Experimental study on glued laminated timber beams with wellknown knot morphology
 486 Downloads
 1 Citations
Abstract
Nowadays, the impact of knots on the failure behaviour of glued laminated timber (GLT) beams is considered by subjecting the single lamellas to a strength grading process, where, i.a., tracheid effectbased laser scanning is used to obtain information about knot properties. This approach singlehandedly defines the beam’s final strength properties according to current standards. At the same time, advanced production processes of such beams would allow an easy tracking of a scanned board’s location, but, at this point, previously obtained detailed information is already disregarded. Therefore, the scanning data is used to virtually reconstruct knot geometries and group them into sections within GLT beams. For this study, a sample of 50 GLT beams of five different configuration types was produced and tested under static fourpointbending until failure. As for each assembled lamella the orientation and position within the corresponding GLT beam is known, several parameters derived from the reconstructed knots can be correlated to effective GLT properties. Furthermore, the crack patterns of the tested beams are manually recorded and used to obtain measures of cracks. A detailed analysis of the generated data and their statistical evaluation show that, in the future, dedicated mechanical models for such timber elements must be developed to realistically predict their strength properties. A potential approach, using fluctuating sectionwise effective material properties, is proposed.
1 Introduction
Wood is a naturally grown material and therefore is gaining popularity for its sustainability as well as its ecological and often also aesthetic advantages over other construction materials. However, the natural growing process also leads to fluctuations in both morphological features and mechanical properties. The morphology is dominated by the fibre course, which in turn is strongly influenced by knots (Phillips et al. 1981). On the microscale, wood shows an orthotropic material behaviour: the stiffness and strength values are orders of magnitude higher in longitudinal direction of the fibre than in the radial or tangential direction. Since the material orientation follows the fibre course, complex stress and strain fields are observed within naturally grown wood even under simplest loading conditions (Hankinson 1921; Goodman and Bodig 1978). Thus, the morphological fluctuations have direct impact on the mechanical behaviour of wood.
The impact of knots and other growth irregularities within mechanical systems obviously depends on the dynamic boundary conditions and the length scale of the corresponding system. The concept of GLT aims at a homogenisation of the material and, in that way, a reduction of the negative aspects of the natural growing process. Nonetheless, for predicting the structural strength of wooden structures, the knot morphology, particularly in the regions exposed to tension stresses, plays an important role.
One of the first more advanced models for the strength of GLT was presented by Foschi and Barrett (1980). Therein, each lamella was divided into equally sized cells, which were assigned a random mass density and knot diameter, from which the stiffness and strength were estimated through regression.
In Ehlbeck et al. (1985), a numerical model for describing the strength of GLT beams was presented. In Colling (1990), important parameters for the GLT bending strength are discussed and a statistical model for predicting the strength of GLT beams was developed. Furthermore, Colling (1995) investigated the relation between characteristic values of GLT bending strength, board tensile strength and finger joint strength. In accordance thereto, in Falk and Colling (1995) the lamination effect – describing the homogenisation of material fluctuations from individual timber boards to GLT structures—was investigated in detail and its ranges were calculated.
Serrano and Gustafsson (2007) discussed different fracture mechanicsbased approaches for assessing the strength of timber structures. The concepts are implemented into a probabilistic framework in the work of Danielsson and Gustafsson (2011), where a combination of Weibull weakestlink theory and mean stress method is proposed.
Brandner and Schickhofer (2008) investigated the relation between tensile stiffness and strength characteristics of single boards and bending strength of GLT. As result, requirements in regard to both strength and stiffness for the board material are proposed for specific GLT strength classes.
Also Frese and Blaß (2009) proposed a strength model for spruce GLT and simulated GLT beam tests to investigate the influence of single board and finger joint strength characteristics on GLT bending strength. Thereby, it was concluded that the requirements for boards and finger joints are insufficient to ensure the quality of strength classes according to EN 1194 (1999).
In Fink et al. (2013, 2014), bending tests on glued laminated timber beams with wellknown material properties are reported. Subsequently, Fink and Frangi (2013) presented a model for the bending strength of GLT considering the growth characteristics observed within timber boards. Also, Fink and Kohler (2014) presented a model for predicting tensile strength and stiffness of knot clusters in structural timber.
Recently, Blank et al. (2017) discussed two GLT models: One considering perfectly brittle material behaviour and one considering quasibrittle behaviour. Thereby, it is concluded that besides the Weibulltype size effect, which accounts for the variability of the material, a second size effect is influencing the load bearing capacity of timber. This second size effect is of deterministic nature and is characterised by the quasibrittle failure behaviour of glued laminated timber.
The work of Kandler et al. (2015) was based on an experimental study performed at Linnæus University, Sweden. While Kandler et al. (2015) covered the linear elastic behaviour, this work focuses on structural failure mechanisms observed during the experimental study. It is the aim of this work to relate (a) the stiffness profile information presented in Kandler et al. (2015) and (b) the knot geometries reconstructed using the approach presented in Kandler et al. (2016) to the failure mechanisms observed in the experimental study.
In Sect. 2, material parameters and methodologies are presented. Section 3 discusses the results and in Sect. 4 conclusions and final remarks are given.
