FTIR analysis of chemical changes in wood induced by steaming and longitudinal compression

Four types of samples were prepared from both beech (Fagus sylvatica L.) and sessile oak (Quercus petraea (Matt.) Liebl.) wood species by implementing following treatment procedures: The raw material used for samples preparation originated from the forest of the Sopron region, Hungary. The material was preselected to include only pieces of heartwood from the same trunk, assuring straight grain orientation, consistent and uniform annual ring structure parallel to the sides and without knots or any other wood defects. At least two replicas for each sample type were prepared. The initial dimension of green samples was 20 × 20 × 200 mm³ (Radial × Tangential × Longitudinal, respectively). Each wood piece was cut in half along the RL direction to create two twin samples with dimensions of 20 × 7 × 200 and 20 × 10 × 200 (R × T × L) mm³ (Fig. 1). The first twin piece was considered as an untreated reference while the second was subjected to one of the modification processes listed above. Such a procedure of samples preparation allowed direct comparison of treatment effects with untreated samples, considering perfect matching of corresponding surfaces, therefore minimizing variability of results due to sample provenance and within tree variations.

All treated samples were plasticized for 1 h by steaming in a saturated steam at a temperature of 100 °C. One third of these were taken apart and served as steamed samples. The remaining wood was longitudinally compressed with 20% compression ratio (relative to the sample length) in a custom-made accessory integrated with a universal material testing machine Instron 4208 (Instron Corp, USA). The compression velocity was 50 mm/min and was performed with plasticized samples immediately after steaming. Half of the compressed samples (compressed and released) were kept fixated at their compressed state in the experimental stand for 1 min to allow natural relaxation of mechanical stresses induced by the treatment (Báder and Németh 2018a). The time of fixation was set in the control software of the used material testing machine. The other half of the samples (compressed and fixated) were kept compressed in the experimental stand for an additional 18 h after compression process. In that case, the wood sample cooled down in the semi-sealed chamber, minimizing moisture flow and, consequently, avoiding wood drying. All samples were conditioned (to constant mass) before further processing in a dark chamber at a temperature of 20 °C and 65% RH to avoid photodegradation or any other post-process deterioration. The equilibrium moisture content (EMC) of wood samples after different treatments is summarized in Table 1.

Small specimens were cut from the treated and reference wood samples to dimensions of 8 × 5 × 40 (R × T × L) mm³. Such a specimen size eased evaluation of the physical–chemical properties by means of the infrared spectroscopy. The cutting pattern was optimized to generate eight specimens for each replica (or six in the case of compressed and fixated). The scatter of infrared radiation from the measured surface is highly dependent on its smoothness (Tolvaj and Faix 1995); therefore, special care has been dedicated for proper specimens preparation. A sharp circular saw with high cutting and small feed speeds was used to avoid form deviations and minimize generated surface roughness. Each specimen was labeled with a unique code to allow tracing and referencing for further statistical analysis. The wood specimens were conditioned after processing for 48 h to uniform moisture distribution and surface oxidation state.

Infrared spectroscopy has been selected as an analytical method for assessment of physical–chemical changes to wood samples exposed to steaming and compression. The instrument used was a Jasco FT/IR 6300 Fourier Transform Infrared spectrophotometer (Jasco Inc, USA) equipped with a Diffuse Reflectance Infrared Fourier Transform (DRIFT) accessory, which illuminates the measured surface and, at the same time, collects the reflected light converting it into a spectrum. This technique is used for the characterization of solid samples or powders (both organic and inorganic) having a coarse surface, such as textiles, wood, bones or paper. Particle size, homogeneity and packing density of powdered samples highly influence the quality of spectrum; therefore, special attention should be directed to sample preparation and presentation (Sandak and Sandak 2018). In our case, the most important was to assure constant surface roughness in order to compare treatment effect. DRIFT spectra can result in an increase in resolution, compared to Attenuated Total Reflection (ATR) technique, but the results are influenced by the sample size differences, similar to the problems that occur during KBr pellet production (Beasley et al. 2014). Nevertheless, integration of sophisticated spectrometric measurements, such as a DRIFT spectroscopic technique, along with multivariate analysis method, is the most promising solution for characterization of ligno-cellulosic materials (Toivanen and Alén 2006).

