Univariate and Multivariate Analysis of Tannin-Impregnated Wood Species Using Vibrational Spectroscopy

INTRODUCTION

A cradle-to-cradle approach is used in miming the naturally occurring processes for synthetic systems,^1–3 and protecting wood with the same processes occurring in the plant kingdom represents an interesting solution for a cradle-to-cradle-designed wood preservative. Trees protect their living matter with resistant dead cells in the bark and heartwood. Metabolic pathways are activated during the natural physiological aging process, including a complex series of chemical reactions that produces tannins.^4,5 Protecting wood with tannins has been known for decades.^6,7 Within the vegetal world, such compounds are used as preservatives, offering protection against light (e.g., developing free radicals due to ultraviolet irradiation) and against biological attack by insects, fungi, and bacteria.^8–10 Advanced formulations based on condensed tannins allow us to overcome the considerable drawback of leachability.^11–13 Thus, preservative retention and its distribution in impregnated wood are very important issues relating to the service life of these products. Diluted tannin-based water solutions with low viscosity allow a liquid preservative to penetrate the wood structure and disperse more homogeneously within the wood.^14–18

Thus, a fast and nondestructive method is required for transfer from laboratory to industrial conditions for the analysis of tannin distribution in wood. Fourier transform infrared spectroscopy (FT-IR) can be used to investigate the changes in wood components and preservatives.^19,20 Many studies have examined the use of tannins for various products and their influence on these materials' properties;^21–24 however, using different types of vibrational spectroscopy to examine tannins and their interactions with wood components are rare.

Pizzi ²⁵ mentioned that the infrared (IR) band at about 1450 cm⁻¹ corresponded with the aryl ring vibrations of the tannin resin. Other researchers have characterized various tannin extracts from different sources via FT-IR.^26–28 Kim and Kim ²⁹ analyzed the curing behavior of tannin-based adhesives and found that the different molecule bridges of the pure tannin resin (mimosa tannin hardened with hexamine) can be observed only for the N–H wagging peak at 750 cm⁻¹, whereas Ozacar et al. ³⁰ conducted similar studies related to condensation with formaldehyde to analyze various reaction products. The curing reactions between the mimosa and chestnut tannin and hexamine at various acidic and alkaline pH values were analyzed by Peña et al. ³¹ They concluded that different chemical structures are obtained at various pH values of the solutions. However, studies on the interaction between tannin and wood components by vibrational spectroscopy techniques have not been published to date.

Here, we analyzed of the interaction between tannin and wood components and showed the possibilities of focusing on FT-IR, FT-Raman, and FT near-infrared (NIR) for the classification of various tannin-impregnated wood samples.

EXPERIMENTAL

Materials. The sapwood samples were obtained from Norway spruce (Picea abies (L.) (Karst.)), European beech (Fagus sylvatica (L.)), and Scots pine (Pinus sylvestris (L.)). They were cut to 25 mm³ × 25 mm³ × 25 mm³, without any discolorations and knots, from one board of each of the wood species, and then the samples were dried for two days at 103 °C before impregnation.

Tannin impregnation solutions were prepared with 2.5, 5, 10, and 20% (w/w) mimosa (Acacia mearnsii) extract in deionized water. The pH values of these solutions were corrected with 50% sodium hydroxide at pH 9.0. Finally, a solution of 33% by weight hexamine (hexamethylenetetramine) was added so that the solid hexamine/tannin ratio was always 6 : 100.

Methods. Impregnations. The cubic samples were placed into a desiccator, and a 800 Pa vacuum was applied for 30 min to remove most of the air trapped in the wood cells. Then, the desiccator was filled with the impregnation solution, and the pressure was slowly increased to environmental pressure. The specimens were kept in the tannin-based solution for 24 h before being removed and lightly wiped with blotting paper. The wet samples were kept for 16 h at 103 °C to allow the tannin-based resin to harden. Subsequently, the treated samples were stored in a climatic chamber (20 °C and 65% relative humidity) until constant weight. The weight of the wood samples was monitored during the tannin treatment to evaluate liquid and solid uptake. For every tannin concentration at least three samples were used.

Fourier Transform Infrared Measurements with an Attenuated Total Reflection Unit . The spectra were recorded between 4000 and 600 cm⁻¹ with 64 scans at a resolution of 4 cm⁻¹ using a Frontier FT-IR spectrometer (PerkinElmer) equipped with a Miracle diamond ATR accessory with a 1.8 mm round crystal surface. Samples were measured by exposing the tangential face directly on the diamond eye, without further treatments. For every wood species and tannin treatment, one sample was selected and three single spectra per sample were averaged. All the spectra were ATR corrected and baseline corrected in the wavenumber range between 4000 and 600 cm⁻¹ using the rubberband method and normalized to two for the most intense band.

