Curing reaction kinetics of paper-based phenolic resin laminates—from laboratory measurements to inline quality control

Abstract

We describe a comprehensive analysis of the drying and curing kinetics of resol phenol–formaldehyde (PF) resin utilizing multiple different thermophysical and optical measurement techniques and combinations thereof to gain a comprehensive understanding of the physicochemical processes that take place during large-scale production of paper-based PF laminates. This included thermogravimetric analysis (TGA), differential scanning calorimetry (DSC), Fourier-transform infrared evolved gas analysis (FTIR-EGA), and near-infrared (NIR) spectroscopy. In particular, the tailored integration of TGA with simultaneous near-infrared (NIR) spectroscopy in reflection geometry facilitated the evaluation of NIR spectroscopy’s suitability for monitoring the ongoing process. This led to the implementation of an NIR-based real-time measurement setup at an industrial production site for a feasibility assessment. NIR spectroscopy in combination with partial least squares (PLS) regression modeling showed highly promising results highlighting the advantages of NIR spectroscopy as a tool for real-time inline quality control for large-scale production of PF resin laminates.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1007/s00216-025-05983-0.

Introduction

Resins are a diverse class of solid or highly viscous substances, either from natural or synthetic origin, that play a crucial role in a wide range of industrial applications due to their adhesive properties, chemical resistance, and structural versatility. Many resins can be converted into polymers through curing or polymerization processes, which significantly changes their mechanical properties [1, 2]. Amongst the most important and wide-spread types of synthetically produced resins are phenol–formaldehyde (PF) resins [3–5].

Phenol and formaldehyde can produce two types of PF resins depending on the reaction conditions. Under acidic conditions with less than an equimolar amount of formaldehyde relative to phenol, novolak is formed, resulting in a linear resin [6]. In contrast, under basic conditions with an excess of formaldehyde, resol is produced, creating a branched polymer network [6]. This study focuses on the resol composition, which undergoes chemical processes to transform the educts phenol and formaldehyde from a liquid or soluble state into a rigid, cross-linked, and insoluble thermoset. The process begins with the deprotonation of phenol by a basic catalyst, such as sodium hydroxide, potassium hydroxide, or ammonia. The deprotonated phenol then reacts with formaldehyde in an addition reaction, forming ortho- or para-methylolphenol [7]. Depending on the formaldehyde-to-phenol ratio, mono-, di-, or trimethylolated phenols may be produced as transient intermediates. The polycondensation of methylolated phenols initiates the cross-linking reaction. At higher temperatures, mono- and polymethylolated phenols condense to form methylene bridges (-CH₂-) by bonding at a reactive position with another phenolic compound. Less commonly ether bridges (-CH₂-O-CH₂-) are formed in a condensation reaction with hydroxymethyl groups. Both reactions result in the elimination of water. If enough heat is supplied to the resin, it undergoes further cross-linking and forms a semi-rigid, non-liquid network.

One of the most important applications of resol PF resins is the production of high-pressure laminates (HPLs) which can be found in many different industrial products [8], most often in the form of PF laminates [6, 9], which are produced via impregnation of one or multiple layers of a base material such as paper, fiberglass, or cotton with PF resin. The impregnated base material is then exposed to high temperatures to induce the drying and curing process described above. The result is a semi-finished product, which can be manipulated before undergoing high-temperature pressing to finalize the curing process, solidifying the rigid, fully cross-linked thermoset polymer matrix [10]. Depending on the choice of base material, number of layers as well as processing conditions, the resulting laminate can be tuned regarding its mechanical properties to fit the intended application [6, 11]. To produce a semi-finished product that will result in a HPL with the desired properties, it is important to precisely control the drying and curing process. Due to the large amounts of heat necessary for large-scale production of these laminates, the process offers considerable potential for optimization, particularly in terms of energy efficiency. Furthermore, a more consistent product quality and significant reduction of out-of-spec products can be achieved through real-time process monitoring.

In this work, multiple complementary laboratory measurement techniques are utilized to unveil the polymerization mechanics of the investigated PF resin during the production of paper-based PF resin laminates. Each individual measurement technique provides different insights into the underlying process. The weight loss is characterized through thermogravimetric analysis (TGA) [12], which was equipped with a custom near-infrared (NIR) reflection probe to simultaneously gather in situ chemical information. Additionally, the process was monitored using differential scanning calorimetry (DSC), which reveals the amount of heat required to increase the sample temperature. Thus, information on its heat capacity is provided, as well as the amount of consumed and released heat attributed to solvent evaporation as well as endo- and exothermic chemical reactions [13]. Furthermore, the volatile gas expelled during the curing process was monitored using Fourier-transform infrared evolved gas analysis (FTIR-EGA) [14]. Combining the insights gained from all these analytical methods reveals details on the complex, multi-step drying and polymerization kinetics taking place during heating of the uncured PF rein. After analyzing the curing reaction in the laboratory setting, it was concluded that spectroscopy in the NIR spectral region can be a suitable choice for inline quality control of paper-based high-pressure laminates (HPLs) based on the investigated PF resin. This led to the installation of an inline measurement setup based on NIR spectroscopy in reflection measurement geometry combined with partial least squares (PLS) regression modeling for real-time inline quality control.

Materials and methods

Paper-based phenolic laminates

The liquid phenol–formaldehyde resin solutions investigated in this study were prepared using standard cooking procedures at the supplier’s production facility. This large-scale cooking process of resol resins involves several critical steps to ensure efficient production and consistent quality. Initially, the raw materials, including phenol, formaldehyde, and a basic catalyst such as sodium hydroxide, are prepared based on the desired resin specifications. In a large reaction vessel equipped with stirring and temperature control, phenol and formaldehyde are combined, and the catalyst solution is introduced to initiate resin formation. The mixture is heated to approximately 60–90 °C for 1–2 h to form a prepolymer. In a subsequent vacuum distillation, a large portion of water is removed.

