mid infrared spectroscopy

Mid-Infrared Spectroscopy for Process Monitoring: A Practical Feasibility Framework

admin 17 min read
Mid-Infrared Spectroscopy for Process Monitoring: Principles, Applications, and Feasibility

Answer the feasibility question before choosing an instrument

Mid-infrared spectroscopy deserves process-specific evaluation when its measurement can produce information that matches the reaction or stream decision you need to make. The method is not suitable merely because it is fast, spectroscopic, or described as inline. Start with the decision, then work backward to the analyte, sampling arrangement, reference method, calibration domain, and response time.

Online infrared spectroscopy has been applied in process analytical technology (PAT) to support scale-up involving hazardous species. That establishes a relevant application area, not a universal result for every chemistry or process. Your process still determines whether the measured spectral variation is usable.

ICH Q13 states that a process-monitoring sampling strategy should reflect the intended use of the data and process dynamics. The strategy includes the variables monitored, measurement method and frequency, sampling amount and location, statistical method, and acceptance criteria. Those details turn a general interest in spectroscopy into a feasibility question you can test.

Use this framework to separate five questions:

  • What chemical information can the spectrum contain?
  • What concentration or process value, if any, will a model infer?
  • Can the sampling arrangement present a representative stream under operating conditions?
  • Does the calibration evidence cover the chemistry and concentration range you expect?
  • What decision will the result support, and what happens when the measurement is unavailable?

If those questions remain unanswered, choosing an instrument is premature. Define the information requirement first.

Place mid-infrared spectroscopy within infrared measurement

Infrared spectroscopy is the broader measurement family: it measures how infrared radiation interacts with matter through absorption, emission, or reflection and can help identify chemical substances and functional groups in solids, liquids, and gases.
The practical distinction is more useful than a memorized label. A stated spectral range tells you what radiation the instrument measures; it does not, by itself, tell you which analytes can be quantified in your stream or whether the resulting signal is sufficiently specific. Ask for the operating range, optical path, sample presentation, and intended application together.

The exact range should therefore be tied to the instrument and method under evaluation rather than treated as a guarantee about all mid-infrared applications. Keep the terminology precise: infrared spectroscopy describes the measurement principle, while mid-infrared spectroscopy identifies the spectral region used in the particular method.

Our OrionIR® mid-IR spectrometer provides one spectrum every second across 900–3500 cm⁻¹ through aqueous, strongly absorbing, and low-concentration streams. That is a stated OrionIR® specification, not a definition of every mid-infrared instrument or evidence that every process stream is compatible.

Use the range as a starting point for a technical discussion. Compare it with the bands relevant to your chemistry, the absorbance expected at the selected path length, and the process conditions at the proposed sampling point. A nominal spectral range cannot resolve interference, fouling, temperature effects, or model validity on its own.

Read the spectrum as chemical information

A measured infrared spectrum records how the sample interacts with infrared radiation across the instrument’s measured range. Absorption, emission, or reflection produces spectral features that can help identify chemical substances and functional groups. Infrared spectroscopy can also help analyze molecular configurations.

Those features are chemical information, but they are not automatically a concentration report. A peak or band may reflect a functional group shared by several species, overlap with another component, or change with the physical setting of the measurement. Interpretation depends on the mixture, concentration range, temperature, path length, and sampling configuration.

For a process application, begin by asking what the spectrum could distinguish directly. You might be interested in the presence of a functional group, the appearance or disappearance of a reactant-related feature, or a changing pattern associated with reaction progression. You then need to establish whether that pattern is sufficiently specific and stable for the process decision.

A useful interpretation brief names the chemical question before it names the instrument. Record:

  • The compounds, functional groups, or process attributes you need to observe.
  • The expected concentration range for each relevant component.
  • Components likely to contribute overlapping or strongly absorbing features.
  • The temperature, pressure, phase, solids content, and flow condition during measurement.
  • Whether you need a qualitative change signal, a quantitative concentration estimate, or a process trend.

This distinction prevents a common category error. A spectrum is the measured response; its meaning comes from chemical interpretation and, where needed, a validated relationship between spectral variation and a reference result.