2 Material and methods
2.1 Material
Grading class, number of laminations, distance between the supports L and height h of the different GLT types as well as the minimum, mean and maximum values of the observed bending strength f_{b} and the coefficient of determination R^{2} between maximum load bearing capacity and system stiffness
Type  Grade  Laminations  L (mm)  h (mm)  Min (f_{b})  Mean (\(f_b\))  Max (\(f_b\))  \(R^2(F_\mathrm {max}\ \mathrm {vs.}\ \varDelta F/\varDelta w)\) 

A  LS15/GL 24 h  4  2 340  132  28.5  39.7  52.5  0.24 
B  LS22/GL 30 h  4  2 340  132  35.9  49.3  61.2  0.71 
C  LS22/GL 30 h  7  4 140  231  33.4  45.1  56.3  0.31 
D  LS15/GL 24 h  10  5 200  330  24.8  31.0  35.7  0.15 
E  LS22/GL 30 h  10  5 200  330  31.8  42.4  58.8  0.36 
Since the study is devoted to the impact of knots on the mechanical behaviour, all beams were produced of full length lamellas. In this way, finger joints, which would have represented additional weaksections in the GLT assembly, could be avoided.
Norway spruce boards of grades LS15 and LS22—delivered by a sawmill located in Töreboda, Sweden—were used. The original dimensions were 35 mm height, 95 mm width and 5400 mm length. Each board was marked with a unique ID, so the measured data could always be linked to the corresponding board.
The used boards were inspected using a tracheid effectbased laser scanning device, yielding the fibre angle on all surfaces (Nyström 2003; Simonaho et al. 2004; Petersson 2010). For a more detailed description of the procedure it is referred to Olsson et al. (2013) and Kandler et al. (2015), where the fibre angle measurements have been used to obtain a stiffness profile for each board.
After inspection, all boards were sent to a GLT manufacturer for gluing using a MUFadhesive. Prior to gluing, all boards were planed to a height of 33 mm. Using the unique board number, the position of each board within the GLT assembly could be traced. The orientation of the boards was chosen to meet the Swedish standards for GLT production. That is, the boards were oriented such that the local pith of the topmost and bottommost boards would be located above and below of the beam, respectively. After gluing, the beams were planed to the final width of 90 mm.
The lamella strength classes T14 (LS15) and T22 (LS22) correspond to GLT strength classes GL 24h and GL 30h, respectively (EN 14080 2013).
2.2 Reconstructing the knot geometries within wooden boards
In a first step, a geometrical description of the knot morphology is determined by employing the automated approach described in Kandler et al. (2016). Therein, fibre angle measurements were experimentally obtained from laser scans based on the socalled tracheid effect, which describes the light propagation patterns in wood. From the fibre angle measurements, the knot areas are estimated. In combination with the pith location, all knots were reconstructed using rotationally symmetric cones. It has been observed that this algorithm also works well for small knots. In Fig. 1a–c, the results for an exemplary board are shown. In Fig. 1b, all recognised knots are displayed, whereas in Fig. 1c only the significant knot groups according to a modified version of the criterion presented in Kandler et al. (2016) are shown.
The modified criterion is based on omitting knots with an opening angle \(\alpha <1.5^\mathrm {o}\). Then, for all knots, the corresponding knot areas visible on the board surface are evaluated. Subsequently, all knot area values are sorted according to their value, resulting in a histogram. Then the value situated at the 70% quantile is chosen to be the critical value. All knot areas larger than this value are from now on defined as large knots, forming the 30% largestknots quantile. These knots are assumed to be mechanically significant and are retained. In case the mutual distance in longitudinal direction x of two such knots is below 200 mm, these knots are grouped together. It is assumed that smaller knots are not important for the mechanical model and are therefore omitted. However, in case a smaller knot is in close vicinity to one of the larger knots, it is retained in the model, since it is assumed to lead to interactions in terms of fibre angle course and failure modes. As measure for the vicinity, a distance value in xdirection of 100 mm has been observed to yield good results.
2.3 Stiffness profiles
In this study, two different types of stiffness profiles were assessed. A stiffness profile describes the fluctuation of the local modulus of elasticity E in the longitudinal direction x, i.e. \(E \equiv E(x)\). The procedure for the first type of stiffness profile is presented in Kandler et al. (2015), where fibre angle measurements obtained through laser scanning are directly used to transform the clearwood stiffness tensor. This approach is similar to the approach presented in Olsson et al. (2013), where the clearwood stiffness tensor was obtained by combining dynamic eigenfrequency measurements with a dynamic finite element analysis. Therein, the components of the stiffness tensor were calibrated so that the lowest eigenfrequency of the finite element model would match the experimentally obtained eigenfrequency. Subsequently, in each point the longitudinal stiffness component is extracted from the transformed stiffness tensor. Employing a nearestneighbour interpolation, the stiffness values are homogenised over each cross section, resulting in a stiffness profile for each board. In Olsson and Oscarsson (2017), it has been shown that the weakest value in such a stiffness profile correlates well to the strength of a single board.
In Kandler et al. (2015), the approach is based on the same principle of transforming the clearwood stiffness tensor according to the fibre angle. Therein, in addition to the inplane fibre angle, an empirical model for the diveangle was employed and the clearwood stiffness tensor was obtained from a micromechanical model based on a multiscale homogenisation procedure (Hofstetter et al. 2007). Based on knowledge of the stiffness profiles in combination with the position of each board, an accurate mechanical model for assessing the stiffness of GLT beams could be developed. In Kandler et al. (2015), a linear 2D finite element code was employed to predict the effective stiffness of GLT beams. In this study, on the other hand, it is investigated how well these data can be correlated to failure mechanisms in GLT.