After the distillation process, the resin is cooled and transported to storage tanks from where it is transferred to the impregnation lines. A schematic drawing of an impregnation line with the subsequent industrial dryer where the resin is dried and cured is shown in Fig. 1. Before impregnation, solvents, dyes, and additives are introduced to the resin to modify its properties for specific applications. The addition of solvents such as methanol helps to adjust the viscosity of the resin in a range of 60–90 m Pa s to achieve sufficient impregnability which can be additionally tuned via pre-heating of the paper via heating rolls. Different dyes can be incorporated to obtain a specific color for aesthetic purposes. The impregnation lines are also equipped with metering systems to ensure accurate dosing. Afterwards, the freshly impregnated paper is introduced into the approximately 30-m-long floatation dryer, where it is exposed to elevated temperatures up to 250 °C. This results in drying and curing of the resin, before the semi-finished product is cooled down to approximately 30 °C using cooling rolls. The NIR measurement for quality control, which is described in the “Inline NIR measurements” section, was installed after these cooling rolls, for continuous measurement of the semi-finished product. For routine quality control, the dried, cured and cooled semi-finished product is sporadically measured manually regarding its volatile content and resin flow (see also the “Determination of volatile content” and “Determination of resin flow” sections, respectively), which must be held in a certain range.

Fig. 1.

Simplified schematic drawing of the paper impregnation line followed by the flotation dryer and pellet changing system. The position of the NIR measurement head installed for real-time inline quality control after the cooling rolls is also indicated (see the “Inline NIR measurements” section for further details)

Laboratory investigation

Preparation of laboratory samples

In this study a liquid phenol–formaldehyde resin solution was prepared using standard cooking procedures at the supplier’s facility as described above. After the cooking and mixing process (approximately 10% methanol and black dye for color), the resin solution was allowed to cool and was stored at 6 °C and measured within 72 h. In the laboratory, where the DSC, NIR/TGA, and FTIR-EGA measurements (for details see respective sections below) were performed, a manual impregnation method was used to apply the resin solution onto kraft paper sheets supplied by the manufacturer. In the manual resin application process, the raw paper is clamped and passed through a resin bath; the resin application is controlled by the applied pressure on the rollers. Key parameters such as the dwell time in the resin bath and the manual pressure on the rollers can be adjusted to influence the resin uptake. All parameters were chosen to most closely resemble the impregnation results at the manufacturer’s facility. Immediately after impregnation, small discs were carefully punched from the large sheets for the various analyses: approximately 10 mg for DSC, 35 mg for NIR/TGA, and 224 mg for FTIR-EGA measurements.

Thermogravimetry (TGA) combined with near-infrared (NIR) spectroscopy

Thermogravimetric analysis (TGA) with near-infrared spectroscopy (NIR) was performed using an STA F449 F5 Jupiter simultaneous thermal analysis (STA) system (Netzsch, Germany). The samples were placed on a 10-mm-diameter Al₂O₃ plate. To enable simultaneous monitoring of the curing process, an NIR reflection probe was inserted through an opening at the top of the oven, which was equipped with a gas-tight seal to maintain an inert atmosphere. Prior to the start of the measurement, the NIR reflection probe was positioned at the optimal distance directly above the sample to maximize the collected NIR signal. A baseline measurement was performed and subtracted from the sample data to correct for any instrument background. The schematic drawing of the setup and a photograph of the NIR probe holder are shown in Fig. 2. The measurements were conducted in a nitrogen atmosphere achieved with a constant flow rate of 50 ml/min. The sample temperature was increased from room temperature up to 250 °C with a heating rate of 5 K/min.

Fig. 2.

Simplified schematic drawing of the custom TGA measurement setup with simultaneous NIR spectroscopy. The photograph on the bottom right shows the custom holder of the NIR probe equipped with a high-precision z-stage to optimize the NIR measurement distance

A compact high-throughput fiber-coupled NIR spectrometer (Ibsen Photonics, Denmark) with a transmission grating and a thermoelectrically cooled InGaAs detector array consisting of 256 pixels was used to obtain spectral data in the range from 1100 to 2100 nm. The NIR reflection probe (Avantes, Netherlands) featured six illumination fibers and one signal fiber enclosed in a 2.5-mm-thick and 200-mm-long stainless steel tip. A fiber-coupled halogen light source (Avantes, Netherlands) connected to the illumination end of the NIR reflection probe, while the signal fiber was connected to the NIR spectrometer to acquire high-quality NIR spectral data.

Initially, a dark spectrum was recorded to compensate for detector noise, with the detector thermoelectrically cooled to a temperature of −10 °C throughout the measurement to reduce noise. A background measurement was then performed on the sample holder (Al₂O₃) to calculate the absorbance spectrum for the sample. The resin-impregnated paper samples were placed flat on the sample holder, and reflection spectra from its surface were continuously recorded. For every measurement, 20 spectra with an integration time of 5 ms were averaged to reduce noise, and one averaged spectrum was saved each second. The measurements were controlled and recorded using JETI Versaspec V5.4.2 software (Ibsen Photonics, Denmark).

