Half a point of moisture too much is energy lost in the dryer; half a point too little is yield lost at dispatch. The decision is taken continuously, and the laboratory value arrives afterwards.
The laboratory remains the reference: oven drying, extraction and the protein assay give the accurate value, and the model is built on them. The in-line measurement returns it while the dryer is still running.
The matrix is natural and variable: a flour changes with the harvest, a milk with the season. A ready-to-use model must therefore have been built over several seasons. That is our partners’ trade: their agricultural and food databases span several years of collection and are updated regularly. The model is then verified on your production.
On a shared line, product changeover is timed on the transition measured in the pipe: downgraded product, water and line time all come down.
An installed analyser you no longer trust? Often its values have never been compared with your laboratory, or not since the last season. Before buying anything new, the diagnosis: the hardware is often sound. What to do with an instrument that has fallen out of use.
Where to start, depending on your role
- You run the plant: start from the operation where a point of moisture costs most. Scoping.
- You develop the method: what a proof of concept has to settle. Feasibility.
- You own quality and food safety: ISO 12099, food safety plan, labelling. The texts.
- You decide the investment: a point of moisture reads as points of yield, calculated on your history. Cost of a control.
- You want examples: eight real situations, presented without names. Experience.
Before the factory: grain, forage, material on receipt
Agricultural and food measurement share the same physics and the same working model. What changes is when the decision is taken and where the instrument sits.
| Where | What is measured | The decision it allows |
|---|---|---|
| At the trailer, on receipt | Moisture and protein on cereals and oilseeds | Accept, reject, send for drying, apply the payment scale |
| At the silo | Moisture of the stored grain | Decide on aeration while the grain is still in condition |
| In the field and in the barn | Dry matter and feed value of forage and silage | Choose a harvest date, adjust a ration |
| On the line | The same variables, continuously | Steer a drying, a blend, a despatch |
The season is a calibration parameter
An agricultural matrix changes with the variety, the origin, the year and the growing practice. A model built on one harvest describes that harvest, and it is widened over two or three seasons. That spend belongs to the project plan from the start, alongside the instrument.
The same instrument exists in field, laboratory and line versions, which lets you keep a single measurement reference from the field through to despatch. Our manufacturer partners.
The three variables that carry the value
Moisture and dry matter
The leading variable of the sector, and the most profitable. It governs yield, drying time and storage stability. It is also the one whose reference method has to be frozen first: two oven protocols do not measure the same water. What a moisture measurement actually covers.
Blend homogeneity
Incorporating an additive, blending powders, seeding: the question is not the mean content but its spread. An in-line measurement follows that spread as it falls and gives an objective stop criterion. Measuring content uniformity.
Checking on receipt
Verifying identity and conformity before unloading changes the nature of the job: the decision moves to the weighbridge, ahead of the transfer. Identifying a raw material.
Fat and protein come out of the same acquisition as dry matter, and each constituent brings its own reference method, its own sample campaign and its own model. One sensor, three variables, three calibrations: a figure worth having in the budget from the start.
Where the measurement sits, operation by operation
| Unit operation | What is measured | What it changes |
|---|---|---|
| Receipt of agricultural materials | Identity, moisture, content of a key constituent | Refusal is decided before unloading. Incoming variability is documented batch by batch instead of being absorbed. |
| Blending | Spread of the content of a major constituent | Blend time stops being a fixed duration: you stop on a criterion reached, and the blends that need longer show it as they run. |
| Drying | Residual moisture during the cycle | Following the trajectory lets you stop at the right moment rather than at the end of the planned time, and the margin you no longer pay for shows on every batch. |
| Cooking or heating | Dry matter, change in composition, the tipping point | An instrumental criterion replaces an operator’s judgement: the result stops depending on which shift is on. |
| Separation | Composition of each outlet, carry-over of valuable material | Valuable material carried into a secondary stream shows continuously, rather than in a mass balance afterwards. |
| Flow in a vessel or a pipe | Phase distribution, settled bed, product changeover front | A blockage is seen coming, and a changeover is set on the real transition rather than on a timer. |
| Packing | Composition and moisture of the finished product | Conformity is checked on the stream rather than by pallet. A deviation is attributed to a moment in the cycle, not to a whole batch. |
| Finished product control | Presence, absence and integrity of the contents through non-metallic packaging; colour in the visible; counting and morphology on the line | The check is on the packed unit, without opening or destroying it. Metallic foreign bodies remain a job for X-rays. |
A spectroscopic measurement analyses the few milligrams it illuminates. Repeating the acquisition in the same place reduces noise, and measuring several portions of the stream is what makes the result speak for the batch. That is a question of representativeness before it is one of sensitivity.
