Bill, Tom and Dog Visit the Fertiliser Factory

Bill, Tom and Dog were out for a drive one morning when they came across a large new factory on the edge of town.

A sign at the gate proudly announced:

Turning Waste into Sustainable Fertiliser

"Looks interesting," said Tom. "Let's have a look."

Inside, the plant manager welcomed them.

"We take waste material in here," he explained, pointing to a large receiving area. "We add the nutrients, dry it using heat, process it, pelletise it, and send the finished fertiliser out the other end."

He pointed proudly around the factory.

"Everything is balanced. The fuel comes in here, the heat dries the material, the gases go through the cleaning system, and the finished product comes out there."

Bill nodded.

"So you know how the plant behaves?"

"Of course," said the manager. "We designed it for these conditions. We know what goes in and what comes out."

Tom looked impressed.

"So if you put the same things in, you get the same result?"

"Exactly."

While the three of them were talking, Dog wandered away.

Dog was always interested in important technical investigations.

Especially ones involving piles of interesting smells.

A few minutes later Bill noticed him standing beside the incoming material.

"Dog," said Bill suspiciously, "what have you been doing?"

Dog looked away.

The plant manager frowned.

"Don't tell me..."

Bill looked down.

Dog had added a little extra moisture to the feed.

The old-fashioned way.

The manager laughed.

"That's nothing. A little bit of extra water won't matter. The plant controls will handle it."

And at first, he was right.

Nothing dramatic happened.

The plant kept running.

The conveyors moved.

The dryer worked.

The fuel system adjusted.

The operators watched the gauges.

"See?" said the manager. "The system is stable."

But then something changed.

The wetter material needed more heat to dry.

The dryer demanded more energy.

The fuel system increased its output.

The temperature in the process changed slightly.

That changed how the material reacted.

The product leaving the system was a little different.

The operators adjusted the settings.

The plant responded.

But now it was not quite where it started.

The next day, Bill and Tom met the manager at the Pub. He had spent the night looking over the operating data.

"It's odd," he said.

"The plant is still operating, but it has moved."

Tom looked confused.

"What do you mean, moved? It is still making fertiliser."

"Yes," said the manager. "But not exactly the same way. It's running differently".

Bill pulled out his notebook.

"That is the interesting part."

He drew a simple diagram.

"Most people imagine a factory like this."

He drew:

Inputs → Plant → Outputs

"You put material in. The factory does something. Something comes out."

"That is how we usually think about machines."

Then he added another arrow.

"The problem is that this factory is not just a machine. It is a connected system."

He drew:

(now) Inputs → Plant → Outputs → (next minute) Inputs → Plant → Outputs → (repeat)
      
"The output affects what happens next. What happens next changes the plant behaviour. The plant changes the output again. It's a loop"

Tom looked at the drawing.

"So the factory talks to itself?"

"Exactly," said Bill. "The history of the plant changes the way it performs in the future"

"And small changes can move the system into a different operating state."

He drew a line.

"At first, the plant was here."

He marked:

State A — normal operation

"Everything was balanced."

Then he drew a second point.

"But after the change, it moved here."

State B — another stable operation

"The plant did not break. Nothing failed. It simply settled into a different way of operating."

The manager looked at the drawing.

"So it can have more than one normal condition?"

"Yes," said Bill.

"And that is the question we need to ask about complex plants."

Tom nodded.

"Not just: 'Does it work when everything is exactly as expected?'"

"Exactly," said Bill.

"The bigger question is: 'What happens when real life changes the inputs?'"

"What if the material is wetter?"

"What if the fuel quality changes?"

"What if the feed composition changes?"

"What if operators adjust the system?"

"Does it return to the original condition?"

"Or does it settle somewhere else?"

The manager thought about the factory.

The gauges were still moving.

The machines were still running.

The exhaust stack was still producing a steady stream of gases, but they were different gases compared to yesterday's.

But Bill pointed at the diagram.

"The important thing is that a plant does not only have outputs. It has behaviour over time - a memory"

"And that behaviour determines what comes out."

Dog wagged his tail.

He had no idea he had just demonstrated a principle of engineering systems.

He only knew that sometimes a little extra moisture changes things.

What Tom learned — A plant can have more than one ‘normal’

Dog's contribution to the fertiliser factory was funny, but the idea behind it is important.

A factory like this is not simply a row of machines where one thing happens after another. The parts interact.

The moisture in the incoming material affects how much drying is needed. Drying affects how much heat is required. Heat demand affects the thermal process producing that heat. What happens there affects the heat available for the dryer. Material that is not quite right may be recycled through part of the plant, changing how much material is already in the system.

And all of those things are changing over time.

That makes the plant a coupled dynamic system.

What does that mean?

Coupled simply means the different parts affect each other.