The recorded NIR spectra were mathematically corrected to reduce unwanted effects such as scattering and intensity fluctuations and improve signal clarity. Due to the different measurement setups used in the laboratory and for the inline measurements at the production site (see also the “Inline NIR measurements” section), different processing methods for the acquired spectral data were applied to achieve the best results. Multiple spectroscopic data processing methods and combinations thereof were evaluated, including baseline fitting (first- and second-order polynomial), Savitzky-Golay filtering [15] with different window sizes from 5 to 25 points and either first- or second-order polynomial, standard normal variate (SNV) transformation [16] and multiplicative scatter correction (MSC) [17]. For the laboratory measurement data, a Savitzky-Golay filter with a window size of 15 and a second-order polynomial in combination with a first-order baseline fit was identified as the best suited combination to remove unwanted distortions of the spectral data, while still conserving important characteristic changes in the NIR spectra recorded at different points in the monitored process. All processing of the NIR spectral data was done using the built-in functions in PLS Toolbox 8.0.2 (Eigenvector Research, USA).

Differential scanning calorimetry (DSC)

High-precision calorimetric data were obtained by Differential Scanning Calorimetry (DSC) measurements, using a DSC 204 F1 Phoenix instrument (Netzsch, Germany) equipped with an automated sample handling system. This configuration ensures better reproducibility and resolution compared to the DSC functionality of the STA system. The DSC instrument was calibrated using standard reference materials—namely indium, tin, bismuth, and zinc—to ensure accurate temperature and enthalpy measurements. Calibration was performed under identical conditions to those used for sample measurements (i.e., with the same gas atmosphere, heating rate, and crucible configuration), which minimizes potential systematic errors. The calibration procedure involved recording the melting transitions of these pure substances and comparing the observed melting temperatures and associated enthalpies to their known reference values. A polynomial fit was applied to generate a sensitivity curve, and baseline drift (including effects from crucible asymmetry) was corrected. This quantitative calibration ensured that heat flow measurements (in J/g) were reliable for subsequent sample analysis.

The samples were sealed in 25 μl aluminum crucibles equipped with lids featuring a small vent hole to allow for the continuous evaporation of volatiles. This design mimics the open system used for FTIR-EGA and TGA/NIR measurements conducted on an Al₂O₃ plate where evaporation and condensation reactions are measured. Crucible and lid were welded together before the measurement. Measurements were performed under a nitrogen atmosphere with a flow rate of 20 ml/min to maintain an inert environment. The temperature was increased from room temperature to 250°C at a heating rate of 5 K/min. The measurement was conducted in duplicate to ensure reproducibility.

DSC and TGA data matching

The data acquired with the TGA setup with integrated NIR reflection probe and the DSC data were both used to gain a more detailed insight into the curing process. The TGA/NIR system allows for simultaneous acquisition of TGA and NIR data, providing spectroscopic information that is directly matched to the temperature profile of the controlled heating process. In contrast, the DSC measurement was used to obtain calorimetric data via measurement of sample heat flow. By matching the temperatures of the TGA and DSC measurements using similarly prepared samples and the same heating rate, the acquired NIR spectra can be directly correlated to the acquired DSC data. This correspondence could not be directly obtained from a fully simultaneous measurement, as the covered aluminum crucibles used for DSC prevent the NIR reflection probe from detecting the sample surface.

The temperature onset (T_on) and offset (T_off) for the exothermic condensation reaction were determined from the TGA data using a tangent intersection method [18]. In this approach, tangents are drawn on the TG curve immediately before and after each transition of the exothermic event where the DSC curve can be approached by linear fit to the data. The intersection points of these tangents then define the respective threshold temperature T_on and T_off. These temperatures are then used to further analyze the DSC data. Notably, the TGA data are used for this purpose because the DSC measurements capture both the endothermic evaporation and the exothermic condensation processes simultaneously, resulting in overlapping peaks that make it difficult to directly determine the precise start and end of the exothermic reaction.

For the DSC measurements, the exothermic condensation reaction region was isolated in the corresponding temperature range obtained from TGA analysis. A Bézier curve was employed to define a baseline under the exothermic peak. The degree of conversion, α_DSC(T), was calculated by integrating the DSC heat flow over the interval [T_on, T_off] and normalizing it by the total heat released over the reaction:

$$\begin{array}{c}{\alpha }_{\mathit{DSC}}\left(T\right)=\frac{{\int }_{T_{\mathit{ON}}}^T\dot{Q}\left(T\right)dT}{{\int }_{T_{\mathit{ON}}}^{\mathit{OFF}}\dot{Q}\left(T\right)dT}\end{array}$$ 1

where $\dot{Q}\left(T\right)$ denotes the heat flow as a function of temperature. This produces a dimensionless conversion curve ranging from 0 to 1.

Fourier-transform infrared evolved gas analysis (FTIR-EGA)

Fourier-transform infrared evolved gas analysis (FTIR-EGA) measurements were carried out using an INVENIO-S FTIR (Bruker, USA) equipped with an external coupling module, which encompasses a liquid nitrogen cooled MCT detector. This external coupling module connects the FTIR to an STA F449 F1 Jupiter (Netzsch, Germany). Approximately 224 mg of sample was placed onto a Al₂O₃ crucible plate with a diameter of 17 mm and mass of about 1.43 g. Subsequently the sample was heated at a rate of 5 K/min from approximately 30 °C to 250 °C under a synthetic air atmosphere composed of 80% nitrogen and 20% oxygen at a flow rate of 100 ml/min. The mid-infrared (MIR) absorption spectrum of the gases emitted from the sample during heating are measured using the FTIR spectrometer in combination with the Bruker OPUS software over a wavelength range from 4500 cm⁻¹ to 650 cm⁻¹, covering the MIR-fingerprint region to allow for precise discrimination of different gases. Each spectrum was measured with a spectral resolution of 4 cm⁻¹ and averaged 32 times, which led to a measurement time of 14 s per MIR spectrum. For calculation of the absorbance spectrum, a background spectrum was recorded prior to sample heating to capture the ambient atmospheric contribution and automatically subtracted from subsequent measurements. In addition, the instrument’s response was calibrated by referencing the water vapor absorption band in ambient air, which serves as an internal standard to maintain consistent baseline correction and sensitivity.

