in situ ftir spectroscopy

How to Decide Whether In Situ FTIR Spectroscopy Fits Your Process

admin 18 min read
How to Decide Whether In Situ FTIR Spectroscopy Fits Your Process

Start with the process decision, not the instrument

The process decision should determine whether you investigate in situ FTIR spectroscopy, not the appeal of a particular instrument. Start by naming the chemical change you need to observe, the time at which that change matters, and the action that the information would support. If an important change could occur between offline measurements and obstruct a defined reaction question, a time-linked in situ measurement merits investigation. If your existing observations already answer the question, the case for adding another measurement is less clear.

The reaction question should be specific enough to test. You might need to distinguish the disappearance of a reactant from the appearance of a product, determine whether an intermediate appears, relate a composition change to operating conditions, or decide when a process has reached a defined state. A general desire for more data is not yet a measurement specification. State the decision first.

Use these questions to frame the evaluation:

  • What chemical or process change must you detect?
  • What could happen between your current offline samples?
  • What decision would change if that event were observed in time?
  • Do you need a qualitative spectral indication, a concentration estimate, a kinetic profile, or a result suitable for a control strategy?
  • Which reference measurement currently supports that decision?

A defined question also gives you a basis for judging the result. A time series may be useful when it closes a specific information gap, but more spectra do not automatically produce a more defensible process decision. Keep the first evaluation narrow: identify the event, the required timing, and the evidence that would make the result useful.

What in situ FTIR measures inside a running process

In situ FTIR measures directly in the reaction medium while the process is running, so the reaction can be monitored without sampling or transfer. The defining feature is the relationship between the measurement and the process material: the spectrum is collected from the reaction medium rather than from a portion removed for separate analysis. That distinction matters whenever handling can alter the material before the spectrum is recorded.

Offline analysis can involve sampling, transfer, dilution, cooling, or exposure to air before measurement. Those steps can change a process sample before it reaches the analytical method, particularly when the system contains unstable intermediates, temperature-sensitive equilibria, or phase-dependent reaction behavior. In situ measurement avoids the specific sampling or transfer step described here; it does not by itself establish that every process question can be answered quantitatively.

The measurement arrangement therefore belongs in the definition of the analytical system. Ask where the process material meets the optical path, whether that location represents the material you need to understand, and how the arrangement behaves under the actual process conditions. A complete process-FTIR evaluation considers the spectrum, calibration, reference method, sampling arrangement, installation, software, and lifecycle support together. Treating the spectrometer as an isolated component can leave the central process question unresolved.

The practical distinction is not simply in situ versus offline as a matter of convenience. It is whether the measurement preserves the connection between the spectrum and the running chemistry closely enough to answer your defined question. Compare the material at the measurement location with the material represented by your offline reference samples, then identify any handling step that could create a difference. That comparison is part of suitability assessment.

What the spectrum can reveal about reaction progress

The FTIR spectrum contains information about molecular vibrations and characteristic absorption bands. FT-IR spectroscopy can characterize molecular structures, monitor chemical reactions, and quantify analytes in diverse materials. In real-time reaction monitoring, changes in those spectral features can be related to reaction kinetics and conversion rates. The spectrum is therefore a source of chemical information; the reported process value depends on how that information is interpreted and validated.

A time-linked series adds the process dimension. This can help you examine when a change begins, how it develops, and how it relates to the process conditions recorded at the same time. The useful output may be a trend, a comparison between stages, or a model-derived estimate.

During reaction development, this time-linked record can make an experiment data-rich. Continuous mid-infrared acquisition can support kinetic modeling, design-of-experiments work, and statistical analysis while reducing reliance on time-consuming offline sampling and sample preparation. Use the series to compare how operating conditions change the trajectory, identify reaction-driving factors, and determine which stages need closer investigation. The practical value is not simply a denser plot: these outputs help shape the next experiment or model. They remain exploratory until the relevant signal is linked to a reference method and a stated model or decision criterion. A rich time series can improve the question asked next without proving that a concentration estimate or endpoint rule is ready for use, and it can clarify which evidence is still missing before operational use or a consequential process action.

