FT-NIR for Online Analysis in Polyol Production

FT-NIR for Online Analysis in Polyol Production — Expert Summary

Significance of the topic:

The hydroxyl value and related parameters (acid number, residual monomer such as ethylene oxide) are critical quality attributes for polyols and polyester resins because they govern polymer chain length, purity and final product performance. Rapid, reliable measurement of these parameters enables timely process control, reduces off-spec production, cuts analytical cost and exposure to hazardous reagents, and supports quality-management programs (e.g., Six Sigma, TQM). Fourier transform near-infrared (FT-NIR) spectroscopy offers a fast, reagent-free alternative to titration and GC methods and is well suited for at-line or near-line monitoring in polyol production.

Objectives and overview of the study/article:

• Demonstrate the capability of an Antaris FT-NIR analyzer to determine hydroxyl value, acid number and residual ethylene oxide (EtO) in a variety of polyol and polyester samples (liquids and translucent solids).
• Compare precision and accuracy of FT-NIR quantitative models with conventional reference methods (titration, headspace GC).
• Show practical sampling approaches, temperature management and chemometric strategies for robust calibrations applicable to at-line monitoring and quality control.

Methodology and used instrumentation:

• Instrument: Thermo Scientific Antaris FT-NIR Method Development Sampling (MDS) system.
• Sampling modules: three-position heated transmission holder for 7 mm disposable vials (liquids) and Tablet Transmission Module for translucent solid resins.
• Sample handling: disposable vials to avoid cleaning, external pre-heating for samples with particulates, temperature control during acquisition (RESULT software) to ±0.1 °C, temperature equilibration delays.
• Spectral acquisition: typical ranges 4000–10,000 cm⁻¹ and 4000–12,000 cm⁻¹; resolution 8 cm⁻¹; 50–100 co‑averaged scans.
• Software and data processing: RESULT for acquisition and temperature control; TQ Analyst for quantitative modelling. Preprocessing included Norris or second derivative, multiplicative scatter correction. Quantification algorithms: Partial Least Squares (PLS) and Stepwise Multiple Linear Regression (SMLR).

Main results and discussion:

The paper presents calibrations developed for three independent sample sets (surfactants, liquid polyols including EtO content, and polyester resins in different production stages). Key performance highlights are:

• Sample Set 1 (surfactants): Hydroxyl value SMLR model using two wavelengths (primary at ~6854 cm⁻¹) produced R = 0.9999, RMSEC = 1.89 mg KOH/g, RMSECV = 2.59 mg KOH/g over a reference range 70–350 mg KOH/g; within-sample repeatability (%RSD, n=5) = 0.55%. Acid number (3‑factor PLS after outlier removal) yielded R = 0.9980, RMSEC = 0.129 mg KOH/g, RMSECV = 0.186 mg KOH/g for 0.8–6.0 mg KOH/g; repeatability %RSD ≈ 1.65%.
• Sample Set 2 (liquid polyols, EtO): Hydroxyl SMLR (three wavelengths) R = 0.9994, RMSEC = 0.780 mg KOH/g, RMSECV = 0.951 mg KOH/g over 23.4–116.0 mg KOH/g. EtO SMLR (three wavelengths) R = 0.9999, RMSEC = 0.400 (units in ppt), RMSECV = 0.460 for EtO range 1.78–100 ppt. Parameter intercorrelation (hydroxyl vs acid) was moderate (r ≈ 0.731) but low enough in this set to allow independent calibration.
• Sample Set 3 (polyester resins, translucent solids): Transmission measurements on molded/hardened samples with pathlength variability compensated by using a denominator wavelength. Hydroxyl calibration (single-term SMLR with denominator) gave R = 0.9995, RMSEC = 0.552 mg KOH/g, RMSECV = 1.99 mg KOH/g over 7.1–62.6 mg KOH/g. Acid number (two-term SMLR with denominator) produced R = 0.9950, RMSEC = 1.31 mg KOH/g, RMSECV = 1.73 mg KOH/g over 8.6–47.8 mg KOH/g. Overall correlation between hydroxyl and acid values depended strongly on early-process samples (global r ≈ -0.709; removing the initial pre-acid charge sample changed behaviour significantly).

Key observations:

• Second-derivative preprocessing and scatter correction were effective to isolate relevant overtone/combination bands (notably O–H overtones) and to compensate matrix/pathlength effects.
• SMLR with judicious selection of primary and compensatory (matrix or denominator) wavelengths produced simple, interpretable models and helped correct for pathlength and matrix variability in at-line conditions.
• FT-NIR proved sensitive enough to quantify low-content residual EtO and acid values with GC or titration-comparable performance for the studied ranges.

Benefits and practical applications of the method:

• Rapid results enabling near-line or at-line process control and faster reaction end-point detection compared to titration workflows.
• Elimination or reduction of hazardous reagents and waste associated with titrations; lower operator exposure and chemical costs.
• Capacity to measure multiple properties from a single spectral acquisition (hydroxyl value, acid number, residual monomer), simplifying QC workflows and reducing sample handling.
• Reduced operator-to-operator variability and automated reporting via RESULT software; improved production efficiency and integration with quality programs.

Practical recommendations from the study:

• Maintain strict temperature control during acquisition (±0.1 °C) and consider preheating viscous or particulate-containing samples to improve spectral reproducibility.
• Use disposable vials for liquids to increase throughput and minimize cleaning; for translucent solids prefer transmission sampling (tablet/transmission module) and apply denominator wavelengths to correct pathlength variability.
• Test for intercorrelation among reference parameters before model building; if strong intercorrelation exists consider separating calibration domains or including additional compensatory wavelengths.
• Employ cross-validation and outlier analysis during calibration development to ensure robust models suitable for at-line deployment.

Future trends and opportunities for use:

• Greater adoption of fully inline FT‑NIR sensors integrated with process analytical technology (PAT) and advanced process control for closed-loop reaction control.
• Improved chemometric approaches including transfer learning, robust model-update strategies, and machine-learning methods to handle broader raw-material variability and long-term drift.
• Miniaturization and cost reduction of FT-NIR hardware to enable more widespread at-line deployment across production lines.
• Cloud-enabled analytics and centralized model management for multi-site standardization and rapid model maintenance.
• Combining FT-NIR with complementary sensors (Raman, mid-IR, online GC) for multivariate monitoring of complex polymerization processes.

Conclusions:

FT‑NIR using the Antaris MDS platform provided accurate, precise and rapid quantification of hydroxyl value, acid number and residual ethylene oxide across multiple polyol and polyester sample types. Well‑constructed SMLR and PLS models with appropriate preprocessing and temperature/pathlength control achieved calibration statistics comparable to reference titration and GC methods while enabling at-line implementation, reduced hazardous reagent use and faster process feedback. These characteristics make FT‑NIR a viable and often preferable alternative for routine QC and process monitoring in polyol production.

References:

The summary is based on the application note (Thermo Fisher Scientific, Application Note 51594, 2008) reporting Antaris FT-NIR method development and calibrations for hydroxyl value, acid number and ethylene oxide in polyol and polyester matrices.