Fourier Transform-Raman . The tangential surfaces of the samples were also examined in the NIR spectral region by FT-Raman using an IFS 66 FT-IR spectrometer (Bruker) equipped with an FRA106 Raman module and laser excitation at 1064 nm. The spectra were recorded from 90 to 3500 cm⁻¹ by acquiring 1000 scans with a spectral resolution (full width half-maximum of the apparatus function) of 4 cm⁻¹. The size of the laser spot on the sample was approximately 100 μm, and the laser power on the sample was about 100 mW (equivalent to an irradiance of roughly 1.3 kW cm⁻²). The spectra were then elaborated with Opus 5.5 acquisition software using the rubberband method for baseline corrections.

Fourier Transform Near-Infrared Spectroscopy . The FT-NIR spectra were obtained from the radial, tangential, and longitudinal surfaces by an MPA spectrometer (Bruker) equipped with a fiber probe (4 mm measurement diameter) at a resolution of 8 cm⁻¹ (32 scans). For every wood species and tannin treatment, a minimum of three samples were selected. Two single spectra of each surface (radial and tangential) per sample were averaged to minimize the influence of various tannin retentions on different wood surfaces.

Data Analysis. Cluster Analysis. The clustering method used was Ward's algorithm implemented in Opus 5.5 software, wherein at the beginning of clustering, every spectrum is treated as one specific group and the spectra that show the least scattering are grouped in a cluster. ³² The FT-IR, FT-Raman, and FT-NIR spectra were pretreated by using the second derivative (17 smoothing points) of vector-normalized spectra across selected ranges of wavenumbers: FT-IR, 3794–2787 and 1814–774 cm⁻¹; FT-Raman, 3118–2705 and 1789–242 cm⁻¹; and FT-NIR, 8000–4000 cm⁻¹.

Partial Least Squares Regression (PLSR) Method . The Unscrambler® 9.7 software (CAMO, Norway) was used for the PLSR analysis. The PLSR was applied to find the latent variables in X (e.g., NIR spectra) that would best describe the variables in Y (e.g., dry tannin retention). Esbensen ³³ and Kessler ³⁴ give detailed information about the PLSR method. The NIR spectra were not pretreated, and the NIR data sets of various wood species were regressed against tannin dry retention for the cross-validation (root mean square error of cross validation [RMSECV]) of the models. ³⁴

RESULTS AND DISCUSSION

Retention of Impregnating Solution. The chemical information from vibrational spectroscopy should be dependent on the amount of solid wood preservative deposed on the surface of the sample. The average wet and dry retention obtained for each impregnation series are summarized in Table I.

OPEN IN VIEWER

TABLE I.

Estimated mean and SD of wet and dry retention of various tannin solutions.

Treatment	Wet retention, %	SD, %	Dry retention, %	SD, %
Scots pine
2.5% tannin	163.8	5	3.95	0.1
5% tannin	164.5	1.1	7.85	0.1
10% tannin	162.3	6.8	15.7	0.8
20% tannin	178.8	5	32.4	1
Norway spruce
10% tannin	119.8	18.1	10.4	2
20% tannin	73.3	2.3	10.8	0.6
European beech
10% tannin	105.3	10.4	10	0.9
20% tannin	107.6	14.3	20.1	2.6

The results showed that wood species that are more easily impregnable reach similar values of wet retention when different tannin solutions are used. The deposited tannin retentions for Scots pine and European beech wood samples are proportional to the concentration of the impregnation solution. For the Norway spruce samples, the impregnations were not complete in cross section of the samples and the core was still untreated, but the impregnation process can still be considered successful for our purposes because the spectra reflect the chemistry on the external surface of the sample. Thus, the local concentration of the deposed tannin also can be considered proportional to the concentration of the impregnating solution for Norway spruce samples.

Univariate Analysis Methods. Fourier Transform Infrared with an Attenuated Total Reflection Unit. A series of tannin-treated samples were characterized by FT-IR. To analyze the interaction between the tannin–hexamine preservative and the wood, the absorbance spectra of the 2.5, 5, and 20% tannin-treated surface of Scots pine was compared with that of untreated Scots pine (Fig. 1). The spectra of various tannin-treated Scots pine samples presented the same peak profiles as the untreated sample, with some significant changes. This tendency is detected for all peaks at 1454, 1270, and 1230 cm⁻¹, although the most considerable increase is for the peak at 1610 cm⁻¹ that is also the significant peak for the mimosa tannin (C=C stretching of the aromatic ring). In general, the intensities of the bands that increase are typically the aromatic signals of wood,^35–38 suggesting that the polymer renders the surface of the treated sample more aromatic. Otherwise, each of the signals that modify after the tannin treatment can be explained by simple overlapping of the wood and externally added tannin.