An infographic checklist for reading chemical information from an infrared spectrum. It lists target information, concentration range, overlapping features, measurement setting, and required output.
A spectrum becomes useful only when its chemical question and process setting are explicit.

Separate direct spectral interpretation from concentration prediction

A spectrum is a measurement, whereas a model-based concentration is an inference from that measurement. Direct interpretation examines spectral features and patterns for chemical meaning. Quantitative prediction uses a relationship learned from spectra paired with reference measurements.

PAT methods in pharmaceutical manufacturing commonly combine spectroscopy or other optical sensors with chemometric multivariate data analysis. Chemometrics is useful when several spectral variables jointly carry information about an analyte or process state, especially when individual bands overlap. The model does not remove the need to understand the chemistry; it formalizes a relationship that must be tested against representative samples.

A yeast-fermentation study illustrates the distinction. The monitored analytes were glucose, ethanol, and biomass. Because the fermentation medium was complex, multivariate methods, including partial least-squares (PLS) and multiple linear regression (MLR) models, were often needed to evaluate the spectra. That example supports the role of models in complex mixtures, not a claim that the same model or analytes apply to your process.

Treat each output according to how it was produced:

  • A raw spectrum is direct instrument output.
  • A selected band or spectral feature is an interpreted signal.
  • A predicted concentration is a model output tied to calibration and reference data.
  • A process decision is a further use of that output under defined acceptance criteria.

Ask which step your proposed application depends on. If you need a concentration value to trigger an action, the model’s calibration population, reference method, validation samples, and operating domain matter as much as the optical measurement. If you need only a reaction trend, a different evidence package may be relevant, but the trend still needs to be shown under your process conditions.

Map the process questions spectroscopy may answer

Process monitoring uses spectroscopy to connect changing chemical information with a defined process question. The relevant question is not simply whether a spectrum changes, but whether the change relates to the reaction or process variable you need to understand.

A chemical-development example used an online spectroscopic process signature representing starting-material concentration against reaction time. The signature showed that a reaction at reflux temperature of 76 °C proceeded faster than one at 65 °C, while the lower-temperature reaction did not reach completion. This shows how a time-resolved signal can compare process trajectories and expose a difference between operating conditions.

You can frame a similar evaluation around questions such as:

  • Is a starting material being consumed at the expected rate?
  • Does an intermediate appear or decline within the process window?
  • Does a change in temperature or another operating condition alter the reaction trajectory?
  • Can a process trend be observed at the point where you need to act?
  • Is the measurement intended for development understanding, scale-up support, monitoring, or a defined control decision?

The last question sets the evidence requirement.

Use application evidence without turning examples into guarantees

Published applications show where infrared or related spectroscopic measurements have been used; they do not establish suitability for your process. Keep the matrix, measurement mode, analytes, scale, and model conditions attached to every example.

  • In the chemical-development example, an online spectroscopic process signature tracked starting-material concentration over reaction time and distinguished reactions conducted at 76 °C and 65 °C. The result was specific to that reaction and development setting.
  • In laboratory-scale downstream-processing runs, at-line FT-MIR was used to monitor monoclonal-antibody titer, host-cell protein levels, and antibody aggregate amount. The measurement mode was at-line and the setting was laboratory-scale downstream processing.
  • In a yeast-fermentation study, the monitored analytes were glucose, ethanol, and biomass. The complex medium meant that multivariate methods, including PLS and MLR models, were often needed to evaluate the spectra.
  • In a continuous powder-blending study, multivariate analysis of an in-line NIR method was combined with residence-time-distribution models to describe process dynamics and support soft-sensor process-control tools. This is an NIR example, not direct evidence for a mid-infrared implementation in your stream.

These examples help you form a test plan. They do not transfer their calibration, accuracy, sampling arrangement, or operating limits to a different chemistry without application-specific work.

