Today
Buffers set by convention
- Blend target set comfortably above specification by convention.
- Property behaviour approximated linearly in planning.
- Quality confirmed by the lab after the blend.
Every blend above specis margin you already paid for.
INSUS predicts finished blend properties such as octane from component data and recipes, states the uncertainty of each prediction, and helps size the quality buffer to what that uncertainty justifies. It works beside your existing blend optimiser. Quality staff certify product.
01
Property estimates with uncertainty, not just a linear index.
02
Buffer sized to modelled risk, not convention alone.
03
Advisory only. Alongside your optimiser; quality staff certify.
The problem
Buffer is a rational responseto uncertainty.
Quality giveaway is the gap between the product a refinery makes and the product it had to make. Being off-spec costs far more than being over-spec, so buffers widen. Octane blending is nonlinear, while planning models often rely on linear indices that miss component interactions. Lab results arrive hours after the blend is made. Some giveaway is structural (tankage, specification, inventory and measurement precision), while the rest can be addressed through process and scheduling. Margin earned in the process units is then surrendered at the last step.
01Recipe decisions during the blend
02Property estimate as the blend proceeds
03Lab certificate arrives hours later
- 01Recipe decisions during the blend
- 02Property estimate as the blend proceeds
- 03Lab certificate arrives hours later
The shift
From buffering against uncertaintyto measuring it.
The specification does not move, and the risk of off-spec product is real. The question is how much buffer the actual uncertainty justifies.
With INSUS
Buffers sized to modelled uncertainty
- Target set against a modelled property estimate with stated uncertainty.
- Nonlinear component interactions learned from the refinery's own blend history.
- Giveaway measured per blend and attributed to structural or addressable causes.
How it works
From component propertiesto a tighter target.
This fits OptiStudio as a property-prediction and uncertainty layer beside the existing blend optimiser. Access is read-only. It does not certify product or replace closed-loop blend control.
Connect
Link components, recipes and blend certificates
Connect read-only to component inventories and measured properties, blend recipes as executed, inline analyser readings, LIMS certificates for finished blends, tank and header conditions, specification limits by grade and season, and reblend history.
Understand
Learn how this refinery's blends actually behave
The model learns how finished properties have related to component properties and recipes in this refinery's own history, including the interaction effects linear blending indices miss.
Predict
Predict finished properties with an uncertainty band
For a candidate recipe, predict each finished blend property with an uncertainty band. The band is the product, because the justified buffer is a function of it.
Advise
Size the buffer to the risk you choose
Advise a recipe that targets the specification with the buffer sized to modelled uncertainty and the refinery's chosen off-spec probability. That probability is set by blenders and product quality staff as a commercial risk decision.
Verify
Report giveaway and off-spec rate together
Every prediction is checked against the finished blend certificate. The scoreboard is the giveaway distribution and the off-spec rate together. A system that reduces one by raising the other has failed.
Getting started
Start with one or two gradesand the certificates you already hold.
For refinery blending where quality buffers sit above specification by convention. Delivered through OptiStudio as a prediction layer beside the existing optimiser. Expert Services can help with data assessment and validation. Advisory and read-only. Discovery should establish whether headroom remains after the incumbent.
- What we need. 12 to 24 months of executed blend recipes with component properties, finished blend certificates, analyser data where available, specification history and reblend records. No new hardware for the core. Inline analysers improve what is achievable but INSUS does not supply them.
- What you see first. Your own giveaway distribution for one or two grades on one blend header, split as far as the data allows into structural and addressable causes, alongside an assessment of the incumbent optimiser, including if there is nothing worth pursuing.
- Where the limits are. Advisory only. Does not certify product or replace closed-loop blend control. A large share of giveaway can be structural. No giveaway-recovery percentage is quoted before seeing your certificates.
Related for processing
Other processingstarting points
Blend quality giveaway sits beside preheat fouling economics and amine circulation on the same processing site.
- 01Preheat Train Fouling & Cleaning AdvisorTrack heat exchanger fouling in crude preheat trains and advise cleaning timing against fuel and lost production.
- 02Amine Circulation & Regeneration AdvisorRecommend the lowest solvent circulation and reboiler duty that still protect treated-gas quality and rich loading.
- 03Processing industryThe wider picture for treating, recovery and conversion: amine, NGL, fractionation and the energy work underneath product quality.
Questions blending and quality teams ask us
Blenders hold a quality buffer because blend properties are nonlinear and lab results arrive after the blend. This use case predicts finished properties with stated uncertainty so the buffer can be sized to modelled risk for quality staff review.
12 to 24 months of blend recipes as executed with component properties, finished blend certificates, analyser data, specification history and reblend records. No new hardware for the core.
A property-prediction and uncertainty layer feeding it. If your incumbent already predicts well and your giveaway is mostly structural, there is no headroom, and discovery should tell you that.
Off-spec rate is a hard constraint and a reported KPI. The acceptable off-spec probability is set by your quality staff, not by the model.
No. Blenders and quality staff make the decisions. Product certification and release stay with quality staff. Closed-loop blend control stays with your existing system.
Through OptiStudio as a prediction layer beside the existing optimiser. Expert Services can support data assessment, integration and validation with your blending team.
A large share of giveaway can be structural. No recovery percentage is quoted before seeing your certificates. The category has well-established incumbents, so the case is measurable headroom over the existing workflow.
Processing · Product blending
Before you reduce giveaway, measure it.
The first deliverable is your own giveaway distribution, split into what is structural and what is not, including if the answer is that there is nothing here worth doing.