The operator panel shows a concentration value that is missing, noisy, drifting, or slow to follow the process. Start by proving that near-infrared spectroscopy can distinguish peracetic acid from the rest of the reaction mixture through the proposed optical path. Then select the analyzer, sampling arrangement, calibration model, and system integrator as one measurement system; buying an online NIR instrument alone does not solve the application.
Read the symptoms before selecting equipment
Separate an analyzer problem from an application-design problem. A stable spectrum with an incorrect concentration points toward calibration or reference-method bias. A changing spectrum with erratic concentration points toward bubbles, temperature, fouling, optical geometry, or an unmodeled process constituent.
| Observed symptom | Check first | Likely cause class |
|---|---|---|
| No concentration result | Analyzer status and spectral signal level | Blocked optical path, failed acquisition, invalid model range, or communications fault |
| Value is stable but biased | Paired laboratory result and sample timing | Reference-method bias, sampling delay, or calibration offset |
| Value is noisy | Raw spectra and process condition | Bubbles, solids, vibration, unstable presentation, or weak optical signal |
| Value drifts over time | Optical baseline and temperature | Window fouling, temperature change, aging reference data, or process composition shift |
| Analyzer misses real process changes | Calibration range and spectral sensitivity | Model lacks representative samples or other constituents mask the useful response |
| Laboratory and online values disagree intermittently | Sample location, collection time, and reaction progress | The two methods measured different material states |
Do not begin by changing display scaling, filtering the output, or replacing communications hardware when the raw spectrum is unstable. That is not the fault. Fix sample presentation and optical signal quality first.
Prove the measurement mechanism
NIR does not directly count peracetic acid molecules. It measures wavelength-dependent absorption from molecular bonds, then a multivariate calibration converts the spectrum into concentration. The reaction mixture may contain water, reactants, products, and temperature-dependent spectral changes. Their absorption bands can overlap, so a single-wavelength assumption is weak unless testing demonstrates selectivity.
Glass reactors add an optical-path decision. The analyzer might view through the reactor wall, use a probe, or measure a bypass sample. Ordinary process glass is not automatically a qualified NIR window. Its composition, thickness, curvature, coatings, surface condition, and wavelength transmission can reduce signal or introduce a repeatable spectral contribution.
Test the exact proposed geometry. Include the same glass type and thickness, process temperature range, expected concentration range, background composition, mixing condition, and bubble load. If the spectra cannot separate peracetic acid changes from those disturbances, relocate the measurement to a suitable probe or engineered sample interface.
Define the application before requesting quotations
Give suppliers a written measurement specification. Missing process data produces optimistic proposals that cannot be compared.
- State the required concentration range and reporting units.
- List every known constituent and the expected variation of each constituent.
- Describe the glass reactor and the proposed optical access point.
- Identify whether measurement must be through the wall, in situ, or in a bypass loop.
- State process temperature, pressure, phase, solids, bubbles, color changes, and mixing behavior.
- Define the required update behavior, accuracy, repeatability, and maximum acceptable delay without inventing values before the process need is known.
- Describe the laboratory reference method used to generate calibration and validation values.
- Specify hazardous-area, wetted-material, cleaning, containment, and communications requirements from the site design basis.
- Ask who owns model development, model transfer, recalibration, validation, and long-term support.
Potential contacts named for online infrared analysis include ABB Process Analytics, Horiba, and Siemens. SPECAC, based in Kent, was identified specifically for online NIR. Treat these as candidates to qualify, not proof that a standard product is suitable for peracetic acid or for measurement through the reactor glass. A local analyzer system integrator can combine the spectrometer, optical interface, sample handling, controls, and validation work.
Run a representative feasibility and calibration program
- Define the reference: Choose the laboratory method that will supply peracetic acid concentration. Control collection, preservation, transport, and analysis so reaction progress after sampling does not create a false analyzer error.
- Collect representative material: Cover normal operation, startup, shutdown, expected raw-material variation, temperature variation, and credible off-spec conditions. Include interfering constituents that vary independently of peracetic acid.
- Use the production optical path: Acquire spectra through the actual glass arrangement or the proposed probe and sample interface. A model developed in a clear laboratory vial may fail when transferred to curved reactor glass.
- Pair spectra and references: Time-align each spectrum with the corresponding physical sample. Account for transport delay if a bypass loop or remote sample point is used.
- Build the model: Train the chemometric model only within the tested concentration and composition space. Retain independent samples for validation instead of using every result for calibration.
- Challenge disturbances: Test bubbles, fouling, temperature changes, glass variation, mixing changes, and background-composition shifts separately where practical.
- Set validity handling: Configure the system to flag spectra outside the model domain. An out-of-range sample must not appear as a trustworthy concentration merely because the model returns a number.
Skipping independent validation is the most expensive shortcut. A calibration can fit its training samples closely and still fail on a new batch.
Verify the complete online system
Commission the measurement from optics to the control-system value. Do not accept it from a single successful comparison.
- Confirm stable raw spectra under steady process conditions.
- Compare repeated spectra to distinguish analyzer noise from real process variation.
- Collect paired online and laboratory results across the operating range.
- Plot error against concentration, temperature, batch stage, and major constituent changes.
- Check that cleaning restores the optical baseline after deliberate or naturally occurring fouling.
- Verify time stamps, engineering units, signal scaling, quality status, and stale-data handling at the operator panel.
- Force analyzer and communications faults and confirm that the control system rejects or flags the bad value.
- Document the accepted model version, reference dataset, optical configuration, and validation results.
If the NIR value will drive automatic control, qualify it first as a monitored measurement. Enable closed-loop use only after its failure states, delay, and valid operating domain have been demonstrated.
Avoid recurring application failures
- Do not treat IR and NIR as interchangeable. Confirm the proposed spectral region and why it provides usable selectivity for the mixture.
- Do not qualify only the spectrometer. Glass, probes, fibers, flow cells, cleaning hardware, and sample transport all affect the result.
- Do not hide unstable spectra with output filtering. Filtering can make the panel look calm while adding delay and masking bubbles or fouling.
- Do not extrapolate silently. Flag concentrations and compositions outside the calibration domain.
- Do not calibrate against poorly timed samples. A reactive sample can change between the reactor, collection point, and laboratory result.
- Do not accept a black-box model. Define responsibility for model maintenance and the triggers for revalidation.
FAQ
What happens if NIR is measured through the glass reactor wall?
The glass becomes part of the optical system. Test its transmission, thickness, curvature, coatings, and surface condition in the intended wavelength range before committing to through-wall measurement.
What happens if water and other constituents absorb in the same NIR region?
The calibration must separate overlapping spectral responses using representative samples in which those constituents vary. If their effects cannot be separated, change the optical region, measurement geometry, or analytical method.
What happens if the online value is stable but disagrees with the laboratory?
Check sample timing, sample location, units, and the laboratory reference method first. Then test for calibration bias using independent paired samples across the concentration range.
What happens if bubbles make the NIR reading noisy?
Inspect the raw spectra and correct the sample presentation, mixing condition, probe position, or bypass design. Output filtering alone masks the disturbance and adds measurement delay.
What happens if no supplier can demonstrate peracetic acid selectivity?
Stop procurement and request an application feasibility study using the real mixture and optical path. Escalate to the analyzer manufacturer's official application-support channel when spectra remain unstable, the model cannot distinguish process constituents, or material compatibility and containment cannot be resolved; involve a qualified analyzer system integrator when the sample interface requires custom engineering.