Determining the optimum layer combination for cross-laminated timber panels according to timber strength classes using Artificial Neural Networks

Abstract

The primary aim of this work was to determine the effects of production parameters, such as wood species and timber strength classes, on some mechanical properties of cross-laminated timber (CLT) panels using artificial neural network (ANN) prediction models. Subsequently, using the models obtained from the analyses, the goal was to identify the optimum layer combinations of timber strength classes used in the middle and outer layers that would provide the highest mechanical properties for CLT panels. CLT panels made from spruce and alder timbers, as well as hybrid panels created from combinations of these two wood species, were produced. The strength classes of the timbers were determined non-destructively according to the TS EN 338 (2016) standard using an acoustic testing device. The bending strength and modulus of elasticity values of the CLT panels were determined destructively according to the TS EN 408 (2019) standard. According to ANN results, the optimum timber strength classes and layer combinations were determined for bending strength as C24-C27-C24 for spruce CLT, D18-D24-D18 for alder CLT, C30-D40-C30 and D18-C30-D18 for hybrid panels; and for modulus of elasticity, C22-C27-C22 for spruce, D35-D30-D35 for alder, C16-D24-C16, and D24-C24-D24 for hybrid panels.

Full Article

Determining the Optimum Layer Combination for Cross-Laminated Timber Panels According to Timber Strength Classes Using Artificial Neural Networks

Engin Derya Gezer,^a,* Abdullah Uğur Birinci,^a Aydın Demir,^a Hasan Öztürk,^b,c Okan İlhan,^a and Cenk Demirkir ^a,c

The primary aim of this work was to determine the effects of production parameters, such as wood species and timber strength classes, on some mechanical properties of cross-laminated timber (CLT) panels using artificial neural network (ANN) prediction models. Subsequently, using the models obtained from the analyses, the goal was to identify the optimum layer combinations of timber strength classes used in the middle and outer layers that would provide the highest mechanical properties for CLT panels. CLT panels made from spruce and alder timbers, as well as hybrid panels created from combinations of these two wood species, were produced. The strength classes of the timbers were determined non-destructively according to the TS EN 338 (2016) standard using an acoustic testing device. The bending strength and modulus of elasticity values of the CLT panels were determined destructively according to the TS EN 408 (2019) standard. According to ANN results, the optimum timber strength classes and layer combinations were determined for bending strength as C24-C27-C24 for spruce CLT, D18-D24-D18 for alder CLT, C30-D40-C30 and D18-C30-D18 for hybrid panels; and for modulus of elasticity, C22-C27-C22 for spruce, D35-D30-D35 for alder, C16-D24-C16, and D24-C24-D24 for hybrid panels.

DOI: 10.15376/biores.19.3.4899-4917

Keywords: Artificial neural network (ANN); Cross-laminated timber (CLT); Strength class; Non-destructive testing

Contact information: a: Karadeniz Technical University, Faculty of Forestry, Department of Forest Industry Engineering, 61080, Trabzon, Türkiye; b: Karadeniz Technical University, Arsin Vocational School, Department of Materials and Material Processing Technologies, 61900, Trabzon, Türkiye; c: Trabzon Technology Development Region Manager, HC Panel Forest Products Consulting Services Industry and Trade Limited Company, 61081, Trabzon, Türkiye; *Corresponding author: gezer@ktu.edu.tr

GRAPHICAL ABSTRACT

INTRODUCTION

Wood is recognized for its carbon storing ability, renewability, reusability, recyclability, lower carbon impact, it is environmentally benign, and promotes good health (Di Bella and Mitrovic 2020). Wood construction technologies are widely employed throughout many regions globally. Out of all the options, cross-laminated timber (CLT) stands out as the most captivating and groundbreaking material for wooden construction. CLT is considered a desirable material for various types of buildings, including detached houses, multi-story buildings, schools, auditoriums, exhibition centers, sports halls, theaters, commercial establishments, and religious buildings. This is due to its advantageous characteristics, such as its lighter weight compared to concrete and steel, its quick construction time, its low carbon footprint, and its ability to withstand earthquakes (Di Bella and Mitrovic 2020; Hindman and Golden 2020; Hematabadi et al. 2020).