Complementing the approach described in the previous paragraph, the reconstructed 3D knot geometries as described in the previous section are also considered. The 3D knot geometries of each knot group were used to compute the 3D fibre angle course within the volume of the board using the procedure outlined in Hackspiel et al. (2014) and Lukacevic and Füssl (2014). This procedure follows the socalled grainflow analogy (Foley 2003). Therein, the fibre angles in the longitudinaltangential plane are assumed to follow the trajectories of a fluid in laminar flow where knots constitute elliptical obstacles. For the diveangle, polynomials fitted to photographs of knot sections are used. Subsequently, the knot geometries and numerically obtained 3D fibre angles are used within a linear elastic 3D finite element analysis. Within this analysis, each knot group (in total more than 3000 groups) was loaded in tension to estimate its effective stiffness. At one end, the knot group is fixed in longitudinal direction and at the other end all nodes are exposed to a predefined displacement in longitudinal direction. The remaining boundary conditions were applied such that the specimen could freely expand in the other directions. From the resultant forces, the effective elastic modulus in longitudinal direction of each knot group could be computed. Based thereon, an alternative stiffness profile was developed, as can be seen in Fig. 1d. From Fig. 1d, it can be seen that for the knot sections a notable stiffness difference exists. The main reason is that the approach by Kandler et al. (2015) considers the local fibre deviation in each measured point but it does not consider interactions between knots. The 3D approach, on the other hand, inherently takes account of interactions between knots. Therefore, the 3D approach generally leads to lower stiffness values.
2.4 Strength profiles

Knotarearatio (KAR), which is the ratio between the knot area of all knots of a knot group projected on the cross section and the cross section area.

Knot area, which denotes the knot area visible on the surface of the board. In addition thereto, the weighted knot area distinguishes between topbottom and leftright sides of the board,

Knot volume.

Interfaces between knots and the surrounding wood matrix.

Foleyarearatio (FAR), which—in analogy to the KAR value— describes the ratio of the area disturbed by fibre deviations to the cross section area.
 1.
IP 1 \(\sim\) KAR + weighted knot area + knot volume + interface + FAR + interaction terms,
 2.
IP 2 \(\sim\) KAR + knot area + knot volume + interface,
 3.
IP 3 \(\sim\) KAR, and
 4.
IP 4 \(\sim\) KAR + knot area + knot volume.
Similar to the stiffness profiles, for each knot group the strength was estimated employing indicating properties 1–4. For the clearwood sections, the tensile strength is computed following the approach proposed in Hackspiel et al. (2014), which is based on scaling empirically based material properties according to results obtained by the micromechanical model for elastic behaviour.
2.5 Experiments
After testing, the geometry of all cracks occurred was manually recorded and documented for all sides of the 50 GLT beams. For cases where the point of crack initiation was not straighforward to identify, all possible crack initiation locations weres documented. In Fig. 4, the documented crack pattern is displayed exemplarily for one beam. Figures of all documented results for types A to E are available as electronic supplementary material, where within each type, beams are numbered from 1 to 10.
The recorded crack measurements were used to estimate the crack area \(A_\mathrm {crack}\). Assuming that all cracks range from one side of the beam to the other side of the beam, the crack area was computed according to \(A_\mathrm {crack} = L_\mathrm {crack} b/2\), where \(L_\mathrm {crack}\) is the accumulated crack length on all sides of the beam. For all beams, the width was b = 90 mm.
2.6 Simplified models for prediction of GLT stiffness and strength
All collected data was assembled in a database and statistically evaluated in the commercial software tool optiSLang\(^\mathrm {TM}\).
In addition, an approach of directly employing the stiffness profiles E(x) was pursued. This approach is inspired by the promising procedure presented in Olsson et al. (2013), Oscarsson et al. (2014), and Olsson and Oscarsson (2017). Therein, the minimum value of the stiffness profile E(x) of a board is assumed to present the weakest defect. Consequently, this value is used as indicating property for the effective strength of the whole timber board, resulting in remarkable prediction results with values of \(R^2 \approx 0.7\).
3 Discussion of results
3.1 System properties/effective properties of GLT specimens
From the plots shown in Fig. 5a, e, it can be seen that the two grading classes reach different maximum load peaks.
In accordance with EN 408 (2010), the system stiffness \(k = \varDelta F/ \varDelta w\) is computed from a linear regression of the load displacement curve in the range of \(0.1 F_{\max }\) and \(0.4 F_{\max }\), where a coefficient of determination \(R^2> 0.99\) is guaranteed. In Fig. 5b, d, f, the relation between system stiffness and maximum load are displayed. While a trend can be clearly observed, the values within each type are not very strongly correlated as the \(R^2\) values given in Table 1 suggest.
3.2 Encountered failure mechanisms
In Fig. 5, the load deflection curves \(F = F_\mathrm {left} + F_\mathrm {right}\) of all types are displayed. It can be seen that after an initially linear curve, 12 beams show nonlinear behaviour before the system load bearing capacity \(F_{\max }\) is reached. These nonlinearities are on the one hand cracks on the tension side, resulting in a spike in the loaddisplacement curve and, on the other hand, plastifications on the compression side of the specimen, leading to a reduction of the loaddisplacement gradient. Thereafter, brittle system failure due to initiation of cracks is observed. The transition from a linear to a nonlinear curve can be explained by local plastificationlike effects in the compression zone of the beams, as can be seen in Fig. 4a. Computing \(f_b\) according to Eq. (4) does not reflect these local plastifications which lead to a nonlinear normal stress distribution over the crosssection height, therefore rendering Eq. (4) inaccurate. Rather, \(f_b\) has system character and represents a stress quantity corresponding to conventional brittle strength theory (Blank et al. 2017).