Real-reaction spectra can also expose features that are difficult to anticipate from starting materials and final products alone. In a complex reaction system, in situ spectra can show whether spectral changes are subtle or significant and can indicate possible intermediates through new peaks absent from the starting materials and products. Treat a new feature as a question to investigate, not as an identified compound by default. Assignment requires appropriate chemical interpretation and supporting evidence.

An endpoint-related signal must be tied to a defined decision rule. A change in an absorption band may indicate reaction progress, but it does not alone establish that the process has reached an acceptable endpoint or that a concentration result is fit for a consequential use. If you need a quantitative endpoint, define the reference method, expected range, validation criteria, and permitted uncertainty before treating the spectral trend as a process result.

Match measurement timing to the change you need to observe

A time-resolved measurement can support several kinds of work:

  • Following reaction progress during development experiments.
  • Comparing kinetic behavior across operating conditions.
  • Linking composition changes with scale-up observations.
  • Supplying measured attributes for monitoring or process-control work when the measurement and model have been established for that use.

These uses require different evidence. A development trend, a concentration estimate, and a control input should not be treated as interchangeable outputs.

Your timing specification should include the event, the expected duration of the change, the interval between current samples, and the time by which a decision must be made. If your current sampling schedule already resolves the event and supports the decision, additional time-linked data may add little. If the event can occur between samples, define how observing it would change the next process action before choosing a measurement frequency.

The required frequency is therefore a property of the process question, not a specification to judge in isolation. A slowly changing composition trend may need a different observation interval from a short-lived intermediate or a rapid endpoint transition. Map the expected process event against the current sampling interval and the decision deadline. Then ask whether the proposed measurement can produce observations at the points that distinguish the possible process states.

Define the stream and sampling interface before judging suitability

The process domain should be documented before you judge whether an in situ arrangement is suitable. Define the analytes, concentration extremes, composition changes, temperature, pressure, water, solids, bubbles, and required decision. These facts describe the material and operating envelope that the spectrum and any later model must represent. A single convenient sample may not represent the range encountered during the process.

Create a process record with the following fields:

  • Target analytes, known interferents, and expected concentration range, including low and high extremes.
  • Composition changes during the reaction or process step.
  • Temperature and pressure at the proposed measurement location.
  • Water content, solids, bubbles, phases, and any changes expected during operation.
  • The location from which material will be measured and why that location represents the decision-relevant stream.
  • The optical path or sampling interface, including the conditions that could affect contact with the material.
  • The required measurement timing and the process event it must resolve.
  • The reference method and the process decision that the result must support.

The sampling interface deserves its own review because it connects the process to the spectrum. Ask whether the measured material is representative of the stream at the moment of interest, whether the optical path remains usable across the operating range, and whether the arrangement introduces a difference between the measured portion and the material used for reference testing. Treat turbidity, solids, bubbles, and potential fouling as process-specific feasibility questions rather than assuming that a configuration suitable for one stream will suit another.

Review the interface as a chain from process material to reported spectrum. For each operating condition, record whether the unresolved issue concerns representativeness, optical access, or the model domain. Then identify what observation would distinguish those possibilities: a comparison of material from different locations, a check at a different process stage, or reference data linked to the same condition. If the distinction cannot yet be made, label it as a feasibility question rather than treating a clean spectrum as proof that the interface remains suitable. This record also helps explain why a configuration that works during one stage may require additional examination when solids, bubbles, temperature, pressure, or composition change.

The full system includes more than the measurement location. Evaluate the spectrum, calibration, reference method, sampling arrangement, installation, software, and lifecycle support as one measurement system. This approach helps separate a chemistry problem from an interface problem, a calibration problem, or an implementation problem. Record the unanswered questions before comparing instruments.

The operating envelope should describe changes, not only nominal conditions. Record the lowest and highest expected temperature, pressure, concentration, water content, solids burden, and gas or bubble condition at the measurement point. Note when those conditions occur and whether they coincide with the process decision. A measurement interface that is representative during one stage may require separate examination during another stage with a different composition or physical state.

Representativeness also includes the relationship between process material and reference sample. If the material changes during sampling, transfer, dilution, cooling, or air exposure, the reference result may describe a different state from the spectrum collected in the process. Document the collection time, location, handling steps, and process conditions for both measurements so that later calibration comparisons have a defined basis.

