NPK fertilizer control with AI — an off-grade bag is not an incident, it is a fine and a lost customer.

The problem

The bag does not lie — but the lab arrives four hours late.

The manager of an NPK fertilizer plant lives with three realities that are rarely in sync:

1. N-P-K dosing changes with every raw material — urea, ammonium nitrate, monoammonium phosphate (MAP) and potassium chloride (KCl) each arrive with their own supplier analysis, and the real nutrient percentage of the batch does not always match the nominal recipe. The LIMS knows it; the granulator operator, almost never in time.
2. The granulator and the dryer do not always deliver the same thing — drum speed, binder, ambient humidity and knife wear all change the particle size distribution. A percentage of fines above expectations and the bag segregates in transport: the last customer receives a different fertilizer from the first.
3. The regulation does not accept a letter of explanation — EU Regulation 2019/1009 sets tight tolerances for the grade declared on the bag. An inspection re-grading or an importer analysis below tolerance means recall, fine and a customer who buys elsewhere next year.

The lab analyzes by sampling and returns the result hours later. By then the batch is bagged, palletized and, with luck, still in the warehouse. The problem is not the recipe, it is that the classic system always arrives late — and every inspection re-grading or every importer return shows up across the whole financial year.

How it fits the IRIS system

iLEAN does not change your granulator — it puts the missing intelligence on the data you are already generating.

The problem in NPK fertilizer control is not a lack of information: it is information living on islands — the LIMS knows how the raw material came in, the SCADA knows the drum speed and dryer temperature, the ERP knows which importer the pallet is going to and with what tolerance, but none of them talks to the next one at bagging speed. iLEAN acts as the putty that fills those gaps, without asking you to change your SCADA, your LIMS or the plant.

Edge sees particle size and moisture in line. Connect reads the LIMS analysis and the importer's spec. The agent cross-checks against the EU tolerance and, if something does not add up, holds the batch before bagging. The person signs — never the other way round.

The three iLEAN pieces applied to NPK fertilizer control:

• Edge — a terminal with machine vision (CNN) over the granulator outfeed conveyor (classifying 1-2-3-4 mm particle size in real time), an NIR moisture sensor after the dryer, and capture of the dosing from every weigh hopper. If the particle size distribution drifts, it triggers a status light and notifies the operator. It works with no network.
• Connect — captures the raw material analysis whether it comes from the LIMS, from the lab spreadsheet or from the supplier's PDF. And it also captures what arrives from outside (the EU Regulation 2019/1009 in force, the African or European importer's spec, the carrier's email with storage conditions) at second zero, with nobody forwarding anything.
• Agent — cross-checks the real N-P-K dosing, the measured particle size, the moisture and the customer spec against the EU tolerance. If there is a deviation, it holds the batch and alerts the quality manager through whatever channel they use. It also generates the conformity dossier — which raw material came in, which grade came out, which particle size — ready for inspection or for the importer. The person validates and signs; the pallet does not ship on its own.

See the full IRIS architecture →

Before and after

Manual control vs. cross-checked control with iLEAN

Aspect	Manual control + lab by sampling	With iLEAN Edge + Connect + Agent
Detecting off-grade N-P-K	When the lab returns the result (hours)	Continuously during granulation
Particle size distribution	Manual sieve once a shift	Vision over the conveyor, batch by batch
Moisture after the dryer	One-off sample	In-line NIR, continuous
Cross-check against EU 2019/1009 tolerance	Documentary review after the batch	Real-time validation, hold if out of range
Traceability by pallet/importer	Rebuilt by hand	Per-batch dossier, automatic
Operation with no network	n/a	Edge keeps running on the panel's own light

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Mid-sized granulated NPK plant (blending urea/nitrate, MAP/DAP, KCl/SOP), 1-2 granulation and drying trains, bagging in 25-50 kg bags and big bags.
• Edge pilot on one train (vision over the granulator outfeed conveyor + NIR moisture + integration with hopper dosing and the LIMS). First value expected in a few weeks.
• Indicative payback between 4 and 9 months, depending on the frequency of inspection re-gradings, rework due to segregation and importer returns documented in recent years.
• The hard lever is a single re-grading avoided at an importer: product held, transport, re-analysis, fine for an out-of-tolerance declaration. The expected reduction in in-line rejects is ≥ 30% over the current baseline.

And the technical director's reasonable doubt

“What if the AI gets the vision-based particle size wrong and holds a batch that was fine?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely classifies particle size against a distribution trained on your own batches, the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN holds and the person signs. The three safety rings exist precisely for this, and the agent never touches the functional safety electronics around the nitrate.

^[1] OpenAI paper “Why Language Models Hallucinate”, 2025 — on the reliability of AI in anchored tasks.