Introduction

Melamine-urea-formaldehyde (MUF) or urea-formaldehyde (UF) resins are extensively used in the production of medium density fibreboard (MDF) panels as well as other wood composite panels such as oriented-strand board (OSB), laminated veneer lumber (LVL), plywood, glue-laminated beams (Glulam) and particle board. These composite wood products are engineered for specific structural performance, particularly LVL and Glulam beams, which are engineered to span distances greater than that of conventional wooden beams. Because these wood products are used in commercial and residential construction, the resins are often modified to impart moisture resistance or to minimise formaldehyde emission, particularly to meet the stringent Californian EPA Air Resources Board emission standards which, from January 2011, require formaldehyde emission from finished MDF panels to be less than 0.11 ppm [1]. As a result, manufacturers of wood composite products are concerned with characterisation of MUF resins used in their manufacturing process.

Often, the MUF resins are manufactured at a resin plant distant from the MDF production facility, requiring the resin to be manufactured and then transported to the panel manufacturing plant where it is transferred to a storage tank. Resins are manufactured in batch conditions to an individual client’s specific requirement, based generally on the moisture resistance and formaldehyde content. At the time of manufacture, the resin batches are tested for a variety of properties including viscosity, pH, solids content [2] and then a Certificate of Analysis is issued. These analyses are rudimentary, in that they are simple to perform but do not often relate to key performance properties of interest by the panel-production facility, such as molecular weight distribution and degree of cross-linking. Furthermore, once the fresh consignment of resin arrives at the panel mill, the contents of the transport vessel are transferred to storage tanks which may or may not contain large volumes of existing, aged, resin. This results in an unknown degree of dilution of the fresh resin with older resin so that the properties of the resin in the storage tank no longer resemble the resin described by the Certificate of Analysis. It is not normal practice to perform exhaustive tests of the resin in use as they are often lengthy (solids content) or subjective (gel time) or of little practical value to a MDF manufacturing plant (pH). Instead during the manufacture of MDF panels, a selection of panels are regularly sampled for quality control involving not only formaldehyde emission testing but also mechanical performance via internal bond (IB) strength, thickness swell, modulus of elasticity (MOE) and modulus of rupture (MOR). All these analyses are performed after the panels are manufactured and provide a feed-back loop to the production process. Knowledge of the MUF resin properties that are of significant value to the manufacturing process (reactivity as a function of gel time, extent of cross-linking, molecular weight), at the point of being sprayed onto the fibre mat, would provide a feed-forward loop to further aid process optimisation. The requirement therefore is to provide on-line (or at-line) determination of resin properties that are of meaningful utility to panel manufacturers. This might include properties such as the viscosity, pH, gel time and solids content of the resin, but also knowledge of the chemical functionality of the resin such as the extent of methylene branching, the amount of free formaldehyde or the ratio of formaldehyde to urea. This may allow for optimised press conditions to maximise bond formation and reduce the amount of residual free formaldehyde in the final panel. Such a feed-forward/feed-back PAT system using chemometric analysis of spectroscopic data has successfully been established for hardboard production [3].

Due to the qualitative nature of mid infrared spectroscopy, considerable research has been conducted in this region of the electromagnetic spectrum [4–7]. Similarly, nuclear magnetic resonance (NMR) spectroscopy, particularly carbon-13 nuclear magnetic resonance (13C NMR) spectroscopy, has proven to be very successful in characterising the speciation and extent of polymerisation and polymeric branching in resins, and MUF resins specifically [8–13]. Near infrared (NIR) spectroscopy, while lacking the qualitative chemical speciation of mid infrared, does allow superior in-, on- and at-line application due to the ease of use of high transmission fibre optic cables and/or much simpler sample preparation and presentation to the instrument and has been used as a means to determine urea content of UF resins [14–16] and to monitor the long-term stability of melamine formaldehyde resins [17]. Near infrared spectroscopy has also been used with success to follow the curing of acrylate and silicone adhesives [18, 19], for analysing the glueline in laminated timber products [20] and to predict the phenol-formaldehyde resin content of oriented strandboard panels post manufacture [21]. While other NIR calibrations of resin properties have been studied and 13C NMR spectroscopy has been used to characterise resins, this paper reports the calibration of NIR spectra with MUF chemical functionality derived from analysis of corresponding 13C NMR spectra and the use of the resulting calibrations in providing daily control chart data in a MDF production facility. This paper is quoted in parts from an industry report with permission from Forest and Wood Products Australia Ltd [22].