At present, the primary raw material for producing CLT panels on a commercial scale is derived from three main coniferous wood species: spruce, pine, and fir (Srivaro et al. 2020). Recent studies have explored the use of fast-growing tree species to broaden the raw material source and enhance the qualities of CLT or hybrid CLT (Dong et al. 2023). Studies have identified poplar, eucalyptus, magnolia, and bamboo as often utilized fast-growing tree or non-wood species in the manufacture of CLT (Marko et al. 2016; Liao et al. 2017; Dong et al. 2023; Satir 2023; Shi et al. 2023). There is a lack of research in the existing literature regarding the utilization of alder timbers, a rapidly growing tree species indigenous to Turkey, in the manufacturing of CLT. This study successfully manufactured CLT panels utilizing alder and spruce timbers, as well as hybrid panels combining these two wood species.

The selection of wood species for the manufacture of structural material is crucial, with the quality and strength class of the chosen species being key criteria that significantly impact the quality of CLT. The TS EN 338 (2016) European standard has established a strength classification for coniferous wood species, consisting of 12 groups (C14, C16, C18, C20, C22, C24, C27, C30, C35, C40, C45, and C50). Similarly, deciduous wood species have been categorized into 8 groups (D18, D24, D30, D35, D40, D50, D60, and D70). These classifications not only ascertain the caliber of the timber but also have a substantial impact on its selling price. According to literature research, European big enterprises that produce CLT typically utilize timbers classified as C24 strength classes based on the TS EN 338 (2016) standard (Wieruszewski and Mazela 2017). Furthermore, companies such as Binderholz, from Austria, utilize a minimum of C18 strength values. In addition, Stora Enso employs C16 and C30 values, Thoma Holz GmbH uses C16 values, Martinsons from Sweden uses at least C14 values, and Lignotrend AG from Switzerland can incorporate both C20 and C24 strength values (Wieruszewski and Mazela 2017). Previous studies in the literature have noted that coniferous wood species are commonly utilized in the production of CLT. However, these studies have primarily focused on a single group, typically C24, instead of comparing strength values. Among the notable studies on this subject are those by Hassanieh et al. (2017), Reynolds et al. (2017), Follesa and Fragiancomo (2018), Brandner (2018), Turesson et al. (2019), and Lie et al. (2020). In addition, strength classes, such as C16 (Kippel et al. 2014), C18 (Soriano et al. 2016; Luengo et al. 2017), C22 (Guo et al. 2017), and C30 (Hadigheh and Dias da Costa 2020), have also been favored. O’Dowd et al. (2016) reported the utilization of timbers from two distinct strength classes in the manufacturing of a single CLT panel. This study specifically chose three distinct strength classes of spruce and alder timbers and utilized various layer combinations to create CLT. It is worth noting that a full investigation into strength classes of this nature has not been documented in the existing literature.

Cost and time-effective testing can be conducted to determine the technological qualities of CLT materials. Nevertheless, altering the production parameters of CLT could result in a procedure that may incur expenses and delays when establishing the potential strength values, considering various elements. Currently, it is feasible to forecast the strength characteristics of wood material under various manufacturing settings by utilizing an Artificial Neural Network (ANN) approach to model the data acquired from existing testing. According to literature research, ANNs have been effective in optimizing structural timber materials. Demir et al. (2023) investigated the impact of certain production parameters on the structural behavior of plywood-sheathed walls when subjected to lateral load. They successfully obtained optimal outcomes by utilizing the ANN to model the nail spacing during production. Öztürk et al. (2022) investigated how press parameters impacted the mechanical properties of composite plywood. They utilized ANN models to determine the optimal press settings that yield the maximum strength values. In addition, Demirkir et al. (2013) utilized ANN models to identify the optimal values for certain production factors that yielded the greatest adhesive strength in plywood. The objective of this study was to assess the impact of production characteristics, specifically wood species, and wood strength class, on the mechanical properties of CLT panels using ANN. In addition, the objective was to identify the optimal combinations of timber strength classes in the intermediate and outer layers of CLT panels that would yield the maximum mechanical properties. This was achieved through optimization studies utilizing ANN models.