Before using an application as a precedent, classify what it actually demonstrates. A study may show that a signal tracks a changing condition, that a model predicts a named analyte within a stated range, or that a sampling arrangement produced usable spectra at a particular scale. Those are different forms of evidence. Record the matrix, analyte, measurement mode, output type, calibration or reference method, and operating conditions for each example. Then compare those fields with your own process rather than comparing instrument labels alone. A close match in measurement mode does not compensate for a different matrix or decision; a similar analyte does not establish transfer across concentration range or process conditions. This comparison also identifies the missing experiment: for example, a process may need representative online sampling rather than another literature example, or unseen-batch validation rather than a broader qualitative description. Where evidence is indirect, describe it as a reason to test a hypothesis, not as proof that the same output will be available in a new stream.

The downstream-processing study reported antibody aggregate quantification down to 1% w/w, host-cell protein monitoring down to 700 ng ml⁻¹, and antibody-titer determination with high accuracy for samples above 0.7 g L⁻¹. Those figures belong to that study’s samples, method, and laboratory-scale context.

Do not use them as acceptance limits for another process. Instead, define your own required range, decision threshold, sample matrix, reference method, and performance criterion before comparing a proposed model with published examples.

A stat-row infographic showing reported application figures: antibody aggregates at 1%, host-cell protein at 700 ng ml⁻¹, and antibody-titer samples above 0.7 g L⁻¹. The figures are presented as study-specific results.
Published performance figures illustrate evidence boundaries rather than transferable process limits.

Treat water and composition as first-order feasibility questions

Sample composition determines how much useful chemical information reaches the detector and how much interference the model must handle. A study measured pure water and aqueous 1.0 g/dL glucose solutions across 8000–950 cm⁻¹ at temperatures from 30 to 42 °C in 2 °C steps. It found that temperature effects were strongest near strong absorption bands.

The measurement path length in that study was 50 micrometres in the mid-infrared region and 0.4 mm in the near-infrared region. Path length is therefore an analytical variable, not a minor hardware detail. A path that works for one matrix or concentration range may not present the same signal balance in another.

We note that fermentation broth is mostly water, which absorbs strongly across large parts of the mid-infrared spectrum. For an aqueous stream, ask how water absorption, target concentration, co-components, temperature, and selected path length interact. The relevant question is whether the target information remains usable under the actual process conditions.

Prepare a composition brief before testing:

  • List the target analytes and their expected concentration ranges, including low and high conditions.
  • List major matrix components, solvents, water content, solids, gases, and likely impurities.
  • Identify temperature changes, dilution changes, phase changes, and composition shifts across the process.
  • State whether the output must distinguish a component, estimate its concentration, or follow a process trend.
  • Define which changes would make the sample outside the intended calibration domain.

This brief gives the feasibility test something concrete to challenge. It also makes clear whether a poor result reflects spectral interference, insufficient concentration, an unsuitable sampling arrangement, or a model that was trained on the wrong process domain.

Use the composition brief as a failure-diagnosis tool, not only as an inventory. If a target feature is obscured by strong absorption or overlap, the question is whether path length, matrix composition, or concentration range can be changed or bounded for a meaningful test. If spectra are clear but predicted concentrations disagree with reference results, investigate the reference method, calibration pairing, and model domain rather than treating the optical signal alone as sufficient. If a laboratory sample behaves well but the process-stream measurement does not, compare sampling location, flow presentation, temperature, pressure, solids, bubbles, and fouling exposure. If the signal follows a reaction but arrives too late for the intended action, the remaining issue may be measurement location or required response time rather than chemical detectability. Keep these diagnoses separate in the test record. A failed model and a failed sample presentation require different next steps, so combining them into one pass/fail label can hide the actual feasibility question. Water and temperature effects are especially important in aqueous work because the matrix can occupy much of the measured response; the test should show whether the target remains distinguishable within that response.

Specify the physical measurement arrangement

The physical measurement arrangement determines what material the spectrum represents.
Describe the proposed measurement point and configuration rather than evaluating the spectrometer in isolation. Document the optical path length, flow condition, temperature, pressure, solids content, bubbles, fouling exposure, cleaning approach, and required response time. These are fit questions for the process arrangement; generic instrument capability does not answer them.

The water and glucose study used a 50-micrometre mid-infrared path length and a 0.4 mm near-infrared path length. That contrast shows why path length should be stated with the spectral region and sample composition. It does not establish the correct path for your stream.

