Polymer plants: quality known hours late, long grade transitions

Melt index and density come from the lab only every few hours

What happens. The melt index and the density of the polymer are measured on samples in the lab. Between two samples the reactor runs on what the operators hope is the right setting.

What it costs. Off-grade material produced between the samples, which is sold at a lower price or reworked.

Why the usual fix fails. Taking samples more often costs staff and still gives results with a delay. Control from the lab value alone reacts to something that happened hours ago.

How a fixed-price remote diagnosis checks it. From reactor conditions (temperatures, pressures, feed ratios, catalyst and hydrogen data) and the lab history I check how well melt index and density can be computed continuously. The fixed-price remote diagnosis tells you if the value is good enough for closed-loop control, as it is used for this task in polymer plants.

Background: Polymer melt flow index soft sensor

Transitions between grades take too long

What happens. Moving from one grade to another takes many hours, because the operators move the variables stepwise and wait for the lab.

What it costs. Off-grade tonnes during each transition and capacity lost to them.

Why the usual fix fails. A fixed recipe for the transition is conservative. Faster moves risk upsets and a second round of off-grade material.

How a fixed-price remote diagnosis checks it. From past transitions (setpoints, responses, time to spec) I check how fast the plant can move safely and what an optimised path looks like. The fixed-price remote diagnosis tells you how many hours per transition might be won.

Background: Trajectory optimisation and setpoint generation

The exothermic reaction runs near its runaway limit

What happens. The polymerisation gives off heat, and the cooling has limits. Operators keep a distance to the limit that is set by caution rather than by calculation.

What it costs. Lost rate, since the reactor is held below what it could do, and the risk of a trip if the limit is reached anyway.

Why the usual fix fails. A standard temperature controller reacts after the temperature rises. More margin is the safe answer, but it takes production.

How a fixed-price remote diagnosis checks it. From temperature, cooling and feed data I check how close the reactor can be run to the limit with a controller that looks ahead. The fixed-price remote diagnosis tells you how much of the margin predictive control could safely use.

Background: Predictive control of process plants

Catalyst activity varies from lot to lot

What happens. A new lot of catalyst has a different activity. The reaction rate and the quality move, and the operators correct the feed.

What it costs. Rate swings and quality variation after every lot change.

Why the usual fix fails. A fixed catalyst feed assumes an average activity. Correction after the lab result is too late and overshoots.

How a fixed-price remote diagnosis checks it. From the reactor’s heat balance, rate and the product data after past lot changes I check whether activity can be estimated and the feed adapted. The fixed-price remote diagnosis tells you how well that works on your data.

Background: Adaptive predictive control

The molecular weight distribution is not measured online

What happens. The distribution of the chain length is measured in the lab, if at all, and not in the running process. It decides stiffness, processing and clarity.

What it costs. Customer complaints about processing that the standard properties did not predict.

Why the usual fix fails. Melt index alone does not describe the distribution. Online instruments for it are rare and need upkeep.

How a fixed-price remote diagnosis checks it. From reactor conditions and the lab distributions I check whether the chain length can be estimated with a model of the reaction. The fixed-price remote diagnosis tells you how far that carries.

Background: Computed measurements

Residual monomer in the product is known late

What happens. After the devolatilisation, the residual monomer in the product is measured with a delay. Odour or a rise in emissions is noticed before the number is.

What it costs. Emissions, odour complaints and rework of material that has already been bagged or stored.

Why the usual fix fails. Running the devolatilisation hot and long is safe but costs energy and can harm the polymer. Lab values come too late to steer.

How a fixed-price remote diagnosis checks it. From temperatures, vacuum, residence time and the lab results I check whether residual monomer can be computed online. The fixed-price remote diagnosis tells you how well it follows the lab.

Background: Computed measurements

Pellet quality and melt temperature swing with throughput

What happens. At the extruder and pelletiser, the melt temperature and the pellet shape change when the throughput changes. Fines and irregular pellets follow.

What it costs. Fines, rejected lots and extra work with the cutter and the die.

Why the usual fix fails. Loops for speed, temperature and cutter run separately, and each change in one disturbs the others.

How a fixed-price remote diagnosis checks it. From extruder, die and cutter data and the pellet quality I check the coupling and whether coordinated predictive control would keep the quality steadier. The fixed-price remote diagnosis tells you what to expect.

Background: Predictive control of process plants

The viscosity end point of resin batches is found by sampling

What happens. In a resin batch, the operators take samples to see whether the viscosity has reached the target. The batch continues until it has.

What it costs. Long batches, because the reaction runs on while waiting for a result, and variation of the end point.

Why the usual fix fails. Sampling more often disturbs the batch and costs staff time. A fixed time ignores the differences between batches.

How a fixed-price remote diagnosis checks it. From batch temperatures, stirrer power and the sample results I check whether the viscosity can be estimated and the end point set by it. The fixed-price remote diagnosis tells you how much batch time it could save.

Background: Computed measurements

Recycled feedstock changes the behaviour of the reactor

What happens. When recycled feedstock is added to the process, the reaction behaves differently. Impurities and a changing composition affect activity and quality.

What it costs. Quality problems and process upsets, which limit how much recycled material can be used.

Why the usual fix fails. Fixed recipes assume a clean feed. Waiting for the lab to show the effect means the batch has already been affected.

How a fixed-price remote diagnosis checks it. From feed records, reactor data and lab results I check what can be estimated about the feed from the process signals and how control can adapt. The fixed-price remote diagnosis tells you how far you can go with recycled share.