Name, average mass density \(\rho _\mathrm {avg}\), dynamically obtained stiffness \(E_{\mathrm {GLT,exp}}^\mathrm {dyn}\), effective stiffness \(E_{\mathrm {GLT,exp}}\) under quasistatic fourpointbending, ultimate bending strength \(f_b\) and number of failed lamellas \(n_\mathrm {lam,failed}\)
Name  ρ_{avg}(kg m^{−3})  \(E_{\mathrm {GLT,exp}}^{\mathrm {dyn}}\,({\hbox {MPa}})\)  \(E_{\mathrm {GLT,exp}}\,(\hbox {Mpa})\)  \(f_b\,(\hbox {MPa})\)  \(n_\mathrm {lam,failed}\) 

A1  432  10,664  10,228  52.5  2 
A2  443  11,436  10,229  34.7  3 
A3  414  10,781  10,348  43.5  3 
A4  417  10,701  10,237  42.2  3 
A5  413  9,950  9,163  40.1  3 
A6  455  11,167  9,503  35.6  2 
A7  428  10,363  8,818  37.1  3 
A8  437  11,090  10,295  45.5  3 
A9  404  9,381  8,871  28.5  3 
A10  472  11,419  10,908  37.6  3 
B1  464  15,213  13,898  47.4  2 
B2  478  15,615  16,191  61.2  2 
B3  466  15,283  14,832  56.1  3 
B4  442  12,166  10,735  35.9  4 
B5  470  14,521  13,640  59.3  2 
B6  450  13,507  11,997  50.5  2 
B7  418  12,132  11,395  39.9  2 
B8  450  13,607  12,799  43.5  2 
B9  445  13,073  11,652  47.3  2 
B10  451  13,921  13,456  51.4  1 
C1  474  14,672  13,942  45.7  2 
C2  474  14,835  13,426  39.1  4 
C3  463  13,892  14,213  51.3  2 
C4  458  13,662  12,641  33.4\(\bullet\)  2 
C5  455  13,155  12,351  43.9  6 
C6  448  12,862  13,316  42.3  3 
C7  492  15,300  15,359  47.0  3 
C8  468  14,657  12,950  46.9  3 
C9  476  14,101  13,019  45.5  3 
C10  474  13,880  14,317  56.3\(\bullet\)  4 
D1  408  10,146  9,725  33.5  6 
D2  422  10,220  9,109  24.8  4 
D3  430  10,693  10,781  35.7  7 
D4  434  11,096  10,392  32.3  7 
D5  447  11,507  10,934  26.1  4 
D6  443  11,182  10,638  28.3  5 
D7  420  10,521  9,766  30.7  7 
D8  437  11,013  10,341  30.5  7 
D9  431  10,857  10,216  31.9  6 
D10  424  10,831  10,995  35.7  5 
E1  452  13,555  13,070  50.3  2 
E2  474  14,580  13,819  37.2  3 
E3  455  13,390  13,185  39.1  3 
E4  442  13,326  13,014  39.3  5 
E5  435  12,903  13,252  44.7  2 
E6  448  13,266  12,726  38.8  4 
E7  441  12,679  12,338  31.8  5 
E8  437  12,950  11,871  44.2  2 
E9  489  15,023  15,502\(\bullet\)  58.8\(\bullet\)  5 
E10  474  14,576  13 805  39.3  3 
To identify patterns in the crack directions, for each segment of the recorded crack geometries the height difference \(\varDelta z\) between end and start point were computed. Subsequently, for each beam the sum of those values was computed to obtain a measure for the steepness of the cracks: \(L_{\mathrm {crack},z} = \sum \varDelta z\). Similarly, the component related to the xdirection, \(L_{\mathrm {crack},x}\), was computed from the sum of differences \(\varDelta x\). In Fig. 8, the ratio of those two results \(L_{\mathrm {crack},z}/L_{\mathrm {crack},x}\) is displayed for each beam. Therein, it can be seen that this ratio lies in the same range for beams of grading class LS22 and seems to be independent of the number of laminations. For example, a crack that spans 1000 mm in xdirection is, on average, accompanied by a zincrement of 40 mm. Thus, such a crack crosses approximately one lamination (recall that all laminations have a thickness of 33 mm). Conversely, for grading class LS15, the ratio \(L_{\mathrm {crack},z}/L_{\mathrm {crack},x}\) is significantly larger and a crack with \(\varDelta X = {1000}{\hbox {mm}}\), on average, crosses at least 2 lamellas.
This behaviour can also be observed by comparing the visualizations of the crack patterns of the two grading classes for the same numbers of laminations, i.e. Figures E.1 with E.2 and Figures E.4 with Figures E.5, respectively. For the lower grading class of LS15 crack patterns seem to remain more localized with respect to their extension in longitudinal direction, which can be explained by the higher propability of adjacent weak sections compared to the higher grading class of LS22, which is emphasized by the higher density of colored patches within the plots of the former, showing the locations of knots, and also by the higher amount of blueish/darker colors, denoting higher single knot volumes and, thus, bigger knots.