Methodology

Sampling

Two differing commercial resins were sampled during the study: a “conventional” MUF resin and a water-resistant MUF resin which will be identified as Resin A and Resin B, respectively. Absolute values of formaldehyde content in particular are not reported in this paper for commercial reasons.

The trial was conducted at the Laminex Group’s MDF manufacturing plant near Gympie, Queensland, Australia. Samples of resin were collected from the two glue lines in the morning and early afternoon in 200 mL HDPE sample bottles. NIR spectra were acquired on-site on the fresh resins within half an hour of collection. The 13C NMR spectra were acquired off-site within 36–48 h of sample collection.

Near infrared spectroscopy

Transmission NIR spectra of the samples were acquired using a ThermoFisher Scientific Antaris II FT-NIR spectrometer (ThermoFisher Scientific, Waltham, MA, USA, www.thermofisher.com). The spectra were acquired between 4000 and 10,000 cm−1 at a spectral resolution of 8 cm−1. A sample of resin was transferred to a Ziploc bag and the bag compressed to a 1-mm pathlength using the tablet transmission accessory. This had the added benefit of simple disposal of the resin following analysis with no clean-up being required. Representative spectra of resins A and B are shown in Fig. 1 along with spectra of the three monomeric species. The third and fourth stretching overtones of the carbonyl are evident at 6097 and 5265 cm−1 in the spectrum of formaldehyde [23].

Fig. 1
figure 1

Representative NIR spectra of resin A, resin B and the three monomeric species. Note differences in the 5000 cm−1 range (N-H stretching of urea)

Full size image

13C NMR spectroscopy

13C NMR spectroscopy was undertaken on samples of resin using a Bruker Avance 400 NMR spectrometer (400.13 MHz 1H, 50.3 MHz 13C). In order to eliminate the nuclear Overhauser effect (NOE), an NMR sequence without NOE enhancement was employed to acquire quantitative spectra using a 5-s relaxation delay. In excess of 8000 scans were acquired per sample during overnight acquisition. A small quantity of DMSO-d6 was added to provide signal lock and spectral referencing (DMSO-d6, septet, 39.50 ppm). A representative spectrum is shown in Fig. 2 with assignments of resonances given in Table 1.

Fig. 2
figure 2

Quantitative 13C NMR spectrum of fresh MUF sample showing chemical shift (top) with expansions of the two regions (bottom). Reproduced with permission from Meder et al. [22]

Full size image
Table 1 13C NMR chemical shift assignment for MUF resins (after Ebdon and Heaton [8])
Full size table

Data analysis

The 13C NMR spectra were integrated according to regions identified by Panangama and Pizzi [12] to determine the chemical speciation of the resins (Table 2). Using these regions, the ratio of formaldehyde to urea is derived according to:

Table 2 13C NMR integration areas (normalised to 100% area) for the major species in the designed resins. The integration areas are those described in Table 1 (after Ebdon and Heaton [8])
Full size table
$$ \mathrm{F}:\mathrm{U}\ \mathrm{mole}\ \mathrm{ratio}=\frac{{\displaystyle \sum \left(\mathrm{A}46:\mathrm{A}93\right)-\mathrm{A}49-\mathrm{A}55}}{{\displaystyle \sum \left(\mathrm{A}160:\mathrm{A}163\right)}} $$

Multivariate analysis of the NIR data and resin properties was performed using The Unscrambler v10.4 (Camo Software AS, Oslo, Norway, www.camo.com) to undertake principal component analysis (PCA) and projection to latent structures (PLS) regression. Calibration was performed on either raw spectra or following first or second derivative transformation of the spectra using a Savitzky-Golay transform [24] with 15 points and second order polynomial fit.