Background: Adaptive predictive control

Refrigeration and steam for the reactor run conservatively

What happens. The cold and the steam for the reactor are provided with a margin, so that the reactor has enough at the worst moment. Most of the time the margin is not needed.

What it costs. Energy for the compressors, the steam and the pumps.

Why the usual fix fails. Reducing the margin by hand risks the reactor at the next peak. The utilities and the reactor are controlled separately.

How a fixed-price remote diagnosis checks it. From utility and reactor data I check how much of the margin is used and what economic operation at the constraint would save. The fixed-price remote diagnosis tells you whether it is worth it.

Background: Economic predictive control

Reactor wall fouling grows unnoticed

What happens. A layer grows on the reactor wall and reduces heat removal. It is not visible, and the temperatures change slowly.

What it costs. Lost heat removal and rate, and shutdowns to clean the reactor.

Why the usual fix fails. Cleaning on a fixed interval is either too early or too late. A single temperature alarm comes when the layer is already thick.

How a fixed-price remote diagnosis checks it. From the heat balance of the reactor over past campaigns I estimate how fast fouling grew and when it should have been acted on. The fixed-price remote diagnosis tells you if you can track it online.

Background: Computing unmeasured states

The grade sequence is planned by habit

What happens. The order of the grades follows what has always been done. Each transition between two grades costs something, but the cost is not part of the plan.

What it costs. Transitions that could be avoided and inventory that builds up.

Why the usual fix fails. Spreadsheet plans consider orders but not the transition cost between pairs of grades.

How a fixed-price remote diagnosis checks it. From orders, grades and the known transition losses I check what a campaign schedule could save. The fixed-price remote diagnosis tells you how large the gain is.

Background: Trajectory optimisation and setpoint generation

Silo blending to hit the specification is done by rule of thumb

What happens. Lots of different quality are blended in silos to land within the specification. The mix is decided by rule of thumb.

What it costs. Off-spec lots and more lab checks, with material blended more than necessary.

Why the usual fix fails. The rule assumes that the lots are as stated. The real quality of each lot is estimated roughly.

How a fixed-price remote diagnosis checks it. From production data and lab values per lot I check how well lot quality can be estimated, and then optimise the blend. The fixed-price remote diagnosis tells you how many lots would end up within specification.

Start-ups produce off-grade material for hours

What happens. After a start-up, the reactor takes hours to produce the right quality. The operators bring conditions up stepwise.

What it costs. Waste and a loss of rate.

Why the usual fix fails. A fixed start-up procedure is slow because it must be safe in all cases. Experience of the best operators is not written down.

How a fixed-price remote diagnosis checks it. From past start-ups I check where time was lost and what a better trajectory would look like. The fixed-price remote diagnosis tells you how much waste it could save.

Background: Trajectory optimisation and setpoint generation

Bed level and fluidisation in gas-phase reactors are hard to see

What happens. In a gas-phase reactor the state of the fluidised bed cannot be seen directly. Pressure differences and temperatures are the only hints.

What it costs. Sheeting on the walls, forced shutdowns and quality variation.

Why the usual fix fails. Alarm limits on single signals come late. The bed state depends on many things at once.

How a fixed-price remote diagnosis checks it. From the reactor’s pressure, temperature and flow signals I check whether the bed state can be estimated and early signs of sheeting found. The fixed-price remote diagnosis tells you what could be warned about.

Background: Computing unmeasured states

Filler content varies in compounding and the properties drift

What happens. In compounding, the share of filler in the product varies. The mechanical properties drift with it.

What it costs. Rejects and complaints.

Why the usual fix fails. Feeders are calibrated, but the real flow varies. The content is checked later in the lab.

How a fixed-price remote diagnosis checks it. From torque, pressure, energy and the lab content values I check whether the filler content can be computed and the feeders adjusted. The fixed-price remote diagnosis tells you how well that works.

Background: Computed measurements

The online rheometer is down for maintenance and the plant runs blind

What happens. The online rheometer that measures the melt properties is out of service for maintenance or a fault. The plant continues without it.

What it costs. Operation without a measurement, with more off-grade material.

Why the usual fix fails. Waiting for the repair means running blind. A spare instrument is expensive.

How a fixed-price remote diagnosis checks it. From the data recorded while the instrument works I check whether an estimate can bridge the gaps. The fixed-price remote diagnosis tells you how well it would follow the real value.

Background: Computed measurements

A new, larger line copied the old control

What happens. A new, larger line was built with the control concept of the old one. The dynamics are different, but the settings are the same.

What it costs. Unstable operation and a long time until the line runs well.

Why the usual fix fails. Retuning on the running line costs production. The old logic does not fit the new scale.

How a fixed-price remote diagnosis checks it. From the new line’s data and the design I check what has changed in the dynamics and what control concept suits it. The fixed-price remote diagnosis tells you where to start.

Grade recipes depend on senior operators

What happens. Senior operators adjust the recipes for each grade. Their way of doing it differs from shift to shift.

What it costs. Variation between shifts and a dependency on a few people.

Why the usual fix fails. Recipes on paper show the base values, not the adjustments. Training takes years.

How a fixed-price remote diagnosis checks it. From the history I extract how experienced operators adjust the setpoints and what they respond to. The fixed-price remote diagnosis tells you whether model-based setpoint advice could pass this on.