An infographic checklist for a representative measurement system: process domain, operating conditions, measurement location, timing requirement, and reference basis.
A suitability review starts by describing the material, interface, timing, and decision.

Use chemometric calibration to translate spectra into process information

Chemometric calibration connects spectral features with a process quantity or identity that you need to estimate. Chemometric models correlate reference concentrations or identities with spectral features, while held-out validation data tests whether the model measures the relevant signal rather than fitting random noise. Without that relationship, a changing spectrum may show that the material is changing, but it does not automatically provide a defensible concentration or process value.

The reference method sets a material boundary on the model. Predicted values cannot be more accurate than the reference method used for training, and the reference method’s reproducibility sets the model’s accuracy floor. Your calibration plan should therefore specify how reference samples are collected, how they cover the process domain, and how their results will be judged for consistency. A sophisticated model cannot remove limitations already present in its training values.

The calibration set must span the expected concentration range, including the extremes. That requirement is not satisfied by collecting only easy, central, or late-stage samples if the process also produces low or high concentrations. Include the composition conditions that the model will encounter during the intended use, then examine whether the available reference data actually cover them.

Ask what the model is intended to report and what evidence supports that report. A model for qualitative change detection has a different task from a model that estimates concentration for a process decision. Define the analyte or identity, reference procedure, concentration domain, sampling relationship, and decision threshold separately. The model should be evaluated against that stated task rather than against a general impression that the spectra look informative.

The calibration domain should also reflect the process conditions that influence the spectrum. Changes in composition, temperature, feedstock, or product specification can alter the relationship between spectral features and the reported quantity. Keep those conditions linked to the reference values used for model development. If an intended operating state is outside the represented domain, treat extension of the model as further technical work rather than silently extrapolating.

Validate the model with independent process evidence

Independent process evidence is needed to test whether a multivariate model works beyond the data used to build it. Validate the multivariate model with independent samples against predetermined performance criteria rather than judging it only on calibration data. Held-out validation data helps test whether the model measures the intended spectral relationship instead of overfitting random noise.

Set the validation plan before examining the result. Define:

  • Which samples, batches, runs, or operating conditions will remain independent from calibration.
  • Which reference method will supply the comparison values.
  • Which performance criteria will determine whether the model answers the stated process question.
  • Which concentration and composition regions must be represented.
  • What action follows if a sample falls outside the calibration domain or fails the criterion.

Predetermined criteria keep a favorable calibration plot from becoming the only basis for a process decision.

We develop models by calibrating against your chemistry and reference method, then validate them on unseen batches. That work remains tied to the chemistry, reference procedure, and intended process domain you provide. It does not turn a general FTIR principle into evidence that a particular stream or decision is already supported.

Transfer should be treated as a technical question. A change in measurement conditions or instrument can require adjustment, additional evidence, or model redevelopment. Include the proposed measurement location, process range, installation conditions, and any change in reference practice in the validation discussion. A model that performs under one set of conditions is not automatically established for another.

Where a proposed implementation changes from laboratory evaluation to a pressurized installation, treat model transfer as a hypothesis to verify. Paeonia Innovations states that the OrionIR High-Pressure Version uses the same measurement principle as the Lab version, allowing chemometric models built during evaluation to transfer to a pressurized installation. That statement can inform the evaluation plan, but it does not remove the need to confirm the process domain, measurement conditions, reference relationship, and independent performance evidence for the intended installation. Record which conditions remain unchanged and which require additional evidence.

Review validation as an evidence chain rather than a single statistic. The independent sample must represent the material and conditions for the intended use, the reference value must come from the defined reference method, and the performance criterion must correspond to the process decision. If any link changes, record whether the model remains within its established domain or needs additional work.

An infographic checklist for model validation: independent samples, reference method, performance criteria, domain coverage, and follow-up action.
Independent evidence is strongest when its samples, reference values, criteria, and actions are defined in advance.