Results

Qualitative comparison of resins

Principal component analysis of all the acquired spectra highlights the relationship between individual samples due to the variance in their respective spectra. Figure 3 presents the PCA scores plot of both resin A and resin B, showing clear separation of the resins based on differences in the spectra of the resins. This has already been seen visually in Fig. 1 and is shown qualitatively in the PCA loadings plot in Fig. 4. The loadings plot identifies systematically the region(s) of the spectra that show the greatest variance for, in this case, the first three principal components. The predominant feature that distinguishes between the two resins is the peak at 5200 cm−1 (1908 nm). This is assigned as an N-H stretch overtone band associated with urea. The sharp peak at 4322 cm−1 (2313 nm) is assigned tentatively to urea species [10, 11, 14]. Further, detailed discussion of the calibration and use of the NIR predictions will be made for resin A, the conventional resin, only.

Fig. 3
figure 3

Scores plot from PCA analysis of A and B resins. Reproduced with permission from Meder et al. [22]

Full size image
Fig. 4
figure 4

Loadings plot for the first three principal components of principal component analysis of fresh resin A and resin B

Full size image

Quantitative calibration of NIR spectroscopy with physical properties

Partial least squares regression calibration of the NIR spectra with resin properties was performed in two steps. Firstly, calibrations for the physical properties were made and are summarised in Table 3. The calibrations are made using either multiple Y response-variables (PLS-2) or a single Y response-variable (PLS-1) using a first derivative Savitzky-Golay spectral transform [24]. It can be seen that physical properties such as the results from an Automated Bond Evaluation System (ABES) [25] at 120 s test time and from gel time tests are readily calibrated with R 2(valid) values of 0.85 and 0.69, respectively (Table 3). Gel time is traditionally an operator dependent and highly subjective measurement that involves determining the time point at which gelation begins. Once NIR spectroscopy has been successfully calibrated with gel time assessed by a single operator, subsequent measurements via NIR spectroscopy are operator independent as the large operator bias for the gel time measurement is removed.

Table 3 Summary of NIR calibration statistics for various model combinations. First derivative, second order polynomial
Full size table

Table 3 and Table 4 also show that NIR spectra of the fresh resins can be correlated with several chemical properties of the resin, namely free formaldehyde content, formaldehyde/urea ratio, specific gravity and solids content. The calibration models show some initial potential for calibration of NIR spectroscopy with other properties such as pH and viscosity although the full cross-validation of the model does not produce a reliable validation. This may in part be due to the low number of samples in the model.

Table 4 Summary of NIR calibration statistics for various model combinations. First derivative, second order polynomial
Full size table

Quantitative calibration of NIR spectroscopy with chemical properties

Secondly, PLS regression calibrations were developed for the chemical composition as determined from 13C NMR spectroscopy and these are summarised in Table 4.

Practical use of NIR calibrations for process monitoring

The effect of storage temperature on pH and free formaldehyde for resin A over a period of time based on NIR-predicted values is shown in Fig. 5. Comparison of the actual pH values and NIR-predicted pH values for storage over time at 25 °C (r 2 = 0.98) and 35 °C (r 2 = 0.85) are also shown. The decrease in pH occurs more rapidly at higher storage temperatures and there is good agreement between the NIR-predicted pH and actual pH values.

Fig. 5
figure 5

Top—NIR-predicted pH and actual pH and bottom—NIR-predicted free formaldehyde content of resin A during storage at 25 and 35 °C. Reproduced with permission from Meder et al. [22]

Full size image

Similarly, storage at 35 °C results in more rapid loss of formaldehyde than storage at 25 °C. In fact, it would suggest that loss of formaldehyde during storage at 25 °C is minimal; however, the NIR-predicted values are negative during the later stages of storage. This suggests that the formaldehyde is out of range of the NIR calibration, so while the values are inaccurate the suggested trend is that there is very little free formaldehyde remaining after several days’ storage at either temperature.

Near infrared predicted values of formaldehyde were used to prepare control charts, an example of which is shown in Fig. 6 for free formaldehyde content and Fig. 7 for gel time. These figures show that the formaldehyde content and gel time of the resin are consistently within the 2σ level for the month of February 2011 although significant variation is observable.