Assessed specimens were presented to the instrument, assuring perpendicular wood fiber direction to the optical axis of the illuminating light. The longitudinal-radial surface of the wood was exposed to illuminating light in order to measure an average reflectance of the area covering several annual rings. The mid infrared spectral range included wavenumbers from 400 to 4000 cm⁻¹, which corresponds to light wavelength from 25 µm to 2500 nm. The spectra were acquired in reflection mode with a resolution of 1 cm⁻¹ and expressed as an absorbance. Sixteen consecutive spectra were averaged at each measurement point in order to increase signal-to-noise ratio. The surface of each specimen was measured 5 mm from both ends at the radial section. Therefore, 16 independent spectra were collected for each batch, separately for reference and treated twins. The only exception was compressed and fixated batches where only 6 specimens were cut due to excessive shortening of the original samples that remained as a result of the modification process. In total, 176 spectra for modified wood and 176 for twin references were collected in both wood species investigated.

The infrared spectrum collected from the surface of wood in reflection mode is a superposition of light absorbance by functional groups of wood polymers (chemical information) and light scatter (physical information). Unfortunately, it is impossible to separate both signals from the spectrum as measured by DRIFT. Nevertheless, it is necessary to pre-process spectra before any systematic study and interpretation. The algorithm for data analysis as implemented in this research is summarized in Fig. 2. Both wood species (beech and oak) were separated at the initial stage in order to eliminate variability within the data due to species. Extended multiplicative scatter correction (EMSC) was applied here to minimize an effect of nonuniform optical density for measured wood specimens and related differences in the measured amplitude and baseline deviation (Gallagher et al. 2005). EMSC is a model-based pre-processing algorithm, capable to correct physical and chemical features in IR reflectance. It provides baseline correction and normalization as well as separation and quantification of chemical and physical variations in the vibrational spectra, making the subsequent calibration models simpler and statistically more robust (Panero et al. 2013). Although it loses some information, it estimates mathematical parameters instead, which can be used to study the different effects separately (Afseth and Kohler 2012). The specific settings of EMSC implemented in this research were: mean spectrum as a reference, order of polynomial = 2, correction method by Classical Least Squares and T-squared limit = 0.99.

Purposefully, no specific baseline correction was applied to the data at this stage of analysis to preserve original shape of the spectral curves and maintaining contributions of absorbance and scatter signals. Such adjusted spectra were then sorted into four classes according to treatment type and intensity. Principal components analysis (PCA#1) was performed on that data set in order to identify clustering of specimens originated from different classes. In addition, ANOVA testing of the statistically significant differences between spectra collected from samples after different treatments was performed for oak and beech separately. As a result the grouping of spectra was evidenced and summarized in Table 2, assuming significant differences at p < 0.05. The software used was custom developed in LabView 2018 (National Instruments, USA) and Matlab (Mathworks, USA), by implementing a multiple comparison algorithm according to Bonferroni method.

Variability observed in the sorted spectra was still too high to perform direct spectra confrontation. For that reason, all spectra belonging to each batch (treatment class) were averaged to assure maximum homogenization and generalization of the spectral representation of specimens. Additional analysis (PCA#2) was performed on that limited data set to highlight spectral features differentiating all classes. For that reason, PCA loadings were extracted and analyzed. The averaged spectra were also directly compared, and peaks interpretation was performed using literature references. Both variation of the amplitude and shift were visually assessed by the panel of experts separately for each peak, with all observations systematically summarized in Table 2. The raw and pre-processed IR spectra of measured specimens can be found in https://​doi.​org/​10.​6084/​m9.​figshare.​12024840 (Báder et al. 2020).

Part of the spectra (2800 to 3750 cm⁻¹) corresponding to hydroxy functional groups (–OH) was cropped from the averaged spectrum in order to deconvolve the spectra outline and list all peaks involved. Spectral deconvolution (or peak separation) is a method to identify peak composition in a complex spectrum, which is supposed to be a combination of multiple peaks. It is a case of IR spectroscopy where several broad and overlapping –OH peaks are located in a part of the spectrum scrutinized from the surface of wood. Shape of each peak was approximated as a Gaussian curve with the specified central position and full width at half maximum (FWHM). A single baseline correction was applied to all spectra/batches before cropping, including quadratic polynomial fitting of the extreme (minimum) spectral points at 1028, 1900 and 3812 cm⁻¹. The deconvolution algorithm was optimized based on the probabilistic hybrid local search algorithm that was custom developed and implemented in LabView 2018 (National Instruments, USA) software. The custom algorithm identified an optimal set of peak locations, heights and widths, that were required to correctly reconstruct the original spectrum by superposition of convolved curves. The best software performance was achieved after providing initial guesses for all optimized variables. These were determined manually by following literature references. The objective of the software was to adjust these variables for each peak in the recurrent way, assuming minor but random variation of a single parameter at once. The loop was finalized after the RMS error of the difference between original and fitted outline of the spectrum was not reduced for at least 20 consecutive iterations. In that way it was possible to compromise between optimization procedure (machine learning by software) and scientific foundations for the deconvolution of this complex spectral band. The change of area under the peak as well as its shift were considered as identifiers for treatment intensity. Root mean square difference between original and deconvolved spectra was used as the quantifier of deconvolution success as an alternative to the calculation of reduced chi-squared value for the deconvolved curves (Popescu et al. 2007; Donaldson et al. 2010).