The matrix is natural and variable, which makes this the ground where calibration on samples earns its place. A pure-component model asks for a composition that is known, finite and declarable, which suits a defined bioproduction medium. A dairy product or a flour is served by regression against reference values. What each approach costs.
That brings a coverage requirement particular to the sector: the training population has to contain the seasons, the origins and the suppliers. What counts is a spread of variability, not a number of samples. Thirty samples spread over the year beat two hundred taken in the same week. How many samples you really need.
Which follows: a food project is planned over a raw material cycle, and that is what makes it hold.
Two in-line configurations, and what decides between them
Measurement can be made contact-free, above a belt, a chute or a transfer section: nothing touches the product, and the line is not modified. Installation configurations.
Line instruments are not told apart by their ruggedness first. Two questions decide, and you already hold both answers.
Is the variable in the visible?
Colour, the Agtron value and the degree of cooking are measured below 780 nm. An instrument whose range begins in the near infrared does not see them, whatever its quality elsewhere. Moisture, fat and protein sit in the near infrared and do not call for the visible. A crisp or chip line needs the visible; a dairy or rendering line does not. This is the case of the ZEISS Corona process, which covers the visible and the near infrared: colour and composition on the same acquisition.
Is the measurement point in a classified area?
A starch drying tower, a pneumatic powder conveyor. Not every configuration is declared for zones 21 and 22. The question comes before any discussion of performance, because it opens or closes the list of candidates.
Near an oven or in flour dust, how the sensor holds up is planned for. Depending on the instrument: a cooled housing, an air purge that keeps the window clean, a suitable ingress protection rating, and a working distance that keeps the instrument away from the heat source. The choice is made at scoping, on your real conditions.
The rest — ingress rating, shock resistance, working-distance sensor, lamp redundancy — is settled afterwards, and settled well.
A sensor above an open line: what food safety asks for
The question often comes late, and it can stop an otherwise successful installation: what happens if the window breaks above the line? It is dealt with at scoping, with your food safety manager.
- The window. Usually sapphire glass, a very hard material that resists scratches and thermal shock.
- The protection. It depends on the instrument and the application. It is chosen at scoping, not at delivery.
- The position. The instrument can be placed and oriented so that a fragment cannot fall into the product.
- The check. The integrity of the window is checked periodically, and that check can be built into your glass and brittle materials register.
- CIP. A non-contact measurement stays out of the CIP circuit. A probe in contact is chosen to withstand CIP, and this is checked at scoping, on your products and your temperatures.
The phases of a CIP cycle read on the same measurement. Product, push water, caustic, acid and rinse water differ in conductivity. Electrical tomography follows the front of each phase across the whole pipe section, alongside the return conductivity meter. The water push stops on the real interface, and the end of rinsing becomes a measured state.
The instruments comply with the applicable CE standards, and probes in contact with the product comply with Regulation (EC) No 1935/2004 on materials in contact with food.
What the instrument brings, and what is built at your site
NIR instruments for this sector arrive stable, and stability is exactly what you want from the hardware. The model is built separately, in chemometrics software, on your matrix. Manufacturer documentation describes instrumental stability rather than a performance figure by constituent, and that distinction is the useful one to hold at purchase.