Dynamic means what happens now affects what happens next.

Think about Dog adding some water.

The important question isn't just:

How much extra water did Dog add?

It is:

What does the whole plant do afterwards?

The dryer may demand more heat. The heat-producing process responds. Material moves through the plant. Controls adjust. Some material may circulate back through the system.

Fifteen minutes later, the plant is in a slightly different condition.

That condition becomes the starting point for the next fifteen minutes.

And so it continues.

The plant therefore has a memory of what has happened before. Two apparently identical plants receiving exactly the same material today can behave differently if they arrived there by different paths.

So I built a simple experiment

I obviously don't have enough engineering information for me to build a detailed computer model of the proposed Glan Devon plant.

But I don't need one to test the basic idea.

I built a deliberately simple model containing three things that have to interact:

• The drying duty: how much water still has to be removed from the biosolids;

• The thermal state: the amount of useful heat available from the thermal process; and

• The downstream inventory: material moving through or waiting in blending, granulation, screening and recycling.

I connected them with simple rules representing things real plants contain: feedback, delays, recycling, storage limits, equipment capacity and trips.

Then I let the model move forward in 15-minute steps for 600 hours.

The purpose was not to predict exactly what the Glan Devon plant will do.

It was to test a much simpler question:

Does a plant with this sort of connected structure necessarily settle into one steady operating condition?

That's an important question. If the plant can have more than one steady operating state, the emissions assumptions used in the air-quality model may describe only one of the ways the plant can actually operate.

First I checked whether each case could work at all

I generated 40,000 different combinations of model settings across a deliberately broad range.

Most failed basic water, heat or production-rate checks before the dynamic experiment even started. That's not surprising. I wasn't trying to make every combination represent a realistic plant — I was exploring the boundaries of the model.

5,027 combinations did pass my checks.

That matters. These were not cases that were obviously impossible because there wasn't enough heat or because material could never move through the plant fast enough.

On simple average balances, they could work.

I then ran every one of those cases twice.

The external conditions and model settings were identical.

The only difference was the condition of the plant when I started it — how much drying work remained, how much recoverable heat was already in the system and how much material was already circulating downstream. A bit like Dog adding extra moisture in the yarn: change the drying duty and you change the heat demand, which then changes what happens next.

That little difference meant:

Same plant. Same inputs. Different history.

And the history mattered

In about 26% of the cases, changing only that starting condition changed the eventual behaviour of the system.

That is the important result. In over a quarter of the cases, exactly the same model under exactly the same external conditions ended up behaving differently simply because it started somewhere different.

There wasn't always one inevitable operating state waiting for the plant.

Where it ended up sometimes depended upon where it had been.

Only 15% of the cases that passed the basic balances settled into a fixed operating point, or smoothly approached one, from both starting conditions. The rest did something different.

A large number eventually reached a storage or operating limit and tripped. Everything stopped before my 600 hour test was finished.

Others settled into different operating regimes depending upon their starting condition — different heat demand, different thermal state and different material inventories. Those different operating conditions could also produce different emissions.

Some cycled repeatedly, up/down, up/down, up/down.

Others continued moving around without settling to a single fixed point during the experiment — the modelled plant never reached a steady state during the 600 hours.

Those percentages are not predictions of how often the real plant will behave in those ways. My simple model cannot tell us that.

What the experiment demonstrated is more fundamental:

A steady operating state is not an automatic consequence of having a plant whose average water, heat and material balances work.

Why does that matter for emissions?

This is where Tom's factory visit becomes important.

Suppose the plant starts in State A.

The dryer has a particular heat demand. The thermal process operates at a particular condition. Material moves through the plant at particular rates.

The exhaust gases produced under that state have some emissions profile.

Then something changes — perhaps wetter biosolids, different fuel, different feed composition or simply a different operating history.

The plant responds and eventually settles into State B.

State B might also be perfectly stable.

Nothing has broken.

The plant may still be making fertiliser.

But the heat demand, thermal conditions, material flows and gas production may now be different.

And therefore its emissions may be different.

The experiment did not calculate pollutant concentrations or emission rates. It wasn't designed to.

What it showed was why that question needs to be asked.

If a plant can occupy more than one credible operating state, then demonstrating emissions for one assumed operating condition does not automatically demonstrate emissions for all of them.

That leads to the question Tom would now ask:

What operating states can this plant actually occupy, and have the environmental impacts been tested across them?

That is the point of the coupled dynamic systems experiment.

Not that the plant must become unstable.

Not that it will behave chaotically.

And not that my simple model predicts what the real plant will do.

It showed something much simpler:

A connected plant can have more than one “normal”.

And before relying on a particular set of predicted emissions, we need to know which normals are possible.

© 2026. All rights reserved.