OPEN IN VIEWER

Fig. 1.

ATR and baseline-corrected and normalized FT-IR spectra of untreated and 2.5, 5, and 20% tannin-treated Scots pine samples.

Table II shows the bands that change after tannin impregnation. No evidence of new covalent bonding between the tannin polymer and the wood surface can be observed. This information led to further analyses and an interest in defining the threshold for which the tannin impregnation level is still detectable. Figure 1 depicts the capacity of the ATR unit of the FT-IR spectrometer to detect very small concentrations of tannin. Proportionality can be revealed between the concentration of the tannin solution and the peak at around 1610 cm⁻¹. The importance of this peak was also noted for beech and spruce wood (Fig. 2). Notwithstanding, the signal at 1454 cm⁻¹ increases for the tannin-treated samples, confirming that this wavenumber also gives information related to the presence of wood preservative on the surface. The bands at 1270 and 1230 cm⁻¹ showed very clearly the increase in aromatic compounds and C=O groups ³⁷ of tannins on the surface for spruce, whereas for beech, these bands are overlapped by the absorbance of C–O vibration at 1242 cm⁻¹ from acetyl groups of hardwoods. ³⁹

OPEN IN VIEWER

Fig. 2.

ATR and baseline-corrected and normalized FT-IR spectra of untreated and 20% tannin-treated beech and spruce samples.

OPEN IN VIEWER

TABLE II.

Assignment of the modified bands after tannin treatment.

Wavenumber, cm−1	Wood component	Tannin–hexamine component	Changes in mid-IR spectrum	Remarks
1710		Hexamine activation, C=O stretching a	Small increase	Reaction between tannin and hexamine
1610	Lignin peak at around 1595 cm−1 aromatic skeletal stretching 35	Flavonoid aromatic stretching 36	Very high increase	Vibration superimpose
1454	Overlapping lignin 37 and carbohydrate signals 35	Aryl-ring vibrations25,38, changes in band height imply cross-linking reaction 30	Medium increase	Vibration superimposition
1270	Lignin: G ring plus C=O stretching 37	Antisymmetric C–O–C deformation 38	Small decrease	Vibration superimposition
1230	Lignin: C–C plus C–O plus C=O streching 37	Antisymmetric C–O–C deformation31,38	Small increase	Vibration superimposition

a

G. Tondi, Department of Forest Products Technology and Wood Constructions, Salzburg University of Applied Sciences, unpublished data, 2013.

Fourier Transform-Raman . A similar study was run using FT-Raman (Fig. 3). We found that the tannins triggered a very significant effect of fluorescence in the treated wood samples. A proportional increase of fluorescence in the Raman spectra was observed.

OPEN IN VIEWER

Fig. 3.

FT-Raman spectra without data treatment of Scots pine untreated, 10% tannin-treated, and 20% tannin-treated samples compared with the pure tannin–hexamine polymer.

After baseline correction, it was possible to focus on the region between 1800 and 700 cm⁻¹; however, there is still fluorescence in the FT-Raman spectrum of 20% for the treated Scots pine samples. The spectra in Fig. 4 show the importance of the signal at 1600 cm⁻¹. This aryl stretching peak at 1600 cm⁻¹ increased after tannin impregnation. Also, a clear change in the Raman intensity was seen at around 1660 cm⁻¹, corresponding to an unsaturated molecule (C=C) with a carbonyl group (C=O).^40,41 Kihara et al. ⁴² concluded that the band around 1660 cm⁻¹ is strongly influenced by the ring-conjugated α,β-unsaturated bonds in lignin. A slight change is also seen at 780 cm⁻¹, where the intensity for tannin polymer in beech, pine, and spruce wood after impregnation increased. However, all spectra showed a peak around this Raman shift; therefore, this band cannot be used to evaluate the chemical changes due to tannin impregnation. The low-frequency region below 900 cm⁻¹ of the Raman spectrum is difficult to assign to functional groups of wood. ⁴⁰

OPEN IN VIEWER

Fig. 4.

Baseline-corrected FT-Raman spectra of Scots pine untreated, 10% tannin-treated, and 20% tannin-treated samples compared with the pure tannin–hexamine polymer.

Despite the technical problems due to fluorescence, FT-Raman still indicates the influence of tannin on the spectra through detection of the aromatic ring at 1600 cm⁻¹.