The comparison of GLT beams for the lower grading class of LS15 (cf. Figures E.1 and E.4) shows that the difference in dimensions and bending spans, about twice the length and bending span for the bigger GLT beams, leads to almost twice the steepness of the cracks. This behaviour might be explained by the fact that for the smaller dimensions, the extension of the crack in vertical direction is limited by their height, as after failure of just two laminations already half of the beam’s cross section is cracked. For the bigger GLT beams of LS15, the crack, which as explained for this grading class tends to be more localized and, thus, to propagate more likely in vertical direction, leads to comparatively more failed laminations.
Interestingly, as mentioned above, the steepness measure for the beams of LS22 (cf. Figures E.2, E.3 and E.5) seems to be the same for all dimensions and numbers of laminations. This means in the present case that as the bending span gets bigger and also the extension of the crack in longitudinal direction \(L_{\mathrm {crack},x}\) is getting bigger, more laminations fail. But still even for the biggest beam type only the first few tensileloaded laminations fail, such that the steepness measure stays the same and the size of the crack only depends on the dimensions and/or bending span.
3.3 Models for the bending strength
For the linear regression model based on minimum values of the stiffness profiles shown in Eqs. (2) and (3), an adjusted coefficient of determination of \(R_{\mathrm {adj}}^2=0.6\) and root mean squared error (RMSE) of \(\sqrt{\mathrm {MSE}}={5.62}{\hbox {MPa}}\) were obtained. In Fig. 10a, the values predicted from the regression model are plotted against the actual values. It can be seen that lower strength values tend to be overestimated while higher strength values tend to be underestimated by the criterion.
Alternatively, a more elaborate model employing a 2D finite element analysis is employed. For this, an approach similar to the mechanical model in Kandler et al. (2015) was chosen. Instead of the continuous laserscan based stiffness profiles, the 3D FE based stiffness profiles according to Fig. 1d are used for describing the longitudinal stiffness of each lamella. In addition, the strength profiles are used to describe the tensile strength of each lamella.
A comparison of the corresponding numerical and the experimental results for the bending strength \(f_b\) is given in Fig. 10b. Therein, the results using the procedure with the four different IPs for the tensile strength property are shown. Results for IP 1 were omitted as the results did not show usable agreement. However, while the correct trend can be observed for IPs 3 and 4, the prediction quality is unsatisfying, the highest coefficient of determination being \(R^2=0.40\) (IP 3). Using the mean stress approach (see Fig. 10c), higher numerical bending strength values are achieved, leading to a coefficient of determination of \(R^2=0.54\), which is still not reliable enough. Therefore, it can be concluded that although the system failure behaviour can be interpreted as brittle failure, such brittle mechanical models do not concur well with the experimental observations. This observation also agrees with the findings presented in the work of Blank et al. (2017).
3.4 Statistical evaluation of the data

General parameters General parameters cover bending span L and height h of the beam and also the average moisture content (MC). Also, the average mass density \(\rho\) of the topmost (tension) lamella is included. Regarding the correlation within this group of parameters, mass density \(\rho\) and moisture content show a linear correlation coefficient of 0.78. This can be explained by increasing weight (and, therefore, increasing values for mass density measurements) of wood with increasing MC. The relationship between these parameters is visualised in Fig. 11b.

Knot morphology parameters The investigated knot morphology parameters comprise the knot volume, the knot area visible on the board surface and the knot interface area to the surrounding wood matrix. Herein, for each parameter the total sum of all knots of the topmost (tension) lamella occurring between the two load application points is used. The linear correlations between knot volume, visible knot area and interface area are between 0.87 and 0.99. Consequently, the correlation with the quantities of interest \(E_\mathrm {GLT,exp}\) and \(f_b\) is approximately the same for these parameters, as can be seen in the three rightmost columns in Fig. 11c. It can be noticed that all three parameters show correlation to the beam length L and beam height h. The reason for this behaviour is the increasing distance between load application points with bigger beam dimensions, see also Fig. 3, which, in turn, leads to an increasing total sum of knot morphology parameters. The distance to the pith did not yield any notable linear correlation to the remaining parameters.

Stiffness related parameters The stiffness related parameters represent the stiffness profiles computed according to the model presented in Kandler et al. (2015) as well as the 3D FE approach. For both stiffness profile types, the minimum value occurring in the tension lamella between the load application points is used as parameter. Also the stiffness profile curvature corresponding to Sect. 3.3 as well as the regression model in Eqs. (2) and (3) belong to this parameter group. Unsurprisingly, the regression model parameter is strongly correlated to the stiffness profile parameter. A notable correlation can be observed between the two stiffness profile parameters and mass density and moisture content measurements. The reason for this observation lies in the micromechanical model (Hofstetter et al. 2005) which was employed for computing the clearwood stiffness tensor within Kandler et al. (2015). For the micromechanical model, mass density and moisture content are the two main input parameters. Also, the two stiffness profile parameters show notable correlation to the knot morphology parameters. The knot morphology can be interpreted as a latent factor influencing both the knot morphology parameters as well as the stiffness profile computation. While the knot morphology is not directly used within the computation of the stiffness profiles, it influences the fibre deviations (Foley 2003) and thus an important aspect of the stiffness profile computation presented in Kandler et al. (2015).

