“Has my blend reached its consistency? Should I carry the dispersion on? Where has the molar mass of my resin got to?” Viscosity decides how an operation is run, and these three questions arise while it is running.
The laboratory measurement, on a sample, gives the value that counts — and it is the value any in-line measurement will be referenced against. What is at stake in addition is the moment: the information that would have settled it was there during the operation, and it is readable.
What a spectrum sees of a viscosity, and what has to be added
Viscosity is a flow property. It is measured by making the material flow, vibrate or rotate, and no spectroscopic method does that. The question to ask of any proposal for in-line viscosity measurement is therefore: on which quantity is the model actually built?
A spectrum, on the other hand, sees very well what governs the viscosity. On a given process, viscosity is determined by a small number of chemical and physical quantities, and those quantities are in the signal.
What governs a viscosity, and what a spectrum sees of it
| What governs the viscosity | Where it happens | What NIR or MIR sees of it |
|---|---|---|
| Molar mass, progress of a polymerisation | Polymerisation, polycondensation, crosslinking | The disappearance of the reactive functions and the appearance of the bonds formed |
| Conversion rate, residual monomer | Synthesis, functionalisation | The bands of the reactants and of the products, quantitatively |
| Solvent or water content | Dilution, evaporation, drying of a varnish | The historical strong point of NIR |
| Solids content, solid fraction | Suspensions, dispersions, slurries | The proportion between continuous phase and dispersed phase |
| Size distribution of particles or droplets | Wet milling, emulsification, crystallisation | Indirectly through scattering, and more directly through a dedicated size measurement |
| Temperature | Everywhere | A shift of the bands, those of water first. It is an effect to correct rather than a quantity to predict: it is measured separately and compensated |
That relation is what gets calibrated, rather than the viscosity itself. The model learns to link your spectrum to your viscosity value, on your process, and it carries no universal law.
What this route commits you to, and the question that comes first
This route is a full calibration on samples. It is the most demanding regime, and there is no shortcut.
- A reference sample campaign, covering the real variability: it is the main work item, and it conditions everything else.
- Your current viscometer becomes the reference method. Its dispersion bounds what the model can do: a prediction cannot be shown to do better than the reference that built it: its measured error includes the viscometer’s.
- A model that lives: it drifts if an unmodelled factor comes into play, a change of supplier for instance. It is monitored and requalified.
The first question to ask is therefore not technical, it is economic. What is it worth to you to know the viscosity continuously rather than every hour?
If the answer is “a lot”, the calibration campaign is justified. If it is “a little”, there is a shorter route, and it is often the right one: not predicting the value, but following the moment it changes regime. A stabilisation, a break in slope, a return to normal are detected without any absolute calibration. That is the logic of the endpoint, and it costs a fraction of the other.
The trajectory says more than the final value
This is the main argument for in-line measurement, and it holds whichever technology is used.
One example among others, on a carbonation. The viscosity rises during the operation, passes through a marked maximum towards the end of the gas introduction, collapses, then rises again. The maximum is no accident: its height follows the introduction flow rate, and a slower rate brings it down markedly. The value measured on the finished product carries none of that.
The same logic holds on a pigment dispersion, on a polymerisation, on an emulsion setting as it cools. That is the process signature, and it reads on the shape of the curve before it reads on a figure.
What we take into account in your project
A measurement project starts with four things you already know rather than with the choice of a sensor.
What you already have
Which viscometer you use today, at which shear rate, with what dispersion between operators. What instrumentation is already on the equipment and what it records. A free nozzle, a window or a probe already fitted changes the cost of the project entirely.
Your objectives
Steering an operation and releasing a batch call for different measurements. The first works from a reproducible signal. The second calls for a value tied to your specification, and so for analytical validation work. Telling the two apart is what keeps a project the right size.
Your constraints
Area classification, temperature and pressure at the measurement point, cleaning in place, window fouling, pace of decision, regulatory and IT requirements. Each one narrows the available catalogue, and they are better set out at the start than at the order.
Your practical frame
The line time available for trials, the number of batches you can sample, the budget and the calendar. Those are what decide between a quantitative model and a state criterion, rather than a technical preference.