Fourier Transform Near-Infrared . From the FT-IR measurements, we observed that the band at 1610 cm⁻¹ corresponded well with C=O stretching in carboxylates of mimosa tannin and that it overlapped the aromatic skeletal vibrations and breathing of lignin at around 1595 cm⁻¹. The NIR spectra as well as the second derivative of NIR spectra also showed changes in the band region 6020 and 5950 cm⁻¹, after the impregnation of tannin in the Scots pine wood samples (Fig. 5).

OPEN IN VIEWER

Fig. 5.

(a) Second derivative of FT-NIR spectra from 5700 to 6100 cm⁻¹ of Scots pine untreated and 20% tannin-treated samples compared with the pure tannin–hexamine polymer. (b) Second derivative of FT-NIR spectra from 4500 to 5700 cm⁻¹ of Scots pine untreated and 20% tannin-treated samples compared with the pure tannin–hexamine polymer.

Mitchell and Schimleck ⁴³ detected a very good correlation between absorbance values of the band at 1595 cm⁻¹ from mid-IR spectroscopic measurements and the intensity values from the band around 5995 cm⁻¹ of the NIR spectra of hardwood samples, which is the first overtone of aliphatic and aromatic C–H stretching vibration.^44,45 Schwanninger et al. ⁴⁶ concluded that this band at around 5970 cm⁻¹ might also arise from involved vibrations of the aromatic groups in softwood lignin. Similarly, the tannin polymer presented a band at around 4690 cm⁻¹, and this band could be also associated to the combination of C–H stretching and C=O stretching of the acetyl groups. ⁴⁶

Moreover, the first overtone of O–H stretching vibration can be observed for the amorphous region of cellulose at 7000 cm⁻¹,^46,47 and for the phenolic OH groups around 6900 cm⁻¹.^47–49 The tannin resin showed an absorption band at about 6960 cm⁻¹. This band influenced detection of the OH groups of cellulose band at 7000 cm⁻¹. ⁴⁶ Such a phenomenon could be observed in all wood species tested.

Multivariate Methods for Analysis. Cluster Analysis for Fourier Transform Infrared. Figure 6 shows a hierarchical cluster analysis that classifies the FT-IR spectra into various groups. The higher heterogeneity level was obtained for the difference between hardwood and softwood species resulting from their distinctions between the chemical components of the wood (e.g., lignin and polyosen). The beech samples can be further differentiated between treated and untreated specimens and also by the difference in tannin concentration, even if they have a small heterogeneity value. The untreated spruce and pine samples were classified into one group and showed distinct differences between the tannin-treated samples. The softwood samples can be differentiated according to the concentrations of tannin on the surface, with low heterogeneity values. The highest tannin concentration (20%) leads to different results. A detailed analysis of the second derivative of the IR spectrum of 20% tannin-treated spruce samples shows greater deviation from the spectrum of 20% tannin-treated pine samples. The distance of the spectra between 20% tannin-treated spruce and pine wood samples is too high, and the cluster algorithm distinguish these samples. This spectrum of the 20% tannin-treated spruce sample also shows differences compared with the 10% tannin-treated spruce samples, and this behavior cannot be explained via tannin treatment alone. Therefore, it seems that the IR spectrum of this sample is unusual compared with that of the other tannin-impregnated spruce samples.

OPEN IN VIEWER

Fig. 6.

Cluster analysis of FT-IR spectra of samples from three tannin treated and untreated wood species.

In summary, FT-IR is able to classify (1) hardwoods and softwoods, (2) treated and untreated samples, and (3) low concentrations of tannin on the surface. The chemical differences caused by the tannin treatments are more significant for the cluster than the differences between softwoods, whereas the influence of low concentration of tannin (2.5, 5, and 10%) is the last parameter that can de differentiated. However, if the concentration of tannin is high (20%), then the cluster analysis was not able to classify the tannin-impregnated samples correctly.

The cluster branching calculated for FT-Raman spectra was comparable to that for FT-IR spectra, whereas a smaller heterogeneity of the FT-Raman data could be observed.

A different behavior of the clustering was obtained for the FT-NIR data (Fig. 7), for which the highest difference was registered for treated and untreated hardwood and softwood samples; therefore, the sensitivity of this technique on externally added tannin can be observed. However, if the concentrations of tannin (2.5, 5, and 10%) on the surface are low, the heterogeneity of the FT-NIR data also is very low, and the Ward algorithms of the cluster analysis are not suitable for the correct classification of the Scots pine samples. Nevertheless, the FT-NIR data provide a classification of the samples into their family as well as their wood species if the tannin concentration on the surface is a maximum of 10%. The penetration depth of FT-NIR radiation is higher than that of FT-IR. ⁴⁶ Therefore, it can assumed that FT-NIR provides information from inside the samples, where the proportion of the tannin decreased and the proportion of untreated wood increased. This sensitivity was exploited to verify the correlation between concentration of tannin solution applied and the intensity of the NIR signal for separate wood species.