The software used for spectra processing and analysis, including EMSC, averaging as well as PCA, was PLS_Toolbox (Eigenvector Inc, USA), available as an extension of the Matlab package (Mathworks, USA). Cross validation of the chemometric models was performed with the venetian blinds approach. In that method all source spectra are divided into s splits of a given number b of objects in each blind. The test set is determined by selecting every s^th blind of spectra from the original dataset. The number of splits was s = 10 with the blind width b = 1. It has to be mentioned that the part of the spectrum ranging from 2600 to 1900 cm⁻¹ was intentionally ignored as it does not contain any relevant spectroscopic information.

A great advantage of infrared spectroscopy is the possibility for direct interpretation of the spectrum as light absorbance is very selective for different molecules and functional groups present in wood. At the same time, wood is highly anisotropic and has a complex polymer matrix where different components interact with each other. Another critical factor is the high hygroscopicity of wood that tremendously affects its spectroscopic characterization. A lot of research efforts have been directed toward systematizing the infrared peaks interpretations that resulted in several scientific publications (Faix 1991; Schwanninger et al. 2004; Popescu et al. 2007, 2006, 2009; McLean et al. 2014; Guo et al. 2018; Moosavinejad et al. 2019). These papers were an inspiration to review the available know-how and adopt it for the specific needs of this research (thermo-hydro-mechanically treated hardwoods). The complete list of peaks observed in investigated experimental specimens is summarized in Table 2.

EMC varied from 12.3% and 12.6% in the case of reference samples to 11.8% and 12.3% with compressed and fixated wood for beech and oak, respectively. Although the differences are small, the EMC of each sample group was altered by the thermo-hydro-mechanical treatments. Oak possessed most steady structure as its change in EMC was statistically significant only after compression treatment and fixation (Table 1). Changes of EMC in beech were significant already after the steaming process. Its cell lumens are not clogged with tyloses (as in oak) that easies washing effect induced by the steaming treatment.

An important difference between chemical content of beech and oak species lays in composition of hemicelluloses. The share of xylan extracted from hydrolysed beech wood is more than double of the same polymer extracted from oak, assuming the same mass loss. Xylose corresponds therefore to 77% of the total monosugars in beech, compare to 49% extracted from oak (Fišerová et al 2013). It was evidenced that xylan is the least stable polysaccharide when exposing various hemicelluloses (including P-glucan, arabinogalactan, arabinoxylan, galactomannan, glucomannan, xyloglucan, and xylan) to thermal decomposition in inert atmosphere (Werner et al. 2014). The variation of the hemicellulose composition in both species contributes even more to discrepancy of the steaming effects considered here as a mild thermal treatment of wood. Even if the total mass loss caused by this treatment is negligible, some chemical alteration of constitutive polymers (such as dehydratation, deacetylation, re-configuration, cross-linking) occurred. The hemicelluloses are the first structural compounds affected even at relatively low temperatures (Sandak and Sandak 2013). The degradation starts by deacetylation causing releasing of acetic acid. Acetic acid acts as a depolymerization catalyst, leading to further polysaccharide decomposition. Simultaneously hemicelluloses undergo dehydratation having effect in decrease of hydroxyl groups (Esteves and Pereira 2009). It will be evidenced below when discussing infrared reflectance spectra, that such chemical changes were more pronounced in beech than oak, confirming observations of changes noticed in EMC.

Compression resulted in the highest alterations in EMC of beech. It can be explained by variations in the number of available sorption sites (such as -OH groups) caused by the rearrangement of constitutive polymers due to mechanical action (extensive compression) in combination with the proceeding mild thermal treatment (steaming).

Figure 3 presents a set of averaged IR spectra collected from beech and oak wood specimens studied in this research (reference, steamed, compressed and released, compressed and fixated). Identified spectral bands are marked and numbered following the literature references as listed in Table 2. A summary of spectral analysis and interpretations induced by visual assessment are also summarized in the same table. Vertical arrows indicate the intensity of alterations (increase or decrease) while changes in the peak position (shifts) toward higher or lower wavenumber are represented by horizontal arrows. In all cases, treated wood spectra were compared to averaged untreated wood (reference) of the corresponding species.