The acquired MIR absorption spectra were then analyzed using a classical least squares (CLS) algorithm in the PLS Toolbox (Eigenvector Research, USA) [19] with pure gas spectra taken from the NIST database, applying non-negativity constraints for both absorbance and gas concentration. The CLS analysis allows to estimate the amount of the individual gases evaporated from the sample as a function of temperature.

Industrial implementation

Determination of volatile content

The determination of volatile components (such as residual solvents and water) used as reference for the spectroscopic analysis involves measuring the weight difference of a sample before and after drying under specified conditions. For that purpose, samples are taken from the product and are tested within 5 min of sampling without any special preconditioning. Circular discs with an area of 100 cm² are punched close to the left and right edges as well as from the center of the approximately 1–2 m broad sheet (depending on the produced format) and weighed with a precision of ± 0.5 mg. These three samples are then dried in a preheated drying oven at 165 °C for 5 min. The volatile content is reported as loss on drying (LOD) in percent for the three samples, which usually differ slightly due to slightly varying degree of paper impregnation over the width of the sheet.

Determination of resin flow

The procedure for determining resin flow involves measuring the percentage of resin that exits the semi-finished product under elevated pressure and temperature. A total of five circular discs with an area of 100 cm² are punched from the test material. The total weight of the five discs is determined before they are pressed between plates at 785 N/cm² (78.5 bar) and 150 °C for a duration of 5 min. After the pressing, laterally extruded resin residuals are removed, and the five samples are reweighed together to calculate the amount of extruded material in percent of the initial weight. Additionally, the breakage of the pressed sample is visually assessed for quality.

Inline NIR measurements

For the NIR measurements conducted inline at the industrial production site, the same high-throughput NIR spectrometer as in the laboratory (see the “Thermogravimetry (TGA) combined with near-infrared (NIR) spectroscopy” section) was used. However, to enable an increased measurement distance from the sample, necessary for both safety reasons and to mitigate influences on the recorded spectra due to slight movement of the sample, a NIR reflection measurement head (tec5, Germany) was used. The measurement head was installed after the floatation dryer and the subsequent cooling rolls (Fig. 1). The measurement head incorporates a 20 W halogen lamp and has a built-in white reference, that can be automatically put in place for periodic reference measurements. This measurement head allows to measure the semi-finished product after the dryer at a measurement distance of approximately 20 cm. The optical signal reflected from the sample was guided to the NIR spectrometer with a 10 m long, low-OH multi-mode optic fiber with a core diameter of 600 µm and a numerical aperture of 0.22 (Thorlabs, USA). As for the NIR spectra taken in the laboratory setup (see also the “Thermogravimetry (TGA) combined with near-infrared (NIR) spectroscopy” section), different combinations of spectroscopic data processing methods were investigated for the data obtained inline at the production site. The optimal processing of the NIR spectral data was identified by minimizing the root mean square error of cross validation (RMSECV) of the subsequently trained partial least squares (PLS) regression model [20] incorporating the reference values for volatile content and resin flow (see the “Determination of volatile content” and “Determination of resin flow” sections, respectively). In addition to the methods already taken into consideration for the laboratory measurements, first- and second-order derivatives of the NIR spectra were tested to compensate signal fluctuations caused by the production process. The best results for the inline NIR spectra were achieved by applying a Savitzky-Golay filter, with an increased window size of 25 (compared to 15 in the for the laboratory NIR spectra) and subsequent standard normal variate (SNV) transformation. Since the application of a first-order derivative on the NIR spectra led to a slight increase of the resulting RMSECV value and a second-order derivative increased it even further, this preprocessing method was omitted. All spectral data processing as well as the subsequent PLS regression modeling was carried out in PLS Toolbox 8.0.2 (Eigenvector Research, USA). Further details on the PLS regression modeling process as well as the achieved results are given below in the “Inline NIR measurements” section.

Results and discussion

Laboratory measurements

In this section, the different phases of the physicochemical curing reaction are analyzed separately. Three different reaction phases, which take place in different temperature ranges, can be identified. These phases are (i) free solvent evaporation, (ii) polymerization, and (iii) final drying and curing phase. Although the phases show some temporal overlap in the temperature range from 30 °C to 250 °C, they can be separated by analyzing the curve of weight loss over sample temperature as determined by TGA measurements shown in Fig. 3.

Fig. 3.

a TGA-measurement data. The top graph shows the mass loss normalized to the total weight loss in the temperature range from 30 °C to 250 °C for two different samples. The two graphs on the bottom show a closer view on the borders between the different process phases. The tangents used to determine the border temperatures are shown as dotted lines in the same color. b DSC data. The top graph shows the power required to heat the sample versus temperature. The bottom graph shows a closer look at the polymerization phase, where also the determined baseline is shown as a dotted line and the resin curing degree is plotted as dashed line in the respective color