Connect model performance to maintenance and process use

Model performance depends on the complete measurement system after installation as well as during development. Spectrum collection, calibration, reference testing, sampling arrangement, installation, software, and lifecycle support should be evaluated together. Your implementation plan should identify who reviews model output, how changes in the process are recorded, and what evidence is needed before the result influences a process action.

We treat residual monitoring as essential to model maintenance. A rising spectral residual indicates that the instrument is detecting something the model was not trained on. The signal should therefore trigger investigation rather than silent acceptance. Examine whether the cause is a changed feedstock, temperature, product specification, process state, measurement condition, or another difference from the training domain before deciding what the model output means.

A rising residual is therefore a boundary signal, not a complete diagnosis. It indicates that the instrument is detecting something outside the model’s training, but it does not by itself identify whether the change arose from feedstock, temperature, product specification, process state, or measurement conditions. Pair residual review with the process record and reference comparisons. If the changed condition is relevant to intended use, decide whether the model domain should be extended, the model revalidated, or the output kept outside the decision path until evidence is available. The maintenance record should preserve that reasoning, so later users can distinguish an investigated change from an accepted result.

A maintenance plan should state how residuals, reference comparisons, and process changes will be reviewed. We offer ongoing chemometric-model support, including revalidation as feedstock, temperature, or product specifications change. If you rely on a model for a recurring process decision, define the conditions that require review and the evidence that permits continued use. A model is a maintained analytical method, not a one-time software setting.

Integration should begin with the decision path. Identify where the spectral or model result will be stored, who will see it, which process variable or measured attribute it informs, and what action is permitted at each confidence or validity state. In-line multivariate measurements can support automated monitoring and process control, but the intended use still requires a suitable model, defined criteria, and site-specific assessment.

Define operational gates for continued use. A valid model output can be reviewed against residual behavior, process conditions, reference comparisons, and the calibration domain before it is used in a process decision. When the signal indicates material outside the training domain, pause the interpretation at that boundary and investigate the changed condition. This creates a clear distinction between collecting a spectrum, accepting a model result, and acting on that result.

The data path should be specified alongside the analytical method. Record the measurement output, model version, relevant process conditions, validity indicators, and the decision that follows. If the result is intended to support monitoring or control, connect the output to the measured attribute and the permitted response. Integration is therefore part of the measurement-system assessment, not a separate task to complete after the model is accepted.

Account for pharmaceutical and continuous-manufacturing expectations

Pharmaceutical manufacturing uses PAT technologies and chemometrics for real-time process monitoring, process control, and Real Time Release Testing in development and manufacturing. Process models may support process design, scale-up, site transfer, process monitoring, and process control, and may form part of a control strategy for consistent product quality. These are potential roles for a validated process model, not automatic consequences of installing an FTIR measurement.

ICH Q13 identifies equipment design and system integration, measured-attribute process controls when used, and justification of control-strategy models as relevant continuous-manufacturing considerations. An ICH Q13 context therefore makes the measurement arrangement, data path, model purpose, and supporting rationale part of the technical discussion. The instrument specification alone does not answer those questions.

For a pharmaceutical or continuous process, connect each proposed output to its intended role:

  • Development data may support process understanding and scale-up work.
  • A monitoring result may show a measured attribute over time.
  • A control input requires defined use conditions, model justification, and system integration.
  • A release-related application requires its own performance and validation basis.

Keep those roles separate in the project plan. Do not describe a feasibility spectrum, a development model, or a monitoring trend as regulatory acceptance.

Your site must determine the applicable validation, quality, data-integrity, installation, and change-control expectations for the intended use. The technical evaluation can prepare the evidence by defining the process domain, measurement arrangement, reference method, model, validation data, and integration path. The final suitability decision remains specific to your chemistry, site, quality system, and intended process use.

The pharmaceutical context changes the questions you must document, but it does not remove the analytical questions. Establish what the measurement represents, how the model was developed, which data tested it, how changes will be managed, and how the output enters the control strategy. For sites in different regions, the applicable quality and regulatory interpretation remains a site-specific matter; the technical record should make the measurement basis and intended use clear.