ICH Q13 identifies measurement interference, the effects of physical sampling on the material stream, and data gaps such as PAT recalibration or system-component failure as considerations when determining whether contingency methods are warranted. Include those events in the test plan rather than treating uninterrupted data as the only operating case.

Ask practical questions:

  • Does the sampling location represent the material relevant to the decision?
  • Can the process maintain the required flow and optical presentation?
  • What happens when solids, bubbles, fouling, pressure changes, or cleaning alter the measurement interface?
  • How will you recognize a missing, abnormal, or out-of-domain result?
  • What alternative information is available during recalibration or system failure?

A physical arrangement is ready for evaluation when its normal state, disturbance states, and data-gap response are explicit.

Test records should separate normal operation from disturbance and recovery conditions. For each condition, note the sampling location, process state, temperature, pressure, solids or bubbles, optical path, and whether a reference result is available. This makes it possible to ask whether an abnormal spectrum reflects the chemistry, the interface, or the instrument. It also prevents a data gap from being interpreted as a stable process value. When physical sampling is used for comparison, record whether removing material or otherwise disturbing the stream could alter the stream being evaluated; the cited process-monitoring guidance specifically treats sampling effects and data gaps as planning considerations. For a proposed inline arrangement, define how spectra are flagged as missing, abnormal, or outside the demonstrated domain, and what evidence permits return to use. This is not a claim that one contingency method fits every site. It is a way to make the arrangement testable before operational reliance. The record should link each flag to the process decision it could affect. That link keeps instrument behavior, sample representativeness, model validity, and decision governance from being treated as interchangeable evidence.

An infographic checklist for specifying a physical measurement arrangement. It covers sampling location, optical path, process presentation, contingency response, and recognition of missing or abnormal data.
A measurement arrangement is ready for testing when its normal state, disturbances, and data-gap response are explicit.

Build the model around the real process domain

A quantitative model is only as relevant as the chemistry and reference data used to build and test it. We require the calibration set to span the expected concentration range, including the extremes.

Those requirements translate into a process-domain question: do your calibration samples represent the material, operating conditions, variability, and concentration limits that the model will encounter? A collection of convenient laboratory samples may not cover changes in temperature, composition, batch, scale, or process stage. The model’s numerical output should be interpreted inside the domain demonstrated by its calibration and validation evidence.

Plan the reference set before collecting spectra. Identify the reference method, sample identity, sampling time, concentration range, operating condition, and batch or run. Preserve the pairing between each spectrum and its reference result so you can investigate disagreement rather than treating it as an unexplained instrument problem.

This is a practical boundary for feasibility work. If the reference result has material variability, the model cannot remove that limitation by producing more decimal places.

Evaluate the model against the reference method’s reproducibility and the process decision’s tolerance. Then test samples that were not used to fit the model. A reported fit to calibration data is not enough to show how the model behaves on new batches or process conditions.

Plan for model drift and measurement gaps

A PAT model is part of an ongoing measurement system, not a finished object that can be separated permanently from the process. Spectroscopic PAT applications often use prediction models whose accuracy can be affected by aging equipment, changes in active pharmaceutical ingredients or excipients, and previously unidentified process variations. PAT models need ongoing updating and management so they continue to reflect process changes and remain accurate.

Define the conditions that would trigger review. Examples include a material change, a process change, a sustained shift in residuals against the reference method, equipment maintenance, a change in sampling hardware, or a new operating range. The trigger should lead to a defined investigation, not an automatic assumption that the model remains valid or that it must be discarded.

ICH Q13 identifies PAT recalibration and system-component failure as data-gap considerations when determining whether contingency methods are warranted. Connect that principle to your operating procedure before implementation.

Document four lifecycle controls:

  1. Detection: Define how you identify drift, interference, abnormal spectra, and missing data.
  2. Decision: State who determines whether the result can support the intended process decision.
  3. Response: Specify the reference measurement, recalibration, investigation, or alternative method used during the event.
  4. Return to use: Define the evidence required before the model or measurement resumes its intended role.

The exact controls depend on your process and intended use. The feasibility question is whether the lifecycle can be governed as clearly as the initial calibration.