Strength related parameters Strength related parameters represent the strength profiles computed in accordance with Sect. 2.4. In analogy to the stiffness related parameters, for each beam the corresponding strength parameter was defined as the minimum value occurring in the tension lamella between both load application locations. Since all four indicating properties are based on the same knot morphology parameters, they are—with exception of IP 1—very closely correlated. For the same reason they show notable correlation to the 3D FE stiffness parameter. However, the linear correlation values \(\delta\) are rather low, ranging from 0.36 (IP 1) to 0.51 (IP 4). The correlation between the 3D FE stiffness profilebased criterion and the strength \(f_b\) is 0.47.

Dynamic stiffness measurement Between the dynamically obtained stiffness value \(E_\mathrm {GLT,exp}^\mathrm {dyn}\) and the remaining input parameters, the highest linear correlation value is observed to the stiffness profile parameter.

Output/result parameters For the results, the input parameters are sorted according to their linear correlation coefficients with the quantities of interest in Fig. 12. For the effective bending stiffness \(E_\mathrm {GLT,exp}\), the parameter with the highest correlation of 0.95 is the dynamically obtained stiffness \(E_\mathrm {GLT,exp}^\mathrm {dyn}\), followed by the stiffness profile parameter according to Kandler et al. (2015). For the remaining parameters, e.g. moisture content, mass density etc., the correlation is relatively low and it is not expected to gain reliable predictions from those parameters alone. Regarding the bending strength \(f_b\), the regression model (2) achieves a correlation coefficient of 0.79, corresponding to a coefficient of determination (CoD) of \(R^2 = 0.62\). The stiffness profile criterion yields a correlation coefficient close to 0.71, corresponding to \(R^2=0.50\). While exposing a clear trend, these results indicate that for the reliable prediction of bending strength, more sophisticated models have to be employed. Interestingly, the knot morphology apparently shows better correlation to the bending strength than the applied indicating properties. For the number of failed lamellas, \(n_\mathrm {lam,failed}\), no meaningful correlation could be identified.
4 Conclusion
To study the impact of knots on the mechanical behaviour of GLT beams, several parameters were investigated with regard to their ability to predict effective GLT beam properties.
The use of the laser scanning information of all boards, obtained before their assembly to GLT beams, allowed a virtual reconstruction of the knot morphology and led to knowledge of the knots’ locations, orientations and sizes within such beams. Furthermore, real failure mechanisms of experimentally tested GLT beams were investigated, by recording crack patterns, which were used to obtain measures of cracks.
 1.
With an increasing number of laminations, the bending strength \(f_b\) decreases. In addition, the grading class also had the expected influence of lowering the bending strength. It was also observed that more laminations lead to more brittle failure modes. With respect to this effect it would be interesting to investigate beams with constant height and varying number of laminations in the future.
 2.
The loaddisplacement curves confirmed that bigger beam types show a more brittle failure behaviour than smaller ones. In addition, the steepness of the observed cracks (\(L_{\mathrm {crack},z}/L_{\mathrm {crack},x}\)) seemed to be independent of the number of laminations for the higher grading class of LS22. For LS15, the crack patterns were more localized, with smaller extensions in longitudinal direction, such that the steepness of the crack was controlled by the number of laminations. Thus, a higher number of such laminations led to steeper cracks. This behaviour can be explained by the higher probability of adjacent weak zones in adjacent laminations for the lower grading class.
 3.
The best estimate for the effective GLT beam bending stiffness, besides a correlation with experimentally obtained dynamic stiffness values for assembled beams, was achieved by correlating them to minimum values of stiffness profiles of the outermost lamination on the side exposed to tension.
 4.
This stiffness profile parameter also led to the best correlations to the effective beam bending strength (\(R^2=0.62\)). Here, an overestimation of lower strength values and an underestimation of higher strength values was observed. In the application of strength profiles to linear elastic FE models the correct trend for the estimation of strength values could be found, but the resulting coefficient of determination was rather low.
Especially the lastmentioned approach of using stiffness and strength profiles, i.e. of fluctuating sectionwise effective material properties for all boards, based on virtual reconstructions of boards and the evaluation of indicating properties, within simple linearelastic FE simulations, showed the problems of such a method: complex morphologies and failure mechanisms of knot sections cannot be easily replaced by sections with effective strength values if failure mechanisms are not taken into account. Therefore, in the future, distinct mechanical models for timber elements with sectionwise effective properties must be developed. This could for example be achieved by additionally assigning fracture energies to the individual sections, as also suggested by Blank et al. (2017). They already showed, by using linear tractionseparation laws in combination with constant fracture energies, good fits to experiments could be obtained. In future, more realistic, variable fracture energies for knot sections might be obtained by simulating those sections with a multisurface failure criterion (Lukacevic and Füssl 2016; Lukacevic et al. 2017), whose combination with the mentioned fiber deviation model shows promising results. Furthermore, in such an approach, additional strength reducing or failure inducing factors, like finger joints, could be easily implemented by introducing additional weak sections.
Particularly, such an approach is needed if the structure’s failure mechanism is rather brittle. In contrast, for ductile failure mechanisms, like observed within crosslaminated timber plates under point loads, it could be shown that the application of strength profiles in the framework of limit analysis (Füssl et al. 2017; Li 2018), i.e. in combination with perfect plasticity models, leads to reasonably effective strength estimates.