Run a bounded feasibility assessment on the real process

Use the following sequence:

  1. State the question and decision. Identify the chemical change, the time window, and the action that the result would support. Investigate time-linked in situ measurement when an important change could occur between offline measurements and obstruct that question.
  2. Document the operating envelope. Record analytes, concentration extremes, composition changes, temperature, pressure, water, solids, bubbles, process phases, and the required decision.
  3. Describe the measurement interface. Identify where the material meets the optical path, how that location represents the relevant stream, and which conditions could affect the measurement arrangement.
  4. Collect spectra from the real process. In a complex reaction system, compare the in situ series with spectra from the starting materials and products when investigating possible intermediates.
  5. Define the reference method. Establish how reference values will be generated, how they relate to the measured material, and what reproducibility they provide.
  6. Build the calibration domain. Ensure that the expected concentration range, including its extremes, and relevant composition conditions are represented.
  7. Test independent samples. Validate the multivariate model against predetermined performance criteria rather than judging it only on calibration data.

The first spectra can answer a limited but useful question: does the real process produce a measurable change that appears relevant to the reaction question? A visible change can justify calibration work, while subtle or confounded changes can indicate that the project needs a narrower question, a different arrangement, additional reference data, or a different analytical approach. Do not treat the first spectral series as a completed method.

Record the assessment as a chain of decisions rather than a collection of plots. For each proposed output, connect the process question to the sampled material, spectral feature, reference method, calibration domain, validation result, and intended action. This record makes it easier to identify whether the remaining uncertainty concerns chemistry, representativeness, optical access, model performance, installation, software, or lifecycle support.

Use explicit decision boundaries at the end of the assessment:

  • Continue if the process question is defined, the measured material is representative, and the spectra show information relevant to that question.
  • Move to calibration work only when the reference method and expected process domain can be described.
  • Extend testing when independent samples, operating extremes, or interface conditions remain unrepresented.
  • Stop or reconsider the approach when the measurement cannot address the defined event or when the remaining evidence cannot support the intended use.

These boundaries keep feasibility evidence proportional to the decision it is meant to inform.

Separate a useful signal from a validated process decision

A quantitative result requires a defensible calibration and reference basis. The calibration range must cover the expected extremes, and independent samples must be tested against predetermined performance criteria. A favorable fit to calibration data is not enough.

A process-control or regulated use adds another boundary: the model, measurement arrangement, integration path, and intended action must be justified for that use. A feasibility assessment can show whether further work is warranted. It does not, on its own, establish control readiness, regulatory acceptance, or replacement of offline methods. Keep those conclusions tied to the evidence actually generated for your process.

Use three separate questions when reviewing the result:

  • Is there a chemical signal? The spectra may show change, a trend, or a possible intermediate.
  • Can the signal support a reported value? Calibration, reference reproducibility, domain coverage, and independent validation must support that interpretation.
  • Can the reported value support the intended action? The process use, integration path, acceptance criteria, and site requirements must support that decision.

A project can answer the first question while the second or third remains open.

A numbered infographic showing three steps from a spectral signal to a validated process decision: identify signal, support a value, and support action.
A visible spectral change may be useful evidence without yet supporting a reported value or process action.

Prepare the process facts for a feasibility discussion

Your next step is to assemble a concise process brief before deciding whether to investigate this approach further. Include the defined reaction or process question, target analytes, concentration range and extremes, composition changes, temperature, pressure, water, solids, bubbles, proposed measurement location, timing requirement, current reference method, and intended process decision. Add the sampling or optical-interface constraints that could affect representativeness.

Use that brief to test the complete measurement system rather than comparing instruments in isolation. The assessment should establish whether a time-linked measurement addresses a real information gap, whether the process material can be measured representatively, whether calibration can cover the expected domain, and whether independent evidence can support the proposed use. If those questions remain open, keep the conclusion at feasibility assessment rather than moving directly to a consequential process decision.

A process-specific record gives you a practical basis for discussing measurement arrangement, reference testing, calibration, validation, installation, software, and lifecycle support. It also makes the boundary clear: the purpose is to determine whether the approach fits your defined liquid-phase process question, not to assume suitability from general FTIR principles. A qualified professional at your site must decide individual process, regulatory, safety, and implementation suitability.



Next step

Have a question for Paeonia Innovations?

Get in touch