An application page headed dairy or cereals describes an aptitude of the instrument. The hardware arrives ready, and the model is built one matrix and one variable at a time, at your site, with your reference analyses. Budgeting both lines from the start is what makes the project land where it was costed.
The two questions that decide the project
What a listed application covers, and what goes with it
A budget built on the sensor alone leaves out the sample campaign, the reference analyses, the line time and the upkeep of the model. All four are known figures, and they belong in the same budget line. The question that brings them out fits in one sentence: on which matrix was this model built, against which reference, and who maintains it as the material moves?
Building the calibration across the seasons
A model built in January on January production works perfectly in February and well in March, and September brings a different material. Seasonal variability is spectral as much as compositional. Two answers, both chosen at the design stage. Spread the training campaign over a full cycle, or run a provisional model with its detection of new territory instrumented, Hotelling’s T², Q residuals, Mahalanobis distance. Either way the first change of harvest is a planned event.
Calibrations that already exist for your materials
On receipt and in the laboratory, you are not obliged to build a model. We work with the calibration sets of two partners on agricultural and food materials: the CRA-W in Gembloux and ZEISS.
Bought or rented, with transfer onto your spectrometer, a ring test that establishes the gap on your own batches, and updating under annual contract — because an agricultural material changes from one season to the next, and a model that does not follow that drift degrades without warning.
With ZEISS the same approach exists in-line on the process, not only in the laboratory. What a turnkey calibration covers.
On forages and animal feed, the coverage reads by family. Twenty-eight materials, in eight families.
| Family | Materials | Variables covered |
|---|---|---|
| Green forage and fresh material | Chopped forage, whole-plant maize, fresh grass, immature cereals | DM, ash, protein, fat, fibre, NDF, ADF, starch, biogas potential |
| Silages | Grass, lucerne, maize, high-moisture maize | DM, ash, protein, fat, NDF, ADF, starch, biogas potential |
| Hays | Lucerne, grass, mixed — ground | DM, ash, protein, fat, fibre, NDF, ADF |
| Total mixed rations | Lactation, dry, beef | DM, ash, protein, fat, NDF, ADF, starch |
| Feeds | Ground maize, compound feed | DM, ash, protein, fat, fibre, starch |
| Cakes and co-products | Soya, sunflower, rapeseed, wheat bran | DM, ash, protein, fat, fibre, and starch on bran |
| Grains, whole kernel | Wheat, barley, maize | DM, ash, protein, fat, fibre, starch — coverage varies from one cereal to another |
| Slurries and digestates | Digestate, cattle slurry, pig slurry | Dry matter and nitrogen |
Grain is measured whole. Wheat, barley and maize are covered without grinding: no preparation, so no delay and no operator.
Biogas potential is covered on eight materials, from the silages to immature cereals. That is not the livestock customer, and it is the same instrument.
Preparation decides as much as the model. “Ground” supposes a forage-sampler take, and pelleted feeds have to be broken. And this table states a coverage, never a performance: no accuracy is announced here, it is established on your own batches.
The texts that frame a measurement in food and feed
Here a standard bears directly on applying near infrared to your materials — rare enough to be worth noting.
- ISO 12099 — Animal feeding stuffs, cereals and milled cereal products: guidelines for the application of near infrared spectrometry.
- Regulation (EC) No 152/2009 — the methods of sampling and analysis for the official control of feed.
- Regulation (EC) No 852/2004 — the hygiene of foodstuffs, and the food safety management plan that follows from it.
- Regulation (EU) No 1169/2011 — food information to consumers. An in-line measurement of moisture, fat or protein then becomes a labelling compliance tool, not only a process control one.
Tell us the product, the operation and the reference method. We will tell you whether the measurement holds.
Forty-five minutes is enough to know which variables are reachable on your matrix, what sample coverage your variability demands, and where the gain is calculated: yield, drying energy or laboratory effort.