Notes
Acknowledgements
Open access funding provided by TU Wien (TUW). The authors gratefully acknowledge the financial support of the Vienna Business Agency. Promoted out of funds of the City Council of Vienna by Vienna Business Agency. A trust of Vienna City council.
Supplementary material
References
 Balduzzi G, Kandler G, Füssl J (2018) Estimation of GLT beam stiffness based on homogenized board mechanical properties and composite beam theory. In: Proceedings of the 6th European Conference on Computational Mechanics (ECCM 6), Glasgow, UKGoogle Scholar
 Blank L, Fink G, Jockwer R, Frangi A (2017) Quasibrittle fracture and size effect of glued laminated timber beams. Eur J Wood Prod 75(5):667–681CrossRefGoogle Scholar
 Brandner R, Schickhofer G (2008) Glued laminated timber in bending: new aspects concerning modelling. Wood Sci Technol 42(5):401–425CrossRefGoogle Scholar
 Colling F (1990) Biegefestigkeit von Brettschichtholzträgern in Abhängigkeit von den festigkeitsrelevanten Einflußgrößen. Entwicklung eines statistischen Modells (Bending strength of glulam beams–Verification of the statistical model by tests). Holz als Roh und Werkstoff 48:269–273 (in German)CrossRefGoogle Scholar
 Colling F (1995) Brettschichtholz unter Biegebeanspruchung (Glulam beams under bending). Fachverlag Holz der Arbeitsgemeinschaft Holz e.V, Düsseldorf, Germany in GermanGoogle Scholar
 Danielsson H, Gustafsson P (2011) A probabilistic fracture mechanics method and strength analysis of glulam beams with holes. Eur J Wood Prod 69:407–419. https://doi.org/10.1007/s0010701004751 CrossRefGoogle Scholar
 Dorn M (2012) Investigations on the serviceability limit state of doweltype timber connections. PhD thesis, Vienna University of TechnologyGoogle Scholar
 EN 1194 (1999) Timber structures—glued laminated timber—strength classes and determination of characteristic values. Standard, European Committee for Standardization (CEN)Google Scholar
 EN 14080 (2013) Timber structures—glued laminated timber and glued solid timber—requirements. Standard, European Committee for Standardization (CEN)Google Scholar
 EN 14081–4 (2009) Timber structures—strength graded structural timber with rectangular cross section—Part 4: machine grading—grading machine settings for machine controlled systems. Standard, European Committee for Standardization (CEN)Google Scholar
 EN 408 (2010) Timber structures—structural timber and glued laminated timber—determination of some physical and mechanical properties. Standard, European Committee for Standardization (CEN)Google Scholar
 Ehlbeck J, Colling F, Görlacher R (1985) Einfluss keilgezinkter Lamellen auf die Biegefestigkeit von Brettschichtholzträgern. Entwicklung eines Rechenmodells (Influence of fingerjointed lamellae on the bending strength of glulam beams). Holz Roh Werkst 43:333–337 (in German)CrossRefGoogle Scholar
 Falk R, Colling F (1995) Laminating effects in gluedlaminated timber beams. J Struct Eng 121(12):1857–1863. https://doi.org/10.1016/j.engfracmech.2017.02.020 CrossRefGoogle Scholar
 Fink G, Frangi A (2013) Modelling the bending strength of glued laminated timber—considering the natural growth characteristics of timber. In: 46th CIBW18 Meeting 2013, 46, CIBW18/46126Google Scholar
 Fink G, Kohler J (2014) Model for the prediction of the tensile strength and tensile stiffness of knot clusters within structural timber. Eur J Wood Prod 72(3):331–341. https://doi.org/10.1007/s0010701407810 CrossRefGoogle Scholar
 Fink G, Kohler J, Frangi A (2013) Bending tests on glued laminated timber beams with wellknown material properties. IBK Report No 350. Tech. rep., Institut für Baustatik und Konstruktion der ETH Zürich. https://doi.org/10.3929/ethza009950793
 Fink G, Frangi A, Kohler J (2014) Bending tests on GLT beams having wellknown local material properties. Mater Struct 48(11):3571–3584. https://doi.org/10.1617/s1152701404242 CrossRefGoogle Scholar
 Foley C (2003) Modeling the effects of knots in structural timber. PhD thesis, Doctorial thesis, Lund UniversityGoogle Scholar
 Foschi RO, Barrett JD (1980) Gluedlaminated beam strength: a model. Journal of the Structural Division 106(8):1735–1754, URL http://cedb.asce.org/cgi/WWWdisplay.cgi?9648
 Frese M, Blaß HJ (2009) Bending strength of spruce glulam. Eur J Wood Prod 67:277–286. https://doi.org/10.1007/s0010700903162 CrossRefGoogle Scholar
 Füssl J, Li M, Lukacevic M, Eberhardsteiner J, Martin C (2017) Comparison of unit cellbased computational methods for predicting the strength of wood. Eng Struct 141:427–443CrossRefGoogle Scholar
 Goodman J, Bodig J (1978) Mathematical model of the tension behaviour of wood with knots and cross grain. In: Proceedings from the First International Conference On Wood FractureGoogle Scholar
 Hackspiel C, de Borst K, Lukacevic M (2014) A numerical simulation tool for wood grading model development. Wood Sci Technol 48(3):633–649. https://doi.org/10.1007/s0022601406290 CrossRefGoogle Scholar
