Your sample is already wrong by the time it reaches the lab

The standard way to find out what is happening inside a reactor is to take a sample out of it. Someone opens a valve, fills a vial, labels it, walks it to the lab, and some hours later a number comes back. That number is treated as the truth about the reactor. It usually is not.

Three errors, stacked

The reaction does not stop when the sample leaves. Unless the sample is quenched instantly and completely, conversion continues in the vial. For a fast reaction the number you eventually measure describes a state the reactor passed through minutes ago and has already left. Quenching helps, but a quench is itself a chemical intervention, and it has to be validated for every chemistry.

The sample is not the bulk. A grab sample is taken from one point, usually a convenient one — a valve at the bottom of the vessel or a bypass loop. In any stream with a solid phase, a second liquid phase, or an incomplete mixing pattern, the composition at that point is systematically different from the composition of the reactor.

The answer arrives after the decision. The lab is not slow because anyone is careless; chromatography takes as long as it takes. But if the result lands two hours after the sample was pulled, it cannot inform a decision about a reaction that finishes in ninety minutes. The measurement becomes a record for the batch report rather than an input to control.

What each error costs

The first error costs accuracy, and it does so asymmetrically — always in the direction of more conversion than the reactor actually had. The second costs reproducibility, and it is the reason process data from a well-run plant can still be noisy for reasons nobody can trace. The third costs everything you would have done differently: the endpoint you would have called earlier, the feed you would have slowed, the batch you would not have lost.

Sampling frequency compounds all three. A process sampled every thirty minutes has, at best, a thirty-minute resolution on any event that happens inside it. Exotherms, phase changes, precipitation onset and impurity formation all happen faster than that, and in a sampled dataset they appear as a single anomalous point or as nothing at all.

What changes with an inline measurement

Moving the measurement onto the process removes all three at once. A flow cell plumbed into a recirculation loop or a representative side stream is still fed by the process, but nothing is carried anywhere, nothing sits in a vial reacting, and the measurement repeats as fast as the instrument can read — every second, in the case of the instrument we build. That changes the questions you can ask.

• Endpoint by observation, not by clock. Instead of holding for the validated worst-case time, you watch the concentration curve flatten and stop.
• Kinetics from one run. A continuous concentration trace is a kinetic dataset. Sampled points are a scatter plot you fit a curve through and hope.
• Transients become visible. Induction periods, intermediate accumulation and short-lived species show up as features, not as noise.
• Deviation is detected while it is still cheap. A feed that started an hour ago at the wrong rate is a correctable problem; discovered at the end of the batch it is a disposal cost.

The objection: inline instruments are hard to run

Historically true. A traditional process spectrometer meant an interferometer with moving parts in a plant environment, an optical fibre that could not be bent past a radius, a purge line, and a specialist to interpret the spectra. Every one of those is a reason a good measurement never made it out of the lab.

Three things changed. Instruments got small enough to sit at the process rather than beside it. Removing moving components removed the alignment failure mode that made plant deployment fragile. And chemometric models now convert a spectrum into the concentration a process engineer asked for, so the person reading the screen does not have to be a spectroscopist.

Where offline analysis still wins

Inline measurement does not replace the lab, and any vendor who says otherwise is selling. Chromatography still resolves components an infrared spectrum cannot separate. Trace-level impurity work still needs the sensitivity of a dedicated analytical method. Reference analysis is also what an inline model is calibrated against — you cannot build the chemometric model without it.

The useful split is this: the lab defines what the truth is; the inline instrument tells you where you are on the way to it, continuously, while you can still act.

The OrionIR® is a 0.1 mL transmission flow cell that plumbs into a recirculation loop with hand-tight fittings, and reads aqueous and strongly absorbing streams at a 20 µm path. See the instrument.