EXPERIMENTAL

Wood Material

In this study, spruce (Picea orientalis L.), a coniferous wood species, was selected as one of the most preferred types for CLT production. Additionally, alder (Alnus glutinosa subsp. barbata (C.A. Mey.) Yalt.), a type of deciduous tree known for its rapid growth, was chosen. Timber from both species was sourced from different regions, in three different quality classes for spruce and two different quality classes for alder, according to the appearance characteristics specified in the TS 1265 (2012) and TS EN 14081-1 (2019) standards, respectively. The timber was planed on all four sides and supplied dry. Care was taken to ensure that the timber had a moisture content within the most suitable range for CLT production, which is 12 ± 3%. Before CLT production, the timbers were dimensioned to sizes of 120 × 10 × 1.8 cm³ and 240 × 10 × 1.8 cm³.

Determination of Timber Strength Classes

The parameters used in this study made it possible to find out the strength classes of the woods before they were used to make CLT. This was done by using non-destructive testing methods that followed the TS EN 338 (2016) Standard. The timbers were obtained based on their visual qualities. The Sylvatest 4 acoustic testing instrument, developed by CBS-CBT, was utilized to ascertain the strength classes of the timber (Fig. 1).

Fig. 1. Determining the strength classes of timber with an acoustic testing device

The instrument uses software to determine the bending strength and modulus of elasticity directly from the measurement of the ultrasonic wave flight duration in the wood material between two transducers (the transmitter and receiver). It then displays the strength class in accordance with the applicable standard. It also creates a graph from the acquired data (Fig. 2). Because the alder wood species is not included among the tree species in the device’s software, the MOE values for this wood species were calculated using the Eq. 1 given below,

(1)

where V is the velocity obtained from the device and d is the density of the alder lumber pieces.

Fig. 2. Acoustic testing device software data

A non-destructive test was performed on 1876 spruce and 1063 alder lumber pieces obtained according to visual quality class. The non-destructive tests yielded strength classes C and D, respectively, for spruce and alder timbers, in accordance with the TS EN 338 (2016) standard. Within the parameters of this investigation, timbers of spruce (C16, C22, and C30) and alder (D18, D30, and D40) were chosen for use in CLT manufacturing based on the obtained strength class groups. The distribution graphs of the MOR and MOE for spruce and alder lumbers are shown in Fig. 3.

Production of CLT Panels

A 160 g/cm² polyurethane adhesive solution was put on the wood’s surfaces in the layer combination groups. These groups were made using the types of wood and strength classes listed in the TS EN 338 (2016) standard. The design process for the manufactured CLT panels involved arranging the layers perpendicularly. In the process of drafting, the orientation of the annual rings in the cross-sections of the timbers was considered to reduce the material shrinkage and swelling. This resulted in the creation of drafts using alder, spruce, and hybrid (alder and spruce combination) timbers. The draft boards were prepared to have three layers and then underwent a pressing procedure. The hydraulic press used for the draft pressing was capable of exerting both vertical and lateral pressures, meeting the requirements of industrial settings.

Fig. 3. The distribution of the MOR and MOE for spruce and alder lumbers

The pressing method involved the use of a cold press. The applied pressure was 0.8 N/mm² for spruce and hybrid, and 1.2 N/mm² for alder. To reduce the space between the wood pieces, a lateral pressing technique was employed, applying pressure ranging from 0.276 to 0.550 N/mm² in conjunction with vertical pressing (Karacabeyli and Gagnon 2019). Table 1 presents the 30 test groups that were generated by combining different wood species with the TS EN 338 (2016) standard.

Destructive Testing of CLT Panels

The bending strength and modulus of elasticity tests for the CLT panels were conducted according to the TS EN 408 (2014) standard, which is used to determine the mechanical and physical properties of structural timber and glued laminated timber. For the bending strength and modulus of elasticity tests, six samples (replicates) from each group were taken.

ANN Modeling

Three distinct strength categories were established for spruce and alder timbers utilized in the manufacturing of CLT panels. During the evaluation, consideration was made based on the TS EN 338 (2016) standard that coniferous timber with strength classes C45 and C50 may not be accessible.