Apparent differences between IR spectra of reference and modified woods are more pronounced in beech than oak, which indicates higher susceptibility of this species for chemical-physical changes induced by the pleating process. It was confirmed by ANOVA testing used for evidencing of the statistical differences between infrared absorbance induced by different treatments. Steaming seems to be a relatively mild process only slightly affecting chemical changes of the wood constitutive polymers. In the case of beech, changes in the functional groups induced by steaming are noticeable for several bands; however, the intensity of alterations is evidently lower than influence of compression and fixation treatments. The most noticeable peak drop as well as shift toward lower wavenumber, resulting from steaming processes, was noticed for band 4 (996) and 5 (1044) assigned to C–O valence vibrations, C–O stretching, C = O stretching as well as to aromatic C–H in-plane deformations. Lowered reflectance values signify the functional groups degradation (polysaccharides) and reduced cross-linking of lignin occurring in these particular bands.

Another drop in spectra reflectance can be observed in band 6 (1086), assigned to C–O deformation in secondary alcohols and aliphatic ethers. Bands 18 (2902) and 19 (2940), linked with symmetric CH2 valence vibrations and asymmetric stretching vibrations of CH related to methyl and methylene in lignin, cellulose and hemicellulose, are also lowered but only after compression and compression followed by fixation. On the contrary, several IR bands exhibit increase of reflectance resulting from the modification processes. This is especially noticeable for bands 1 (670), 2 (834) and 8 to 14 (1147–1505). It can be associated to the condensation of lignin occurring after glass transition induced by elevated temperature in combination with mechanical stresses. Corresponding peaks in the IR spectra of oak wood do not exhibit pronounced visual differences. Moreover, some bands, in the case of oak, seem to not be affected at all by the thermo-hydro-mechanical (pleating) process (e.g., 15 (1598) and 16 (1652) being assigned to C = O stretching and aromatic skeletal vibrations).

It has also been observed that noticeable deviations of IR spectra occurred in the range covering the hydroxy functional groups of wood polymers (peak 20 to 25 (3149–602)), which suggested changes of the hygroscopic properties of treated specimens. Dedicated analyses were therefore performed for these particular spectral bands that are reported in a subsequent chapter.

Principal component analysis (PCA) is a powerful chemometric method to reduce highly correlated and multi-dimensional data set to lower dimension. It searches for unique properties of spectral data and separates these into groups of peculiar similarities. Two PCA analysis were performed for the needs of this research. In the first, the whole range of spectra collected from measurements of singular specimens was analyzed, focusing on within batch variation (PCA#1). The second analysis was performed on averaged spectra for each batch, providing a much more generalized spectroscopic description of the studied materials (PCA#2). The data were mean centered before analysis by subtracting the average absorbance spectrum. This pre-processing is a routine procedure performed before PCA computation to minimize the effect of a few dominating peaks on the overall chemometric model.

Results of PCA#1 performed on the spectra of beech and oak wood are presented in Fig. 4. It is evident that treatment batches are better separated in the case of beech where a sequence from reference to compressed and fixated data points follow the PC1 axis direction. This component is dominant in the model and explains nearly 94% of the variance within data. The cluster of reference is clearly separated, confirming hypothesis of the profound effect of steaming and compression on the physical–chemical configuration of beech wood. The cluster of steamed wood slightly overlaps compressed samples and lays around the centre of the PC1. Both compressed sample types are partially overlapped, but some discrimination can be noticed when considering PC2 scores

The spectra corresponding to different batches of oak wood overlapped, creating a single cluster mixing reference, steamed and compressed specimens (Fig. 4). Only compressed and fixated spectra were, to some extent, detached from the rest, creating a separate cluster that slightly overlapped with compressed and released. The first principal component of PCA#1 for oak explains almost 60% of variance while the second more than 20%.

Additional analysis (PCA#2) was performed on the averaged spectra of beech specimens. In that case, each batch was homogenized by computing the average value of absorbance for all wavenumbers, resulting in two data points per treatment type and six points for corresponding reference batches. The resulting score plot is presented in Fig. 5. Improved separability is evident compared to the cloud of data points shown in Fig. 4. Again, first principal component (PC1) separates the reference spectra from those modified, following treatment intensity. Steamed specimens are grouped in a single cluster, clearly separated from other treated woods by means of the second principal component (PC2). Averaged spectra collected from both compression treatments (released and fixated) overlap but, at the same time, are distant from the other clusters.