 Hankinson RL (1921) Investigation of crushing strength of spruce at varying angles of grain. Tech. rep., Air Service Information Circular, 3(259), Material Section Report No 130, US Air Service, USAGoogle Scholar
 Hofstetter K, Hellmich C, Eberhardsteiner J (2005) Development and experimental validation of a continuum micromechanics model for the elasticity of wood. Eur J Mech A/Solids 24:1030–1053CrossRefGoogle Scholar
 Hofstetter K, Hellmich C, Eberhardsteiner J (2007) Micromechanical modeling of solidtype and platetype deformation patterns within softwood materials. a review and an improved approach. Holzforschung 61(4):343–351CrossRefGoogle Scholar
 Kandler G, Füssl J, Serrano E, Eberhardsteiner J (2015) Effective stiffness prediction of GLT beams based on stiffness distributions of individual lamellas. Wood Sci Technol 49(6):1101–1121. https://doi.org/10.1007/s0022601507455 CrossRefGoogle Scholar
 Kandler G, Lukacevic M, Füssl J (2016a) An algorithm for the geometric reconstruction of knots within timber boards based on fibre angle measurements. Constr Build Mater 124:945–960. https://doi.org/10.1016/j.conbuildmat.2016.08.001 CrossRefGoogle Scholar
 Kandler G, Lukacevic M, Füssl J (2016b) From the knot morphology of individual timber boards to the mechanical properties of glued laminated timber. In: Proceedings of the World Conference on Timber Engineering (WCTE 2016), Vienna, AustriaGoogle Scholar
 Li M, Füssl J, Lukacevic M, Martin C, Eberhardsteiner J (2018) Bending strength predictions of crosslaminated timber plates subjected to concentrated loading using 3D finiteelementbased limit analysis approaches. Submitted to Composite StructuresGoogle Scholar
 Lukacevic M, Füssl J (2014) Numerical simulation tool for wooden boards with a physically based approach to identify structural failure. Eur J Wood Prod 72(4):497–508. https://doi.org/10.1007/s001070140803y CrossRefGoogle Scholar
 Lukacevic M, Füssl J (2016) Application of a multisurface discrete crack model for clear wood taking into account the inherent microstructural characteristics of wood cells. Holzforschung 70(9):845–853. https://doi.org/10.1515/hf20150162 CrossRefGoogle Scholar
 Lukacevic M, Füssl J, Eberhardsteiner J (2015) Discussion of common and new indicating properties for the strength grading of wooden boards. Wood Sci Technol 49:551–576. https://doi.org/10.1007/s0022601507121 CrossRefGoogle Scholar
 Lukacevic M, Lederer W, Füssl J (2017) A microstructurebased multisurface failure criterion for the description of brittle and ductile failure mechanisms of clearwood. Eng Fract Mech 176:83–99Google Scholar
 Neuhaus FH (1981) Elastizitätszahlen von Fichtenholz in Abhängigkeit von der Holzfeuchtigkeit (Elasticity data of spruce wood depending on wood moisture). PhD thesis, RuhrUniversität Bochum, URL http://permalink.obvsg.at/AC01538444, in German
 Nyström J (2003) Automatic measurement of fiber orientation in softwoods by using the tracheid effect. Comput Electron Agric 41:91–99. https://doi.org/10.1016/S01681699(03)000450 CrossRefGoogle Scholar
 Olsson A, Oscarsson J (2017) Strength grading on the basis of high resolution laser scanning and dynamic excitation: a full scale investigation of performance. Eur J Wood Prod 75(1):17–31. https://doi.org/10.1007/s0010701611026 CrossRefGoogle Scholar
 Olsson A, Oscarsson J, Johansson M, Källsner B (2012) Prediction of timber bending strength on basis of bending stiffness and material homogeneity assessed from dynamic excitation. Wood Sci Technol 46(4):667–683. https://doi.org/10.1007/s002260110427x CrossRefGoogle Scholar
 Olsson A, Oscarsson J, Serrano E, Källsner B, Johansson M, Enquist B (2013) Prediction of timber bending strength and inmember crosssectional stiffness variation on basis of local wood fibre orientation. Eur J Wood Prod 71(3):319–333. https://doi.org/10.1007/s0010701306845 CrossRefGoogle Scholar
 Oscarsson J, Olsson A, Enquist B (2014) Localized modulus of elasticity in timber and its significance for the accuracy of machine strength grading. Wood and Fiber Science 46(46)Google Scholar
 Petersson H (2010) Use of optical and laser scanning techniques as tools for obtaining improved FEinput data for strength and shape stability analysis of wood and timber. In: IV European Conference on Computational Mechanics, Paris, France, MayGoogle Scholar
 Phillips G, Bodig J, Goodman J (1981) Flowgrain analogy. Wood Sci 14(2):55–64Google Scholar
 Serrano E, Gustafsson PJ (2007) Fracture mechanics in timber engineering—strength analyses of components and joints. Mater Struct 40(1):87–96. https://doi.org/10.1617/s1152700691210 CrossRefGoogle Scholar
 Simonaho SP, Palviainen J, Tolonen Y, Silvennoinen R (2004) Determination of wood grain direction from laser light scattering pattern. Optics and Lasers in Engineering 41(1):95–103. https://doi.org/10.1016/S01438166(02)001446
Copyright information
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.