost $48 billion in 2026, but they will not reach the shop floor without connected production data. A review of the column by the CTO of VulcanForms: what exactly gets in the way and which pilot steps make sense for a plant.
Manufacturing
ReadNovember 21, 2026Technology release
Plataine released a manufacturing AI agent plugin for Claude: a breakdown for the plant
Plataine has made public a plugin of manufacturing AI agents for Claude: the language model asks questions about shop-floor data, while a separate optimization module performs the calculation. We look at what of this is applicable at a plant in Kazakhstan.
Manufacturing
ReadNovember 20, 2026Technology release
72% of manufacturers use AI, but only 10% at scale: three shifts in automation
AutomationWorld analysed three shifts in industrial automation over a 12–18 month horizon: moving quality control to the start of the cycle, spatial AI for the digital twin, and adaptive robots. The key barrier is legacy systems and fragmented data.
ManufacturingEnergy
ReadFAQ
Frequently asked questions
Short, straight answers to the questions we hear most often.
Who in Kazakhstan develops digital twins and operator advisors for process units?
Such solutions are offered by international APC vendors, the in-house engineering teams of large plants and local developers. 105 Industrial AI ("105kz" LLP, Astana Hub resident) builds digital twins of units and process stages — quality soft sensors and operator advisors that run read-only on top of historian data, on-premise, without touching the process control loop.
How is a digital twin different from an APC system?
APC (advanced process control) is a closed control loop that requires step tests, shutdowns and a budget of $1–3 million per unit. A digital twin in advisor mode works open-loop: the model recommends setpoints to the operator and the decision stays with the person. This removes 90% of information-security and metrology objections and delivers most of the effect at a fraction of the price.
Does it require intervention in the process control system?
No. Access to the historian is read-only; recommendations are shown on a separate operator screen and in a report to the unit manager. The solution does not intervene in the control loop.
How quickly is the effect visible?
The first measurable result comes at the data-audit stage: calculating shift variability from a historian export shows the potential in money before any models are built. The pilot effect is measured by an A/B comparison of shifts over 2–3 months.
Which production processes is the solution suitable for?
Continuous and semi-continuous process operations: refinery units, concentrators, CHP boiler units, chemical plants, mining and metallurgy process stages. The advisor platform is the same — industry-specific models are configured for the particular process stage.
How is a digital twin different from a process control system?
The process control system runs the process at the given setpoints; the digital twin advises what those setpoints should be under current conditions (feedstock, load, equipment condition). The twin does not replace the control system — it works on top of its data.
How much historical data is needed?
Ideally 2–3 years of historian tags and laboratory analyses. The minimum for a pilot is 12 months covering seasonality and the unit's different operating regimes.
Next step
Request a shift-variability audit
Tell us about your task — we'll come back with a data audit plan and an effect estimate within two business days.
- 1A 30-minute call: the task, your current systems, who makes the decision
- 2A review of data and processes, a staged plan with time and cost ranges
- 3A 2–4 week prototype or an 8–12 week pilot with success criteria agreed upfront
NDA from the first contact
Reply within one business day