Case study · Optimization
From next-morning KPIs to same-day decisions
A virtual sensor and advisory optimizer for Fatima Group’s PFL ammonia plant. It predicts daily Production and Energy Index from live process signals, then recommends safe setpoint moves that cut energy while holding production.
Predict the day. Recommend the move. Keep the engineer in charge.
Process Mix Optimizer · The design goal
The case file
From challenge to outcome.
Step 01
The challenge
Production and Energy Index were known only the next morning, with no view of whether the operating point was efficient and no quantified basis for trading production against gas burned.
Step 02
Our approach
Predict, then optimize. 127 DCS tags, KPI and lab reports and weather feed one daily pipeline; each day becomes about 890 features; a Ridge and LightGBM ensemble predicts the KPIs.
Step 03
The solution
An advisory layer that scores thousands of setpoint combinations in about 10 seconds. Moves are capped at ±3% per day, nothing writes to the DCS, and the engineer decides.
Step 04
From after-the-fact scores to an in-day virtual sensor
- 0.59%Average error on daily production, across 49 unseen days.
- 2.1%Average error on Energy Index.
- 0.93 R²Variance explained on unseen data.
FAQ
Questions about Process Optimizer.
The short answers, in the order teams usually ask them.
What is the Process Optimizer?
A virtual sensor that predicts the plant’s daily Production and Energy Index during the day, plus an optimizer that recommends safe setpoint moves.
Does it control the plant?
No. It’s advisory only. Nothing writes to the DCS, and an engineer approves every move.
What data does it use?
127 DCS tags from the historian, daily KPI and lab reports, and ambient weather.
How accurate is it?
0.59% average error on daily production and 2.1% on Energy Index, tested on 49 consecutive unseen days.
What if a recommendation is wrong?
Moves are capped at ±3% per day and stay inside proven operating ranges, so the downside of a bad day is an ignored suggestion.
Are the energy savings proven?
Phase 2 recommendations are pending field validation. A sample recommendation predicts a 4.5% lower Energy Index with production held in band.
Pilot the Process Optimizer
on one process unit.
Bring the challenge. Bring the ambition.
We’ll show you what it looks like on your own data.