NIR process analyzer: specifying a model, not just an instrument

Building the measurement

1. Define the measurement and its required accuracy. What decision does the inline number support, and how accurate must it be for that decision? An analyser accurate enough to see a trend is far cheaper than one accurate enough to release material, and the two are frequently confused at specification.
2. Choose the probe position before the instrument. Path length, window fouling, bubbles and temperature at the measurement point determine whether any spectrum is usable. A probe in a poorly chosen position produces uninterpretable data regardless of the instrument behind it.
3. Plan the reference data set honestly. A calibration model needs samples spanning the full range of the variable, including conditions the process rarely reaches. Collecting them may require deliberately running the process off target, and that has to be planned and approved.
4. Decide how the model will be maintained. Raw materials, process changes and probe replacement all shift the spectral response. Agree who owns the model, how drift is detected and what triggers recalibration, before the system is handed over.
5. Qualify the sensor against a reference method. Every inline sensor needs an offline reference to be judged against, at installation and periodically afterwards. A sensor that has never been compared to a reference method is a trend line, not a measurement.

The model is the deliverable

A spectrum is not a measurement. The calibration model that maps spectra to concentrations is built from reference samples, validated against held out data and maintained as the process evolves, and it is the part of the project with the longest timeline and the least visible budget.

Ask a vendor who builds the model, what reference data they need from you, how it will be validated and who maintains it afterwards. A proposal that covers only hardware and installation is covering perhaps a third of the work.

Installing a probe changes the process

A probe is a penetration, a cleaning obligation and sometimes a sterility risk. In a regulated process it is a change requiring assessment, and in a disposable system it constrains which assemblies can be used at all.

Involve the people who own the equipment and the quality system before the probe is ordered. Retro fitting an analyser into a validated process is considerably harder than including it in a new one.

These instruments are calibrated, not absolute

An optical composition instrument does not measure protein or fat directly. It measures absorbance at many wavelengths and predicts the value from a model built on samples whose composition was measured by a reference method. The model is the instrument's real content.

So the questions are which reference method the calibration was built against, how many samples it covers, whether your material falls inside its range, and how the model is maintained. An instrument calibrated on a different variety, region or process will read confidently and wrongly.

Keeping a calibration alive

Models drift as material changes between seasons, suppliers and process conditions, so a programme needs periodic reference analysis of samples the instrument has read, and a procedure for updating or biasing the model. Without it, the instrument slowly stops agreeing with the laboratory.

Ask who owns the calibration, what it costs to update, and whether it can be transferred to a second instrument. Calibration transfer between instruments of the same model is not automatic and is a real question for anyone running more than one.

Condition monitoring answers a different question

Analysing a lubricant or a fuel in service is about trend rather than absolute value: wear metals rising, water ingress, viscosity changing, additives depleting. A single result means little and a series means a great deal.

That makes sampling discipline, consistent sampling points and a database of history more important than instrument precision. Programmes that change sampling point or laboratory lose the history that made the data valuable.

Gas analysis in a process stream

Continuous gas analysis for composition, calorific value or a specific component is specified by the components reported, the ranges, the response time and the sample conditioning that gets a clean, dry, correctly pressured sample to the instrument.

The conditioning system is where these installations succeed or fail, and it is routinely underspecified. Ask what conditioning is included, how it is maintained and what happens to the reading when it is not.

One catalogue heading, several process analyzers

A heading for food instruments mixes optical composition analysers, classical nitrogen and fat determinations, moisture balances, texture instruments and colour meters. They share a market rather than a principle, and most of the fast ones are optical instruments with a calibration for a commodity.

Which you need follows the specification the product is sold against: the reference method named there is the one the result has to be comparable with, and a faster instrument is acceptable only where its calibration against that method is maintained.

An nir analyzer, and the model behind every reading

An NIR analyzer does not measure a concentration; it measures absorbance across overtone bands and applies a calibration model that somebody built from reference samples analysed by a primary method. That model is the instrument's real content, and it is why two analyzers reading the same stream can disagree: different reference sets, different preprocessing, different validation range. Ask how many reference samples the model was built on, which primary method they were measured by, how the model is validated and updated, and what the analyzer does when a sample falls outside the model's space.

A near infrared analyzer, and what it is good at

Near infrared reads overtone and combination bands, which makes it fast, non-destructive and indifferent to sample form, and poor at trace analysis. What it is good at is a bulk property on a stream or a solid: moisture, fat, protein, a polymer's composition, a blend's uniformity, a tablet's content at line. Every one of those is a calibration built from reference samples measured by a primary method, so the analyser is bought with a modelling plan rather than as an instrument. Ask how a new product is added to the model and who maintains it.

Common questions

Why do inline analysers end up unused?

Almost always because the calibration model was never completed or maintained. The hardware works; the model that turns its output into a usable number requires reference data and ownership that the project did not budget.

Is Raman or near infrared the right choice?

Raman handles aqueous systems well and gives sharper molecular specificity, which suits identification and some concentration work. Near infrared is faster, cheaper and well suited to bulk composition and moisture. The decision follows the analyte and the matrix, not the technology.

Are single use sensors, such as an optical dissolved oxygen sensor, as good as traditional probes?

For dissolved oxygen and pressure they are now routinely used in disposable systems, with pre calibration replacing in place calibration. The trade is convenience and sterility against the inability to recalibrate in place, and the supply chain becomes part of the measurement.

What does a handheld instrument actually deliver?

Identification and verification of a material against a library, quickly and at the point of receipt. Quantitative work in a process stream is a different specification and usually a different instrument.

Does an optical composition instrument measure protein directly?

No. It predicts the value from a calibration model built against a reference method. Ask which reference method, how many samples, whether your material is inside the model's range and how the model is maintained.

Why is a single oil analysis result of limited use?

Because condition monitoring is about trend: wear metals rising, water ingress, viscosity changing. Consistent sampling points and a history matter more than instrument precision.

Are catalogue food analysers all the same technology?

No. The heading mixes optical composition instruments, classical nitrogen and fat methods, moisture balances and colour meters. The specification your product is sold against names the reference method the result has to match.

Browse by service class

• Compare Gene synthesis on advertised prices