A great advantage of PCA is the possibility for identification of spectral features supporting discrimination. Two loadings corresponding to PC1 and PC2 are shown in Fig. 5, together with the reference position of identifiable spectral peaks (Table 2). First principal component discriminated reference (negative values) from all the treated samples (positive values). The most noteworthy peaks recorded in the corresponding loading are 8, 9, 13, 14, 16, 17, 21–25 (1174, 1270, 1458, 1505, 1652, 1740, 3280–3602). It reveals that most affected polymers due to treatment are lignin (C–O, C = O, CH₂), cellulose (CH₂) and all hydroxyl groups affecting hygroscopic properties of wood. Low contribution of PC1 to the steamed wood spectra is related to moderately low absolute values of scores. It suggests the intermediate extent of changes, still indicating more similarity to compressed samples than to untreated wood (positive sign of scores). The second principal component allowed discrimination between steamed and compressed samples, where contribution of peaks 7, 8, 14, 17, 18, 19 (1132, 1174, 1505, 1740, 2902, 2940) are most dominant. It corresponds to the spectral changes particularly in lignin (CH, C–O, C = O, CH₂), condensation of guaiacyl as well as in hemicellulose (CH, C = O, CH₂). Information recorded in both principal components corresponds to that interpreted by visual assessment with the exception of peak 17 (1740), possessing a high value in PC2 loading. However, the great importance of hydroxy groups (peaks 21 to 25 (3280–3602)) is evidenced, suggesting alteration of water sorption mechanisms in specimens after thermo-hydro-mechanical treatments.

As stated above, the extent of changes recorded in the FTIR of oak was much less profound than of beech. It was confirmed by PCA#2 for oak (not reported here) where discrimination between clusters corresponding to different treatments was limited.

Substantial changes to the FTIR spectra due to thermo-hydro-mechanical treatments are assigned to the modification of wood hygroscopic properties. Even if EMC variation was not very high in specimens after conditioning (Table 1), many differences were statistically significant (ANOVA p < 0.05). Significant difference was detectable between all of the differently treated beech specimens and between oak reference and fixated specimens. Similarly, the outline of the IR spectra in the range corresponding to the absorbed water and –OH groups in general (3700 to 3000 cm⁻¹) was evidently altered following intensity of treatment, especially in the case of beech (Fig. 3). A great limitation of the direct spectra interpretation in that range is presence of several peaks that are broad and highly overlapped. For that reason, a special software has been developed for deconvolution of spectral peaks, assuming that variations of the deconvoluted peak width and its maximum shift are restricted and controlled.

For the needs of this research, six -OH peaks were identified following literature references summarized in Table 2. Each of these are linked to different wood polymers and involved in uptake of moisture at different sorption stages. The result of compressed and fixated spectra deconvolution is shown in Fig. 6 for both beech and oak specimens. Areas and shifts for peaks of interest are plotted as a bar chart to highlight mechanisms of hygroscopic changes to the wood due to treatment. Even if the IR spectra outlines appear to be similar for both studied species, some slight differences are evidenced in the deconvolved peaks. The highest absorbances were noticed for peaks 21 (3280), 22 (3375) and 23 (3496) although the areas below these are rather constant and do not change due to treatments. On the contrary, peak at 24 (3565) (assigned to weakly absorbed water) decreases along treatment intensity. Such change is more monotonic in the case of oak than beech. The peak at 25 at 3602 cm⁻¹ (weakly H-bonded water) follows the opposite trend and its area increases with treatment intensity. Variations of the peak area are associated to peak shifts. It is especially noticeable in the case of beech wood where three peaks (23 (3496), 24 (3565) and 25 (3602)) are monotonously shifted as related to the reference. This was not that evident in the case of more resistant (to the pleating process) oak wood.

The summary of above observations leads to the conclusion that absorbance of the hydroxyl functional groups associated to the weakly bonded wood increases noticeably following treatment intensity. On the contrary, the strongly bonded water is relatively stable and not affected by the treatment. It confirms the trend noticed with visual assessment of spectra (Table 2) as well as alterations of EMC.

It has to be mentioned that the thermo-hydro-mechanical treatment induces morphological changes to the wood that may affect the vapour transport within bulk (diffusion) and in the border of the wood and surrounding air (absorption). Even if it can not be directly evidenced by infrared spectroscopy, the water sorption sites availability changes also due to mechanical re-configuration of constitutive polymers (translocation and partial densification) in combination with micro/nano cracking of cell walls.