In the top graph in Fig. 3a, the sample weight loss is shown as a function of temperature. The weight loss has been normalized to the total weight loss that happens over the whole temperature range to compensate for slightly different sample sizes and weights. In the two shown curves, acquired from the measurements of two individual samples, two distinct changes of the slope can be identified as indicated by the two dashed, vertical lines. The bottom graphs provide a detailed view of the specific segment of the temperature curves in the vicinity of these two characteristic temperatures. The bottom graphs show that both curves have their maximum slope change at very similar temperatures. The mean value of these determined temperatures was chosen as phase transition temperatures, with their standard deviation serving as a means for error estimation. This results in T_on = 95 ± 1 °C and T_off = 147 ± 1 °C, as temperature borders between the process phases. These temperatures are also used as the onset and offset temperatures for the exothermal condensation reaction of the resin used in the evaluation of the DSC data shown in the top graph in Fig. 3b. The DSC curve shows the power required to heat the sample versus the sample temperature and resembles previously reported DSC curves for phenol-resin-impregnated kraft paper [21]. The dotted line shown in the same graph shows the Bézier curve that indicates the baseline during the exothermic polymerization reaction. The bottom graph gives a closer look at the temperature region between T_on and T_off. Therein it can be seen that at these temperatures, the DSC curve starts to significantly deviate from the baseline. The dotted line in the same graph shows the integrated area between the baseline and the DSC curve, normalized to the cumulated heat flow calculated using Eq. (1). This value gives the curing degree of the resin sample and shows similar behavior for both samples, that are in good agreement with previously reported curves for similar resin mixtures [22, 23]. The enthalpy $\mathrm{\Delta }H$ normalized to the sample weight released by the exothermal reaction during the polymerization process is given by the area between the dotted baseline and measured DSC curve. As indicated in the figure, $\mathrm{\Delta }H$ amounts to approximately −60 J/g for both samples.

Phase (i)—free solvent evaporation

In the initial phase taking place temperatures below T_on, the required power for heating continuously increases until it reaches a maximum at approximately 85 °C. Such behavior is typical for heating processes where evaporation with increasing temperature is the dominant heat loss mechanism [24]. This is supported by the fact that the temperature of peak required heating power nicely coincides with the temperature of maximum evaporation of water and methanol as evident by the FTIR-EGA data shown in Fig. 4. Shortly afterwards, the required heating power drops, as the evaporation of these free solvents decreases.

Fig. 4.

Left—estimation of the gas evaporation from the sample as a function of temperature as determined by CLS regression. The different gases are given in the legend in the respective graph (top and bottom). Right—measured spectrum (black) and pure gas spectra calculated via CLS for T_on and T_off on the top and bottom, respectively

The NIR absorption spectra recorded from the sample surface in phase (i) are shown in Fig. 5a. Figure 5b shows the first principal component (PC1), that describes almost 95% of the change in the spectra in this first reaction phase, and the inset shows the scores on PC1 as a function of temperature. The spectra in Fig. 5a clearly exhibit a decrease in absorption around 1930 nm. This is also reflected by the high loadings on PC1 at the same wavelength and the continuously decreasing PC1 scores with increasing temperature. This absorption band is attributed to the H–O-H bending and O–H stretching combination mode [25]. Therefore, the reduction in absorption at this wavelength is strong evidence for the evaporation of excess water from the sample, which is also supported by the FTIR-EGA data. Furthermore, significant changes in the NIR absorption spectrum are observed in the wavelength region between 1400 and 1700 nm. An increase in absorption is clearly seen at 1440 nm, which is attributed to the first overtone of O–H stretching, related to hydroxyl groups [26] and is again reflected in PC1, which shows pronounced negative loadings at this wavelength. This increase can be interpreted as an increase in concentration of methylphenols in the sample as excess water and methanol are evaporated, resulting in a more pronounced absorption at this wavelength. This effect might be partly superimposed with the shift of the water absorption band around 1450 nm towards shorter wavelengths in the observed temperature range [27], which, together with the strong water desorption, also contributes to the decrease of the absorption signal between 1500 and 1700 nm. However, the latter effect is mostly contributed to evaporation of methanol from the sample [28], which makes up approximately 10% of the original sample mass. This is also clearly indicated by the FTIR-EGA measurements in this reaction phase. Interestingly, the shape of the decrease in PC1 scores as a function of temperature mimics the decrease in sample mass in the phase as shown in Fig. 3a, further supporting the assumption that in this reaction phase, sample changes are predominantly caused by evaporation of volatile compounds.

Fig. 5.

a NIR spectra recorded during phase (i) of the process, color-coded from blue (30 °C) to red (95 °C). b Loadings on the first principal component (PC1), describing almost 95% of the spectral variation in this phase. The inset shows the scores on PC1 as a function of temperature for this process phase

As already mentioned, FTIR-EGA data acquired in phase (i) clearly shows evaporation of water and methanol from the sample. Furthermore, the FTIR-EGA data indicates that formaldehyde, phenol, and formic acid release rates increase as the temperature rises, especially above a temperature of approximately 60 °C. As described in the “Paper-based phenolic laminates” section, the initial PF resin mixture is heated to approximately 60–90 °C for 1–2 h with a subsequent vacuum distillation to remove volatile compounds. As formaldehyde has a boiling point of approximately −19 °C, it would be expected that free formaldehyde in the mixture is completely removed already in this production step. The fact that it is clearly detected in the FTIR-EGA data can be attributed to steric hindrance causing residual formaldehyde in the sample mixture after this initial process step. As the temperature is increased, molecular mobility increases, which can eventually lead to the observed evaporation of formaldehyde from the sample. Another hint towards steric hindrance is the evaporation of phenol, as it would be expected due to the abundance of formaldehyde in the initial mixture that all phenol molecules react with formaldehyde to form methylolphenols. The formic acid is intentionally added to the resin in the production process by the manufacturer. It is not directly involved in the ongoing chemical reaction and therefore also eventually evaporated from the sample as the temperature is increased.