Table 1. Test Groups Formed According to Wood Species and TS EN 338 (2016) Standard

Additionally, no research studies were found regarding the utilization of timber with strength classes higher than C30 in the production of CLT. Thus, the groups were chosen based on the strength class selection, which included C16, C22, and C30 for coniferous and D18, D30, and D40 for deciduous wood species.

The selection was made to ensure equal spacing between the classification rankings specified in the applicable standard, up to the C30 standard. The objective of the ANN modeling in this study was to determine the strength values for different strength classes according to the TS EN 338 (2016) standard using experimental data gathered from three strength classes. Therefore, the study was able to accurately predict the structural strength of CLT panels that would be made from timbers with different strength classes. This meant that expensive and time-consuming tests were not needed, and the results were very reliable. ANN studies were also used to make prediction models that helped find the best combinations of wood strength classes for spruce, alder, and hybrid CLT panels for both the outer and middle layers.

The main process variables (inputs) used in this ANN modeling were the wood species and the strength class values of the timbers used in the outer and inner layers, according to TS EN 338 (2016). The output variables consisted of the bending strength and modulus of elasticity values, which represented the mechanical strength properties of the CLT panels. The MATLAB software package was utilized for the creation, training, and testing of artificial neural networks, ultimately leading to the identification of optimal values. The results were acquired via experimental research conducted in a manner that caused damage or destruction. To find out how the strength properties changed when different wood species and strength classes of wood were used in the outer and middle layers, as described in TS EN 338 (2016), the experimental data was split into training and test sets. Out of the available data, 20 examples were chosen for the training phase of the ANN, while the remaining 10 samples were utilized to assess the ANN’s ability to validate prediction capability. The estimated values obtained from the test process were compared to the actual values. The prediction models that yielded the most accurate predictions were selected based on the root mean square error (RMSE) calculated using Eq. 2 and the mean absolute percentage error (MAPE) calculated using Eq. 2. These diagnostic measures are widely recognized as the best and most commonly used indicators of performance. These equations are as follows,

(2)

(3)

where t_i represents the actual value, td_i represents the model prediction value, and N refers to the number of terms. The network architectures identified as the top-performing prediction models in ANN analyses are shown in Figs. 4 and 5, based on the output variables. The network topologies of the models that yield the most accurate and precise results are shown in the pictures. These structures comprise of 1 input layer, 2 hidden layers, and 1 output layer. Each hidden layer consisted of 4 to 2 neurons for measuring bending strength and 3 to 5 neurons for measuring modulus of elasticity, respectively.

Fig. 4. The ANN architecture selected as the bending strength prediction model

Fig. 5. The ANN architecture selected as the elasticity modulus prediction model

A multilayer feedforward and backpropagation ANN has been chosen as the preferred method for handling the challenges. The activation functions chosen for this investigation are the hyperbolic tangent sigmoid function (tansig) and the linear transfer function (purelin). The learning rule chosen for the neural network was the gradient descent with momentum backpropagation algorithm (traingdm), while the training technique picked was the Levenberg-Marquardt algorithm (trainlm). The performance function utilized for evaluation was the mean square error (MSE), which was computed using Eq. 4:

(4)

The training and test data were standardized to the range of (-1,1) in the models constructed using the hyperbolic tangent sigmoid function. This normalization ensured that each variable had an equal impact on the prediction models. Afterwards, to guarantee an accurate understanding of the outcomes, the data underwent a reverse normalizing procedure to revert them to their initial values. The normalization (scaling) procedures were computed using Eq. 4. The equation defines the variables as follows: X_norm represents the data that has been normalized, X represents the actual value of the variable, X_min represents the minimum value in the data set, and X_max represents the highest value in the data set.

RESULTS AND DISCUSSION

ANN Modeling

The study successfully obtained extremely dependable and accurate prediction models using destructive structural strength values. During the analysis, a multitude of models were trained and tested. The models that yielded the most accurate predictions were determined based on the MSE values. Figure 6 depicts the variations in these values over the course of the iterations.