In chemicals the constraint is economic and energetic: every safety margin is paid for on every tonne.
A reaction held twenty extra minutes “to be safe” spends steam, cooling and reactor time, because the value that settles it arrives from the laboratory an hour too late.
Titration, NMR and chromatography give the accurate value and build the model; the in-line measurement returns it at the pace of the reaction, even in an opaque medium or an ATEX zone.
An analyser still running, and no one looking at it? An online analyser whose model has drifted, and whose alarm is no longer read in the control room. Before buying anything new, the diagnosis: often the hardware is sound, and it is everything else that is missing. What to do with an instrument that has fallen out of use.
The three questions that come up most often
- “When can I stop?” End of reaction, distillation cut. Detecting the endpoint.
- “What size am I at?” Crystallisation, milling, dispersion. Particle size.
- “Is this the material on the label?” Goods-in, sorting, charging. Material identity.
Where to start, depending on your role
- You run the plant: start from the operation that runs on “to be safe”. Scoping.
- You develop the method: what a proof of concept has to settle. Feasibility.
- You own quality and safety: ATEX, ASTM standards, what is mandatory. The texts.
- You decide the investment: the cost of a control, calculated on your batches. Cost of a control.
- You want examples: eight real situations, presented without names. Experience.
The three questions, in detail
When can I stop?
End of reaction, exhaustion of a reagent, a distillation cut, end of emulsification. The highest economic stake in the sector, and often the quickest to handle: following a signal to stabilisation frequently works with no absolute value to predict. Detecting the endpoint.
What size am I at?
Crystallisation, wet milling, dispersion. The size distribution governs filterability, downstream drying and the product specification. Some optical measurements return it on concentrated suspensions with no dilution, so the state you are observing stays intact. Measuring particle size.
Is this really the material on the label?
Goods-in checks, polymer sorting, verification before charging a reactor. The check takes seconds at the drum and it protects the batch and the cleaning cycle behind it. Checking raw material identity.
Operation by operation, what is measured and what it changes
| Unit operation | What is measured in-line | What it changes |
|---|---|---|
| Batch reaction | Conversion, disappearance of the limiting reagent, appearance of the product | The end of reaction is observed instead of waited out. The signature of the operation becomes visible, and an abnormal signature shows during the run rather than at final testing. |
| Distillation and separation | Composition of the phase being followed, exhaustion of a constituent in real time | The cut point is made objective rather than judged by eye. Fewer reworks, less reboiler energy. In-line viscosity serves as the criterion where composition is not enough. |
| Polymerisation and functionalisation | Degree of substitution, residual monomer, functional groups | The stop is placed on the reaction itself. Transmission MIR measures viscous media up to around 10,000 cP with no dilution, at the pace this operation is decided. |
| Emulsification | Surfactant incorporation, stabilisation of the dispersed phase | Emulsification is a delayed process rather than an instantaneous one: the oil-phase bands only appear after a measurable lag. The end criterion becomes spectral, driven by the chemistry rather than by a recipe time. |
| Crystallisation | Size distribution, appearance and growth of crystals, supersaturation | Seeding and the cooling profile are set on a measurement rather than on history. Downstream filtration stops being the adjustment variable. |
| Wet milling and dispersion | How the size distribution moves during milling | Milling stops when the target is reached, which keeps both the energy and the product properties where you want them. |
| Drying | Water content and residual solvent during the cycle | The stop follows the state reached rather than a duration, and the safety margin stops being paid in energy on every batch. |
| Flow in a vessel or a pipe | Phase distribution, stratification, settling, product front | Electrical tomography shows what happens inside a closed stainless pipe or vessel with no multivariate model to maintain: the distribution comes out of the measurement physics. |
| Identity checking and sorting | Polymer type, presence of fillers and additives | Sorting happens at source. Raman identifies most polymers in seconds, fillers and additives included, black plastics among them — which few instruments manage. Carbon black absorbs the excitation: the answer is not more power, which would burn the surface, but recovering a weak signal. That is a matter of the collection train, not of the laser. |
None of these measurements analyses the batch. Each analyses the portion of material the sensor illuminates, where it sits. That is a question of sampling before it is a question of instrument.