Phase (ii)—polymerization

In phase (ii), taking place at temperatures between approximately 95 °C and 147 °C, the exothermic condensation reaction of the resin starts, as clearly evident by the DSC measurements. The area between the baseline and the DSC curve in this reaction phase gives an estimation of the heat released by the ongoing condensation reaction, which is around − 60 J/g for both samples.

The NIR absorption spectra recorded during this phase are dominated by a decrease in absorption in the wavelength range between 1450 and 1800 nm, as well as a reduction in absorption above 1950 nm; both changes have already been reported previously [29, 30]. The latter change in the spectra can be attributed to the combination of O–H stretching and deformation modes in the methylol group, which is strong evidence for the ongoing polycondensation reaction. Absorption in the wavelength range between 1450 and 1800 nm can be attributed to C–OH, CHx, and aromatic C-H bonds [31], the observed reduction in absorption in this range is interpreted as methylene bridges being formed between the aromatic rings, as well as the evaporation of phenol and formaldehyde from the sample, as seen in the FTIR-EGA spectra in this reaction phase.

No considerable change in the water absorption band around 1930 nm can be seen in the NIR spectra recorded from the sample surface, as opposed to the spectra acquired in phase (i). This indicates that the water being released by the ongoing condensation reaction is immediately evaporated instead of accumulating in the sample due to the elevated temperatures above the boiling point of water.

As in phase (i), the scores on PC1, which describes over 98% of the total spectral variance, plotted as a function of temperature (inset in Fig. 6a), nicely resemble the weight loss curve in this reaction phase, as shown in Fig. 3a.

Fig. 6.

a NIR spectra recorded during phase (ii) of the process, color-coded from blue (95 °C) to red (147 °C). b Loadings on the first principal component (PC1), describing over 98% of the spectral variation in this phase. The inset shows the scores on PC1 as a function of temperature for this process phase

The FTIR-EGA data recorded in phase (ii) indicate that considerable amounts of water are still being released from the sample, explained by the ongoing condensation reaction. As the condensation reaction ends at approximately 147 °C, the amount of water released from the sample decreases significantly, as no more water is produced as a by-product of the resin polymerization process. The evaporation of formaldehyde, phenol, and formic acid all increase in this reaction phase, eventually each reaching their maximum at approximately 128 °C, 133 °C, and 140 °C, respectively. As already briefly discussed above, the delayed evaporation of these substances might be caused by steric hindrance and might be hinting towards a glass transition in the resin happening in this temperature range. Unfortunately, the standard DSC measurements cannot detect the glass transition, since it is superimposed by the much stronger effect of the ongoing exothermal condensation reaction. Additionally, since the glass temperature is strongly dependent on the curation degree of the resin, which continuously changes in this phase, it might not be feasible to define a glass transition temperature in this case.

Phase (iii)—final drying and curing

At temperatures above approximately 147 °C, the exothermic condensation reaction is mostly completed, as the DSC curve again closely follows the baseline, as evident from the top graph in Fig. 3b. The amount of heat required to increase the sample temperature shows a continuous slight decrease in this temperature range, as also the amount of water evaporated from the sample decreases, as visible in the FTIR-EGA data (Fig. 4). The NIR spectra, however, still exhibit a slight decrease in the absorption above 1950 nm, which might hint towards still ongoing polymerization, albeit at a strongly decreased rate (Fig. 7). Due to this slow reaction rate, this reaction might not be evident from the DSC data.

Fig. 7.

a NIR spectra recorded during phase (iii) of the process, color-coded from blue (147 °C) to red (250 °C). b Loadings on the first principal component (PC1), describing approximately 91% of the spectral variation in this phase. The inset shows the scores on PC1 as a function of temperature for this process phase

A strong increase in absorbance around 1430 nm and decrease around 1550 nm is also clearly visible in the NIR spectra, which are similar to the changes observed in phase (i). The FTIR-EGA data recorded in this phase show that the water evaporation from the sample decreases, as no more water is produced after the condensation reaction is finished. The small amounts of water that are still detected indicate further drying of the sample. Furthermore, a sharp decrease in the evaporation of phenol, formaldehyde, and formic acid is visible. The reason for this is on the one hand that the concentration of these substances in the sample is exhausted and on the other hand that the polymerized resin causes steric hindrance, prohibiting further evaporation, at least at rates that can clearly be measured with this technique. Significant evaporation of both CO₂ and at higher temperatures also CO is evident from the MIR-spectral data, hinting towards thermal degradation of the cured resin. This together with the continued evaporation of water from the sample explains the continued weight loss detected in the TGA data and agrees with literature, where evaporation of CO₂ and CO from cured resins have been reported starting at approximately 150 °C and 200 °C, respectively, both in air and N₂ atmosphere [32].

Inline NIR measurements

From the laboratory measurements discussed in previous sections, it was concluded that NIR spectroscopy should be a useful tool for the real-time quality control at the production line of the semi-finished phenolic resin laminate. NIR spectroscopy combined with chemometric data analysis techniques has already been widely applied to monitor and control biological [33–36] as well as chemical reactions [37–40] including the inline monitoring of resin polymerization processes [30, 41]. It should be noted that Raman spectroscopy is another measurement technique already successfully utilized for monitoring of the curing process of resol PF resins [42]. However, it was found in laboratory measurements (data not shown) that the black dye added to the resol resin investigated herein causes a strong fluorescence signal. This was observed with both 785 nm and 830 nm excitation wavelengths, overshadowing the chemical information encoded in the Raman spectra, rendering it unsuitable for inline process control.