Fig. 6
figure 6

Control chart plot of NIR-predicted free formaldehyde in resin A for daily samples collected over a 5-week period. Adapted from Meder et al. [22]

Full size image
Fig. 7
figure 7

Control chart for gel time of resin A based on NIR-predicted values over a 5-week period. Reproduced with permission from Meder et al. 2009. Adapted from Meder et al. [22]

Full size image

Determination of resin gel time is a highly subjective test and is prone to individual operator bias. PLS regression of the NIR spectra with manually measured gel time results in a calibration that, while it may contain bias from a single operator, is then independent of operators as the bias is inherent in the calibration and not the operator. The correlation of NIR-predicted gel time vs NIR-predicted free formaldehyde for resin B resin has an r 2 value of −0.87, indicating that as the free formaldehyde content decreases the gel time increases.

Four of the A resins were adulterated to alter their properties in order to test the ability of NIR to identify “outliers”. A control chart for free formaldehyde was prepared using the NIR-predicted values for the fresh resin from the reactor vessel (Fig. 8). The free formaldehyde content for the adulterated resins was also predicted from the NIR spectra and plotted on the same control chart. Note that two of the adulterated resins lie outside the ±3σ control lines while two resins are within the control lines.

Fig. 8
figure 8

Control chart of NIR-predicted formaldehyde (top) and pH (bottom) for resin A sampled on a daily basis. M1–M4 are modified A resins, 35C was stored at 35 °C, Fy2% has 2% added formaldehyde, EW2% is a B resin and has 2% extra water, pH12 is a B resin adjusted to pH 12. The dashed lines are the +3σ and −3σ limits for the daily samples. Reproduced with permission from Meder et al. [22]

Full size image

Conclusions and recommendations

In establishing the instrument protocol for acquisition of NIR spectra, it was noted that a rapid decline in NIR-predicted formaldehyde content with time was observed when the resin was stored in resealable zipper bags (Ziploc®) and spectra re-acquired over a period of 4 days (data not shown). There was also some lesser rate of decline observed when samples were stored in HDPE containers and freshly transferred to resealable bags for spectral acquisition. Based on this result, samples stored in resealable bags were not considered stable for longitudinal studies, with the focus being on fresh sampling.

Although this study was conducted using at-line sampling, the use of fibre optic probeheads would enable sampling to occur in situ, either directly in the storage tank or preferably in a recirculating side stream. By contrast, there was no regular resin testing conducted at the manufacturing plant, primarily for reasons of timeliness. Determination of solids content is lengthy as it requires drying the resin in an oven overnight and gel time is subjective to measure. Instead, MDF panels were selected once per 8-h shift for destructive testing. The turnaround time on the analysis meant that it was not until the following day that the results were available to the operations crew to determine whether any deviation in quality had occurred. Installation and commissioning of the NIR in the MDF manufacturing plant now enables at-line testing of the resin at regular 4-hourly intervals (and potentially every 30 s if an in-line system was installed), allowing more timely monitoring and control of resin properties providing proactive control of manufacturing conditions to ensure panel performance remain within specification.

While 13C NMR has previously been used to characterise UF/MUF/MF resins [10–12], this study has established correlation between NIR spectra and NMR-derived speciation of resins such as the formaldehyde/urea ratio, total methylol and the branched methylene content. This provides understanding of the molecular speciation and extent of cross-linking, and hence the extent of reactivity, of the resin.

Unlike Certificates of Analysis which do not represent the actual resin condition once a batch is diluted with aged resin in the storage tanks, near infrared spectroscopy provides not only timely characterisation of the resin but also characterisation of more process-relevant properties (free formaldehyde content, molecular formulation and gel time) in an at-line PAT application. To date, no attempt has been made to determine correlation between the NIR-predicted properties of the resins with observed panel performance, although unpublished data (Meder and Thumm, unpublished) has shown strong correlation of NIR with panel IB strength, MOE/MOR and thickness swell. Application in a MDF manufacturing plant with greater variability in resin properties would be required in order to establish correlation between resin property(ies) and panel performance in a manner such as that established for production of hardboard.