Define the intended decision and sampling strategy

The intended process decision should determine what you measure, where you measure it, how often you measure it, and what result triggers action.
Write the intended use in one sentence before selecting a sampling frequency. For example, you might need to detect a rapid disturbance, follow a process trend, compare reaction progression, or support an established quality decision. Each purpose implies different requirements for response time, data handling, location, and acceptance criteria.

The decision design should answer these questions:

  • Which variable or variables are monitored?
  • Where does the measurement represent the process decision?
  • How frequently must data be collected to observe the relevant change?
  • Which statistical treatment turns measurements into a decision rule?
  • What acceptance criteria distinguish an actionable change from normal variation?
  • Which physical sampling effects or measurement interference could distort the result?

ICH Q13 also identifies measurement interference, the effects of physical sampling, and data gaps such as PAT recalibration or system-component failure as considerations when determining whether contingency methods are warranted. Include those conditions in the decision design.

Do not describe a measurement as useful until you can state what its output changes. A high-frequency stream of spectra may support process understanding, while a concentration prediction may support a narrower decision only after its calibration, acceptance criteria, and failure handling are established.

Keep inline information complementary to laboratory evidence

Separate the roles in your evaluation:

  • Inline or online measurement: observes the process at the selected location and frequency.
  • Reference analysis: supplies the comparison result used for calibration, investigation, or validation.
  • Chemometric model: relates spectral information to a defined reference value when direct interpretation is insufficient.
  • Process decision: applies a stated rule to the result and records what happens when the result is unavailable or outside its demonstrated domain.

Keeping these roles separate helps you identify whether an unresolved issue belongs to sampling, spectroscopy, reference analysis, modeling, or decision governance.

Apply a bounded feasibility sequence

A bounded feasibility sequence turns a general interest in mid-infrared spectroscopy into a set of testable questions. Use the sequence below before treating a process-monitoring result as implementation evidence.

  1. Describe the stream and chemistry. Record phase, major components, target analytes, concentration range, temperature, pressure, solids, flow conditions, and known interference risks. We ask customers to describe their stream and chemistry so we can determine OrionIR® compatibility and configuration.
  2. Define the process decision. State whether you need a direct spectral trend, a predicted concentration, reaction progression, disturbance detection, or another defined output. Set the required response time and acceptance criteria.
  3. Choose representative samples and reference measurements. Pair samples with the reference method that will establish the model relationship or test the spectral interpretation.
  4. Test the proposed measurement arrangement. Examine the sampling location, path length, flow presentation, temperature, pressure, solids, bubbles, fouling exposure, and data continuity under relevant conditions.
  5. Check calibration coverage. Include process variability that the model is expected to encounter.
  6. Review performance against the decision requirement, not only the calibration fit.
  7. Define lifecycle handling. State how you will detect drift, manage recalibration, investigate interference, and respond to system-component failure.

Stop or narrow the proposed use when a material fit question remains unanswered. A bounded test should be able to produce a no-go conclusion as well as support further evaluation.

We offer to test OrionIR® on customer process samples when it makes sense. Treat that sample work as a compatibility and configuration evaluation tied to your stated chemistry and process question. Prepare the sample identity, operating condition, concentration range, reference result, and required output so the test addresses the uncertainty that matters.

Choose the next step from the remaining fit questions

Mid-infrared spectroscopy merits further evaluation when you can connect the spectrum, sampling arrangement, reference method, calibration domain, and response time to a defined process decision. If those links are not yet clear, your next task is to document the uncertainty rather than select a configuration by specification alone.

Prepare a concise technical brief containing:

  • The stream composition, phase, concentration range, and operating conditions.
  • The analyte or process change you need to observe.
  • The proposed sampling location and measurement arrangement.
  • The reference method and its reproducibility.
  • The required response time, acceptance criteria, and contingency path.
  • The evidence needed from representative samples and unseen batches.

A qualified professional should determine individual suitability, validation requirements, and any regulatory or safety decisions for your site.

Send the technical brief only after you can state the process question and the evidence needed to answer it. That gives the discussion a clear boundary and makes a compatibility evaluation more useful.



Next step

Have a question for Paeonia Innovations?

Get in touch