The laboratory measurements discussed above clearly show significant and characteristic changes in the NIR spectra for the different phases of the ongoing polymerization process. Therefore, NIR absorption data of the produced phenolic laminates, should provide useful information on their degree of curing. Thus, the NIR data should nicely correlate with the resin flow and volatile compounds left in the sample (see the “Determination of volatile content” and “Determination of resin flow” sections), which are both routinely measured at the production site for sporadic quality control.

To assess the suitability of NIR spectroscopy as a tool for real-time inline quality control, a dedicated NIR measurement head (details described in the “Inline NIR measurements” section) was installed at the production site after the cooling rolls (see also Fig. 1). At this position, the temperature of the sample is relatively stable at 32 ± 2 °C. This is important to mitigate the influence of temperature effects, such as wavelength shifts of different absorption bands that can influence the output of the chemometric model applied to the NIR spectra.

In a feasibility measurement, approximately 18,500 NIR absorption spectra were collected over the duration of roughly 3 h and 10 min, with some short downtimes of the NIR measurement system in between, caused by software issues. During this timeframe, different operation parameters of the dryer were intentionally changed to provoke changes in the curing degree. This was done to introduce the associated changes in the NIR spectra to test if the curing degree can be derived from the acquired NIR spectral data by chemometric modeling. The dryer parameters changed were air fan speed and air temperature, the heating roller temperature as well as the feeder speed of the paper web. As one would expect, lower temperature and reduction in fan speed result in lower curing degree and thus increased resin flow and volatile content in the sample. Increased heating roller temperature is known to cause slightly higher impregnation degree of the paper, which results in a reduction of curing degree. The last parameter is the paper web speed, which when reduced causes a higher curing degree, due to the longer dwell time of the sample in the dryer. The values of these dryer parameters are plotted as a function of time in Fig. 9a and b.

Fig. 9.

Top—The four dryer parameters intentionally altered during the feasibility NIR measurement at the production plant of the paper-based PF resin laminates as a function of time. Bottom—PLS predictions calculated from the NIR spectra acquired at the production plant of volatile content (blue) and rein flow (orange) as a function of time. The blue and orange areas in the graph correspond to the in-spec ranges for the volatile content and resin flow, respectively. The black circles represent the results of the offline reference measurements taken at the production plant

A total of 11 samples were taken for reference measurements of resin flow and volatile content (see description in the “Determination of volatile content” and “Determination of resin flow” sections, all obtained reference measurements are provided as electronic supplemental material in Table S1). It should be noted that both offline reference measurements are strongly correlated, as they are both means to estimate curing degree. Therefore, it was chosen to model both resin flow and volatile content with a single PLS regression model. For each reference sample, 50 subsequently recorded NIR spectra taken roughly in a timeframe of ± 13 s surrounding the passing of the later measured sample section in front of the NIR sensor were considered as reference spectra. For the volatile content, only the reference measurement of the sample taken from the center was considered, since also the NIR-measurement is positioned in the center of the sheet.

As the laboratory investigation of the PF resin curing using NIR spectroscopy showed no significant spectral features below 1300 nm related to sample curing, it was chosen to consider only the wavelength region from 1300 to 2050 nm for the PLS modeling. The number of latent variables (LVs) for the PLS model was chosen via cross validation (CV), where 11 data splits were performed, each excluding the 50 subsequently recorded NIR spectra corresponding to a single reference measurement. The root mean square error of cross validation (RMSECV), showed a clear minimum with 4 LVs for both reference measurements at approximately 0.44 wt% and 0.19 wt% for the volatile content and resin flow, respectively.

Unsurprisingly, the PLS regression vectors for volatile content and resin flow, shown in Fig. 8a in blue and orange, respectively, are almost identical as they extract very similar information from the NIR spectra. A closer look at the loadings of the two regression vectors reveals a significant positive contribution of the NIR absorption around 1930 nm to both volatile content and resin flow. As discussed in the “Phase (i)—free solvent evaporation” section, where a significant peak in the principal component scores at this wavelength was observed, absorption at this wavelength is attributed to the H–O-H bending and O–H stretching combination mode and is strongly correlated to the amount of water present in the sample. Since more residual water directly leads to higher values for volatile content and resin flow, positive loadings at this wavelength are expected for both regression vectors. Absorption between 1570 and 1730 nm also contributes positively to the predicted values as evident from the positive loadings on the regression vectors in this region. This is in good agreement with the laboratory measurements, where positive contributions in this wavelength range have been observed in the loadings of the first principal components. As discussed above, changes in this wavelength region are attributed to methylene bridges being formed between the aromatic rings, as well as the evaporation of methanol, phenol, and formaldehyde from the sample (see the “Phase (i)—free solvent evaporation” and “Phase (ii) of the drying and curing process” sections). Negative loadings on the regression vectors can be seen between 1400 and 1500 nm, which is consistent with the observations made from the principal components in the “Phase (i)—free solvent evaporation” and “Phase (iii) of the drying and curing process” sections. Negative loadings are also visible in the regression vectors around 1800 nm. Similar negative changes in absorption close to 1800 nm were observed in the laboratory measurements in phase (iii). This, combined with the negative loadings in both regression vectors, suggests that changes in the sample occurring in phase (iii) of the ongoing process (see the “Phase (iii) of the drying and curing process” section) have a significant influence on volatile content and resin flow of the product in the industrial production environment.

