Alkyd resins are esterification products of a polyol, a polybasic acid or anhydride, and a fatty acid or drying oil. The reaction is a reversible polycondensation that liberates water, removed at 180-240 C to push conversion forward. Two quality numbers govern the batch: viscosity, which tracks molecular weight build-up, and acid number, which tracks residual carboxyl content in mg KOH per gram of resin. Overshoot the target acid number and the resin sits outside specification; overshoot the other way and the reactor is closer to gelation than the operator can see.

The reference technique for acid number is potentiometric titration under ISO 2114. The result is authoritative. The problem is that in a reaction whose kinetics steepen near endpoint, an authoritative number that lands twenty to sixty minutes after the sample was drawn is a lagging indicator. The industry has spent the last decade putting spectroscopic probes in the recirculation loop to close that gap. This piece sets the two approaches side by side.

What acid number really tells the operator

Acid number is defined as the milligrams of KOH required to neutralise the free acidity in one gram of resin. In an alkyd batch, that free acidity is the residual carboxyl on partially reacted polybasic acid or anhydride, plus any unreacted fatty acid. As esterification proceeds, water leaves, ester bonds form, and the acid number drops. A specification of, say, 8-12 mg KOH/g on a medium-oil alkyd is a specification on how far the polycondensation has gone.

Acid number and viscosity are related but not interchangeable. Viscosity is dominated by molecular weight distribution and by any late-stage branching or gel-precursor formation; it can climb sharply near endpoint while acid number is still moving down smoothly. Serious coatings producers watch both. A viscosity read alone can miss a batch that is still consuming acid; an acid number alone can miss a batch that is about to gel. The article on inline endpoint control for amino-resin polycondensation makes the same distinction for urea-formaldehyde chemistry.

The reference method: ISO 2114 potentiometric titration

ISO 2114 - the standard maintained since 2000 for polyester resins and for binders in paints and varnishes - defines both a partial acid value and a total acid value. Partial covers free acids and half of any anhydride groups; total covers free acids plus fully hydrolysed anhydride groups. The determination is a potentiometric titration with 0.1 mol/L potassium hydroxide in ethanol, on a resin sample dissolved in a solvent blend (typically toluene / ethanol or toluene / 2-propanol depending on the resin’s solubility). Metrohm’s application note AN-T-164 and the SI Analytics / Xylem application paper describe the procedure at working detail; Mettler-Toledo publishes an equivalent workflow for powder coating resins.

The strengths of the method are well established. It is a primary chemistry - stoichiometric neutralisation of an acid by a strong base - so once the sample is properly dissolved the result is traceable and defensible. Modern potentiometric titrators with dynamic equivalence-point detection handle the endpoint automatically; auto-samplers push per-shift throughput to 40-60 titrations without operator intervention. For a lab that already runs hydroxyl number, saponification value, and iodine value on the same platform, adding acid number costs almost nothing.

The weaknesses are operational, not chemical. Sample draw, cool-down, dissolution, titration, and result entry runs 15-45 minutes even in a well-equipped lab. The resin is hot and viscous; the solvent blend is flammable; the samples are consumed. In a plant with several parallel reactors, sample logistics become the rate-limiting step near endpoint. The plant sees the reaction as a series of discrete points along the batch curve, and interpolating between them is a judgement call.

For batch release and compliance testing, ISO 2114 titration remains the anchor. For minute-to-minute reactor control, the plant needs something else in parallel.

What inline spectroscopy actually measures

Neither NIR nor Raman measures acid number directly. What they measure is a spectral fingerprint of the reactor contents; a multivariate model then maps that fingerprint to a predicted acid number. The point matters because “inline acid number” is often written as if it were a primary measurement, which it is not.

The peer-reviewed literature on alkyd polycondensation modelling - Aigbodion and co-workers in the Journal of Coatings Technology and Research is representative - shows that acid number over time on a well-controlled cook is a smooth monotonic function of process variables. That is exactly the kind of trajectory a chemometric model, calibrated on a reasonable range of batches, can track once the sample-presentation problem is solved.