The safety margin is the real cost line. In a chemical plant, uncertainty rarely turns into a rejected batch. It turns into extra cycle time, solvent and steam, applied to all production to cover the unfavourable cases. The gain from an in-line measurement is calculated on one gap: the spread of actual durations and consumptions between the best batch and the worst. Not on the cost of a laboratory analysis: that reversal is particular to this sector. Elsewhere it is the number of controls a year that decides.
That calculation is done on your own history, during scoping, before any purchase. A process already centred and already repeatable will show a small figure, and that too is a result: it tells you the margin is already where you want it.
The case of pharmaceutical synthesis
Part of fine chemicals makes pharmaceutical active substances. The operations are the ones described here — reaction, crystallisation, filtration, drying — but the frame changes: the process is justified under ICH Q11 and run under ICH Q7, and polymorphism becomes a critical attribute alongside yield.
That shifts where the measurement pays: it still serves throughput and energy, and it now also serves the dossier. We cover that case on its own page: drug substance.
Two questions that decide the project
NIR reaches a particle size by correlation, and that has consequences
Particle size changes how light scatters, and therefore the baseline of the spectrum. You can draw a prediction from that, and it can be excellent, but it is indirect. There is no pure spectrum of a particle size: you need a model calibrated on samples, built on reference values. That model holds while the raw material, the filler and the compaction stay inside its range, and it is refreshed when they move. Where materials move often, a measurement whose quantity comes out of the physics, light scattering or tomography, costs less to keep alive.
Mechanical access sets the timetable, and it is settled with projects and maintenance
A process connection, a flange, a probe passing through a jacket, a pipe that has to be opened. These belong to projects and maintenance, not to the analytical department. They set the timetable far more than the chemometrics do. Area classification is settled at the same moment. It belongs to the operator: the employer classifies hazardous places into zones (directive 1999/92/EC; in France, article R4227-50 of the Labour Code), and the equipment is then chosen for the zone set. Explosive atmosphere certification is an early selection criterion, since it exists on some configurations and is worth confirming on the others. Asked in the first meeting, it steers the choice of instrument from the start. We ask it there.
The rest is handled by design: fouling of a window in a loaded medium, thermal drift, and the calibration bounds within which the model is valid. That last point belongs in the feasibility report, written down. A stated range of validity is worth more than a stated accuracy. That is the subject of keeping a model alive, and what turns a good demonstration into an installation that holds.
The texts that frame a measurement in chemicals
No regulatory text plays the part of ICH Q2(R2) here: nobody will impose what a process analytical procedure must demonstrate, and nobody will come to check it. What regulation frames here is the product and the installation. Industry standards do exist, and they say exactly that — they are simply not enforceable: holding to them is your decision.
- Regulation (EC) No 1907/2006 (REACH) and Regulation (EC) No 1272/2008 (CLP) — registration of substances, classification and labelling.
- Directive 2014/34/EU (ATEX) — it decides which equipment is admissible at the measurement point, and often the cost of integration.
- ASTM D6122, Validation of the Performance of Multivariate Online, At-Line, Field and Laboratory Infrared Spectrophotometer, and Raman Spectrometer Based Analyzer Systems — edition in force: D6122-25. It is the closest thing to a validation frame for a process analyser: periodic performance checks, comparison with the primary method, drift criteria.
- ASTM E1655, Standard Practices for Infrared Multivariate Quantitative Analysis — edition 17, reapproved in 2024. Building the calibration itself: sample selection, validation set, out-of-domain sample detection.
Tell us which operation would gain most in cycle time. We will tell you whether it can be measured.
Forty-five minutes is enough to place your process: which measurement physics suits your medium, whether the quantity you are after is measured directly or by correlation, and on which line of your operating account the gain would show. Area classification is handled in the same conversation.