Fig. 8.

a PLS regression vectors for volatile content (blue) and resin flow (orange). b, c PLS prediction vs. offline reference measurement for volatile sample content and resin flow, respectively. The error bars show the standard deviation of the PLS predictions calculated for 50 subsequently recorded NIR spectra

Figure 8b and c show the values calculated from the NIR data using the PLS model vs. the values determined via the corresponding offline reference measurements. These plots show that the standard deviation of the predicted value for the 50 subsequently recorded spectra, indicated by the error bars, is of a similar relative magnitude for both reference measurement methods and amounts to approximately 4% of the respective measurement range. However, the overall fit of the mean predicted values fits visibly better for the volatile content. This is explained by the variation in curing degree across the width of the produced phenolic laminate. While the volatile content is measured in the center of the product, roughly at the position of the NIR measurement spot, the measurement of the resin flow only gives the average value over the entire width of the sample.

The calibrated PLS model allows to evaluate volatile content and resin flow as a function of time, which is depicted at the bottom in Fig. 9 with the same time axis as the different dryer parameters shown in the graph above. The blue and orange areas mark the acceptable ranges for volatile content and resin flow of 6.2 wt%–6.8 wt% and 0.6 wt%–1.0 wt%, respectively. In the beginning both reference measurements are mostly inside the desired range. One exception is the measurement of the volatile content for the second reference measurement, which is only 5.8 wt% and thus slightly too low. The resin flow on the other hand is still within spec, hinting towards differences in curing degree across the sample width. In fact, the measurements of the volatile content at the edges of the sample are higher with values of 6.2 wt% and 5.9 wt%.

After the dryer temperature and fan speed were decreased and the heating roller temperature increased, both volatile content and resin flow rose significantly, which is also nicely reflected in the PLS predictions for both values. Interestingly, the fourth reference measurement is still in-spec for the resin flow, but already out of spec when it comes to volatile content. In the PLS prediction also the resin flow value is out-of-spec at the time of the measurement, which could again be caused by curing degree inhomogeneity across the width of the sample. As expected, further decrease in dryer temperature and increase in heating roller temperature continues the upwards trend of volatile content and resin flow, both in the PLS predictions and offline reference measurements.

After the reduction of heating roller temperature and increase in dryer temperature around 10:50, the curing degree at the end of the dryer slowly increases again before reaching in-spec values according to the PLS prediction shortly after the seventh reference measurement was taken. In-spec values are also measured for the eighth reference sample, which agrees with the inline NIR measurements.

Around 11:40, the dryer temperature and air flow were increased, while sample speed was reduced to provoke a higher curing degree. This heightened curing degree is nicely visible in both the inline NIR and the offline reference data taken thereafter. Around 12:13, both sample speed and dryer temperature were reset to usual values for the product to return to in-spec production. This leads to a sharp increase in the predicted value for volatile content and resin flow in the PLS predictions, which is later confirmed by the delayed offline reference measurement.

Overall, a very good agreement of PLS predictions and offline reference measurements is achieved. The PLS predictions in Fig. 9 nicely demonstrate the quick response time of the NIR measurement which provides much better insight into the ongoing changes in curing degree upon change of the drying parameters. This can be exploited for better process control and energy savings, as trends can be tightly monitored and dryer settings can be set accordingly in real time to produce high-quality products with the desired curing degree, while using only the minimum required heat energy.

Conclusion

In this work, we demonstrated a thorough laboratory investigation of the curing process of phenolic resin using multiple different thermophysical and optical measurement technologies. While TGA and DSC measurements provided information on the thermophysical properties and reaction kinetics, the NIR spectral data that were simultaneously recorded during the TGA, unveiled additional chemical information directly from the sample. This was achieved via a custom-built measurement setup that allowed to introduce a NIR reflection probe directly into the STA system for a simultaneous multimodal measurement. Additionally, FTIR-EGA provided valuable complementary information on the evaporated volatile compounds from the investigated sample, shedding further light on the reaction kinetics of the investigated resol PF resin on kraft paper.

As the laboratory investigation showed a clear correlation between changes in the NIR spectroscopic data and the degree of curing of the investigated sample, it was decided to implement a real-time NIR-based measurement system into a production line for PF resin laminates. To assess the feasibility of the NIR measurement for real-time monitoring and control of the products curing degree directly in the industrial production plant, a measurement campaign was carried out where 18,500 spectra were acquired, and 11 reference measurements were taken. The inline NIR measurements resulted in a tight correlation with the offline reference measurements of volatile content and resin flow, which are both means of estimating the curing degree of the produced semi-finished phenolic laminate. In summary, the study demonstrates a multimodal analysis of the resin curing process to enable detailed insights into the physicochemical drying and curing process of resol PF resin. Building on these results an NIR spectroscopy measurement setup was developed and successfully implemented, proving its suitability as a tool for real-time quality monitoring in the industrial production of paper-based PF resin laminates.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contribution

R.Z., S.F., D.W., and M.B. designed, performed, evaluated, and interpreted the NIR measurements. J.K. designed and performed the TGA and DSC measurements. J.K., T.B., and R.Z. evaluated and interpreted the TGA and DSC measurements. D.L. designed and performed the FTIR-EGA measurements. D.L., R.Z., and M.B. evaluated and interpreted the FTIR-EGA-measurements. G.L and E.P. were responsible for measurements at the industrial dryer and provided samples and reference measurements. R.Z., J.K, D.W., G.L., T.B., and M.B. drafted the manuscript. R.Z. prepared the figures. All authors contributed to editing of the manuscript. All authors approved the final version of the manuscript.

Funding

The authors acknowledge funding by the Produktion der Zukunft project DATA (Digitaler Assistent für Trocknungsanlagen, No.: FO999891216).

Data availability

The data that support the findings of this study are available from the corresponding author, R.Z., upon reasonable request.

Declarations

Conflict of interest

The authors declare no competing interests.