NIR: where it fits

Near-infrared has been used on coatings and resin reactors since the mid-1990s. Its advantages on alkyds are practical: fibre-optic transmission probes are cheap in ATEX-rated form, path lengths in the 1-5 mm range work on medium-viscosity resins, and detection is thermoelectrically cooled InGaAs, which sits in a plant panel without complaint. Overtone and combination bands from the O-H stretch (residual hydroxyl and any water) and from the ester C-H region carry indirect information about esterification progress. The buyer’s guide to inline NIR analyzers sets out the vendor field.

The limits are known. NIR bands are broad and overlap. Building a model that generalises across raw-material lots and across small formulation shifts requires a calibration set larger than most plants want to build. Coloured or high-solids resins in the darker grades attenuate signal to the point where transmission geometry stops working; diffuse reflection helps but adds its own drift. NIR is not a bad choice, but it is a choice that pays back only when the plant runs a stable slate of alkyds and can afford model maintenance.

Raman: where it fits

Raman scattering is a different physical channel: instead of absorption of overtones, it reads inelastic scattering off covalent bond vibrations, most usefully in the 300-1800 cm-1 fingerprint region. For alkyds, three regions carry endpoint information: the ester C-O-C stretch near 1040 cm-1, the aromatic ring modes near 1000 cm-1 (present when ortho-phthalic or isophthalic acids are used), and the C=O stretch of the ester carbonyl near 1600 cm-1. Feasibility work published by process-analytics vendors reports high correlations between Raman-derived predictions and reference acid number and viscosity across the working range of a typical medium-oil alkyd cook.

Raman’s operational profile complements NIR’s. Bands are narrow, so overlaps between analyte and matrix (solvent - white spirit D40, xylene, occasionally butanol) can be resolved by the model rather than avoided by the geometry. The measurement is largely temperature-insensitive over the 20-250 C span of interest. Water does not respond, which matters in a condensation reaction that liberates water. And the immersion probe can sit downstream of the filtration node in the recirculation loop, seeing the batch as the plant sees it.

The costs are also known. Fluorescence from unsaturated fatty-acid residues (linseed, tung, dehydrated castor) can raise the baseline enough to swamp the signal on some resins; 785 nm excitation with careful baseline correction handles most cases, and time-gated Raman handles the harder ones, at higher hardware cost. Model build is slower than for NIR because the fingerprint region has more information but each band carries less. The Raman versus NIR decision framework is where the trade-off should be resolved on a project-by-project basis.

Vendors offering inline analyzers on resin production

The process-Raman field for coatings and resins covers Endress+Hauser (Raman Rxn2 / Rxn4), Mettler-Toledo AutoChem (ReactRaman 802L), HORIBA Scientific, Bruker (via the Tornado line), MarqMetrix under Thermo Fisher, and Gekko Photonics, whose Spectrally INLINE 785 nm analyzer is listed in Gekko’s public catalogue as an application for acid number and viscosity in alkyd resins. On NIR, the field is Bruker Optics MATRIX-F, Metrohm NIRS DS2500, Yokogawa, and PerkinElmer/Revvity. The inline Raman buyer’s guide covers vendor selection in detail.

Practical picture: what a plant actually does

The plants that have solved this problem well do not choose between titration and spectroscopy - they run both. The titrator anchors the specification and the chemometric model. The spectrometer is the near-real-time indicator; it tells the operator when to draw the next confirmatory sample, when to hold temperature, and when to consider quench. On a well-run installation, the number of titrations per batch falls, but titration does not disappear.

Two cautions before committing. Model maintenance is a real cost - a chemometric model that predicts acid number to within a few tenths of a mg KOH/g on the batches it was calibrated on will drift as raw materials, catalyst lots, and formulation targets shift. Every serious deployment budgets technician time to keep calibration current; the article on PAT ROI without inflating numbers is worth reading before the business case is signed. And the value of an inline number is set by what the operator can do with it: if reactor control cannot act on the reading, the analyzer becomes an expensive display.

For batch release, ISO 2114 titration stays the standard. For continuous reactor control, spectroscopy has moved from research to routine on alkyds. The two now coexist rather than compete.