OPEN IN VIEWER

Fig. 7.

(a) Cluster analysis of FT-NIR spectra of samples from various tannin treated and untreated wood samples. (b) Vector normalized and second derivative of averaged FT-NIR spectra from 4000 to 8000 cm⁻¹ for the pretreatment of the cluster analysis.

Partial Least Squares Method . Figure 8 shows the measured versus the predicted tannin dry retention of the pine wood samples. A full cross-validation was performed for every wood species. For Scots pine wood samples (n = 30), the coefficient of determination (R²) was 75% and the RMSECV was 6.1% with five principle components (PCs). PLSR models for beech wood were calculated with 25 samples, resulting in R² = 93% and RMSECV = 1.3% with five PCs. For Norway spruce samples (n = 22), the R² = 91% and RMSECV = 1.4% with two PCs. The sample amount was small and varied for each of the wood species; therefore, the results are not usable for a comparison of the PLSR models for the various wood species. However, these results demonstrate the potential of FT-NIR to estimate the surface concentration of tannin for the tested wood species. Moreover, the relationship between tannin impregnation and FT-NIR spectra demands further consideration.

OPEN IN VIEWER

Fig. 8.

Predicted versus measured values of the PLSR model to estimate the tannin dry retention of the impregnated Scots pine samples with the FT-NIR data.

The Scots pine wood samples were impregnated with various solutions of 2.5 to 20% tannin concentrations (Table I). The data of the tannin dry retention determination showed dependence on the tannin solutions and the amount of tannin in the wood samples. Tondi et al. ¹⁴ concluded that the penetration of flavonoids in Scots pine occurs longitudinally through tracheids with open-bordered pits and across radial direction through parenchyma rays, whereas beech is almost exclusively penetrated in the longitudinal direction through large and easily accessible vessels. Therefore, it could be assumed that the tannin impregnation in the wood is not uniformly distributed in the samples with dimensions of 25 mm ³ × 25 mm³ × 25 mm³. The highest tannin amount is on the surface and with increasing penetration depth, the tannin concentration decreases. Figure 8 shows that with higher tannin retention of the samples, the differences between the real dry tannin retention values and the regression model increased, and a heteroscedasticity in the residue can be seen. The dry tannin retention was measured gravimetrically, and the difference in tannin uptake between cross-section and radial and tangential surfaces was not determined. Therefore, these PLSR models might be influenced by the nonhomogeneous distribution of the tannin within the cross section and the different penetration depths of the tannin in the samples. To improve these regression models, further measurements of smaller samples (e.g., 10 mm³ × 10 mm³ × 10 mm³) can balance the tannin distribution in the samples to minimize the inhomogeneity of the sample thickness.

CONCLUSIONS

This study demonstrated the efficacy of three vibrational spectroscopic techniques to analyze tannin-treated wood surfaces. Each technique can be used to extract different information about the tannin impregnations. Mid-IR spectroscopy allows the detection of the tannin in the treated surfaces when the concentration of the impregnation solution is very limited (2.5%). The peak at around 1610 cm⁻¹ is the peak that most clearly shows the presence of externally added tannins on the surface due to the aromatic chemistry of the flavonoids. The high performance of the analysis of characteristic bands in the spectra of FT-IR was not confirmed by cluster analysis. The results of a cluster analysis indicated a high spectral heterogeneity between the impregnated wood samples and wood species. However, if the concentration of tannin is high (20%), the cluster analysis was not able to classify the tannin-impregnated softwood samples correctly.

FT-Raman is also quite sensitive to tannin. After baseline correction, the aromatic peak at 1600 cm⁻¹ and the increase in the aryl stretching peak are presented as important Raman shifts.

FT-NIR was less successful in the analysis of characteristic bands in the spectra, but it showed good sensitivity to investigate small differences of externally added tannin after cluster analysis and PLSR. These results demonstrate that a classification and estimation of tannin dry retention in various wood species are possible with fast, nondestructive measurement methods.

These findings provide a basis for further investigations of the estimation of externally added tannin content in wood. Moreover, they could serve a basis to establish guidelines for quality assurance of the tannin impregnation process and to estimate the service life performance of the tannin-impregnated wood and also aid in the transfer from laboratory to industrial conditions for consumer applications.