Conditions for success and limits
The correlation is your process’s, not a law
The model links a spectrum to a viscosity because, on your product, the same causes govern both. If a new factor comes in, a changed additive or a raw material from another supplier, the relation can break without anything signalling it.
The remedy is known and it is put in place from the start: monitoring the spectral residual at each acquisition, and keeping a comparison point in the laboratory. A monitored model warns when it leaves its domain.
Shear thinning makes the target move
A Newtonian fluid has one viscosity. A shear-thinning fluid has one per shear rate. The model predicts the one your reference method measures, and nothing else.
That is not an obstacle when the specification is clear: it is enough to calibrate against the measurement that is authoritative at your site. It becomes one when two departments use two different instruments. Two instruments, two results: which one to believe?
A probe looks where it is
In a stirred vessel, the material is not in the same state at the bottom, at the wall and in the shear zone. An immersed probe faithfully returns the state of the zone it occupies, and nothing else.
The remedy is a better chosen location, not a more accurate instrument. And when the distribution is itself the question, the answer becomes an image of the phase distribution.
Heavily loaded media and pastes
Beyond a certain solid fraction, the spectrum is dominated by scattering and the relation with viscosity becomes unstable. Opaque, highly viscous media are the difficult case.
Two routes remain open. Following the dry matter or the particle size, which govern consistency and are measured better than it. Or changing quantity: the power drawn by the agitator is a poor indicator, but a free one, and it is sometimes enough to locate an endpoint.
What it changes, in practice
The gain is not calculated on the cost of the analysis. It is calculated on the duration of operations run by the clock, and on the batches reworked for want of knowing in time.
Three effects come back.
- The stopping point is decided on the real state of the material, not on a duration.
- A deviation is seen during the operation, not at final testing.
- Batches can be compared, because you have a curve and not a point.
The calculation is made on your own data, before any investment: real durations, rework rate, hourly cost of the equipment. It is the first deliverable of a scoping phase, and it sometimes concludes that the investment is not justified: that is a result too.
Frequently asked questions
Which in-line viscometer do you recommend?
None, and that is not an evasion. We do not deploy rheology instruments: our trade is non-destructive measurement by optical and electrical means.
What we do is different. We link your viscosity to the quantities that govern it, and it is those that NIR or MIR follow in line. Your viscometer stays useful: it becomes the reference method of the model.
Is a chemometric model needed?
Yes, if you want a viscosity value. It is a full calibration on samples, with its reference campaign and its upkeep over time.
No, on the other hand, if you want to know when the consistency stops changing. A stabilisation is detected on the raw signal, with no absolute value. That is often the real question, and it costs much less.
Can a batch be released on this measurement?
Not automatically. A specification expressed at a given shear rate is released on a measurement made at that shear rate.
The model can, however, be integrated into a documented control strategy, up to contributing to real-time release. It is then a full analytical validation project, not a consequence of the installation.
How many samples are needed to calibrate?
The question is not the number but the coverage. A model built on thirty batches that resemble each other is less robust than a model built on fifteen batches that sweep the real variability, difficult batches included. How many samples are really needed?
Our product is very viscous and opaque. Does that rule it out?
No, but it points towards diffuse reflectance rather than transmission, and towards a measurement point where the material flows past. Opacity is less of a hindrance than people think: it is heterogeneity at the scale of the probed volume that causes the problem.
And depending on your question, another quantity sometimes answers better. If you are looking for the end of a dispersion, the size distribution is a more direct criterion.
What sets the time before a usable criterion?
Knowing whether the signal exists and whether it discriminates is quick. Building a robust criterion calls for representative batches, having seen the real variability. A trial that has seen only one batch says nothing about routine.
And if the answer points elsewhere?
Then it is better to know it from a feasibility trial than after an installation. A documented conclusion is a deliverable: how the signal behaves, and which quantity would better answer the question asked.
Describe the operation, your current viscometer and your specification. You will know what is reachable.
Forty-five minutes is enough to know which quantity governs your viscosity, whether your reference method is precise enough to calibrate a model, and whether a state criterion answers better than a value. You will leave with all three.