Modeling of mechanical properties of wood-polymer composites with Artificial Neural Networks

Eroğlu, M. A., Altun, S., and Ciritcioğlu, H. H. (2024). “Modeling of mechanical properties of wood-polymer composites with Artificial Neural Networks,” BioResources 19(3), 4468-4485.

Abstract

Mechanical properties (tensile strength (TS), modulus of elasticity in tensile (MET), flexural strength (FS), modulus of elasticity (MOE)) of the material to be obtained depending on the production parameters in the production of high-density polyethylene (HDPE) wood-polymer composites with Scots pine wood flour additive were predicted using Artificial Neural Networks (ANN) model and without destructive testing. In the first stage of the study, an ANN model was developed using data from 56 different studies in the literature on the mechanical properties of wood polymer composites. In the second stage, in order to determine the reliability of the model, output values were estimated using input parameters that had not been used in training and testing of the model. Based on the same input parameters, test specimens were produced and mechanical tests were performed. The results obtained from the experiments and ANN model were compared by considering the mean absolute percentage error (MAPE) value. The coefficient of determination (R²) values obtained in the training and testing phase of the ANN models were all higher than 0.90. In this way, the mechanical properties of the wood polymer composite were successfully predicted by the ANN model. Because most of the MAPE values obtained from the mechanical tests were below 10%, the model was considered a reliable model.

Download PDF

Full Article

Modeling of Mechanical Properties of Wood-Polymer Composites with Artificial Neural Networks

Mustafa Altay Eroğlu,^a,* Suat Altun,^b and Hasan Hüseyin Ciritcioğlu ^c

DOI: 10.15376/biores.19.3.4468-4485

Keywords: Tensile strength; Flexural strength; Modulus of elasticity; HDPE; MAPE

Contact information: a: Department of Interior Design, Safranbolu Şefik Yılmaz Dizdar Vocational School, Karabuk University, Safranbolu/Karabuk, Türkiye; b: Department of Industrial Design Engineering, Technology Faculty, Karabuk University, Karabuk, Türkiye; c: Wood Products Industrial Engineering, Forestry Faculty, Düzce University, Düzce, Türkiye;

* Corresponding author: altayeroglu@karabuk.edu.tr

INTRODUCTION

New material that is formed by the combination of at least two different materials in macro dimensions is called a composite material. In the production of composite materials, materials that are not suitable on their own and cannot dissolve in each other are given new properties such as lightness, cost, flexibility, and strength to make them suitable for their areas of application. Wood polymer composites are also used to replace wood materials. The variety of materials and production parameters used in the manufacture of wood polymer composites results in a wide range of mechanical properties of the resulting material. Determining the mechanical properties of composites of different compositions using destructive testing methods is both laborious and costly. For example, for the validation experiments carried out in the later stages of this study, approximately 40 working hours were spent with two people in the laboratory environment. From a manufacturer’s perspective, extra time spent on production, labor costs, electricity costs, machine depreciation, etc., will arise. Additionally, the manufacturer may not have the necessary equipment and skills to perform mechanical testing, resulting in the need to hire outside services, which can be time-consuming and costly. Using the ANN method, it is possible to reach the same result within minutes by trying different parameters. In this way, predicting the properties of the material to be produced in advance will save time and money.

To effectively use ANN, the usual procedure is to collect information about samples, make generalizations, and then make decisions about those samples using the learned information to make predictions about samples not yet considered in the analysis. Due to these learning and generalization properties of ANN, it finds wide application in many fields of science today and demonstrates its ability to solve complex problems (Ergezer et al. 2003). Therefore, ANN modeling can be successfully applied to the prediction of mechanical properties of composite materials.

One important class of composite materials is wood polymer composites. Only one study on predicting the mechanical properties of these composites with ANN using low-density polyethylene (LDPE) was found (Atuanya et al. 2014). When looking at the studies on determining the mechanical properties of wood polymer composites, it is apparent that HDPE, LDPE, and polypropylene (PP) polymers are often used due to their low cost and ease of processing.

The studies using HDPE as the polymer used pine, poplar, fir, acacia, oak, maple, beech, rubber, and cedar wood flour as fillers, maleic anhydride grafted polypropylene (MAPP), maleic anhydride grafted polyethylene (PE-g-MA) as coupling agent, and various minerals as additives. The tensile and flexural properties of composites made from recycled HDPE are similar to those of composites made from virgin HDPE (Adhikary et al. 2008). Significant increases in mechanical properties have been reported with the addition of coupling agents and additives such as MAPP, PE-g-MA, and silane (Adhikary et al. 2008; Nourbakhsh and Ashori 2009; Mbarek et al. 2011; Koohestani et al. 2017). In general, mechanical properties improve with increasing wood filler content (Bouafif et al. 2009; Nourbakhsh and Ashori 2009; Mbarek et al. 2011; Koohestani et al. 2017; Martins et al. 2017). Modulus of elasticity, flexural values, and tensile strength values increase with increasing wood flour particle size (Bouafif et al. 2009; Rafighi et al. 2014; Chaudemanche et al. 2018). However, some studies reported a decrease in mechanical properties with increasing particle size (Homkhiew et al. 2018), and an increase in wood flour content decreased flexural, tensile, and impact strength (Balasuriya et al. 2001). Rafighi et al. (2014) reported that there was no significant difference in the mechanical properties of wood polymer composites concerning wood species.

In the studies with PP, Scots pine, poplar, maple, oak, Weymouth pine, beech, cypress, spruce, fir, red pine, and eucalyptus, MAPP as coupling agent and various minerals as additives were preferred as fillers. Different results on the mechanical properties of wood-PP composites have been reported in the literature. In some studies, the mechanical properties of the composites were found to decrease with increasing wood flour content (Bledzki et al. 2002; Kaymakçı 2015). The mechanical properties increase when coupling agents are used (Oksman and Clemons 1997; Bledzki et al. 2002; Stark and Rowlands 2002; Bledzki and Faruk 2003; Ndiaye et al. 2010; Nourbakhsh and Ashori 2010; Kaymakçı 2015) and when wood flour particle size is reduced (Nourbakhsh and Ashori 2010; Haque et al. 2019; Leu et al. 2012; Delviawan et al. 2019; Murayama et al. 2019).

It was found that polymer-wood flour blend samples had lower impact resistance values compared to pure polymer (Nourbakhsh and Ashori 2010; Haque et al. 2019). In another study, the notched impact strength increased with increasing wood flour size (Stark and Rowlands 2002). Increasing wood flour content increases flexural strength and modulus of elasticity, decreased impact strength and tensile strength, and adding 5% clay to the samples significantly decreased modulus of elasticity (Ndiaye et al. 2010). Increasing wood flour content increases modulus of elasticity and modulus of elasticity in tensile, decreases flexural and tensile strength, and increases notched impact strength but decreases unnotched impact strength (Stark and Berger 1997; Leu et al. 2012). Minerals increase the impact strength of composites (Oksman and Clemons 1997). Karmarkar et al. (2007) reported that an increase in wood flour content increased tensile strength and modulus of elasticity in tensile and decreased notched impact strength.

In the studies using LDPE, spruce, radiata pine, eucalyptus, acacia, oak, and rubber were preferred as fillers, PE-g-MA as coupling agent, and various minerals as additives. While there are studies indicating that increasing the amount of wood flour positively affects the mechanical properties of the samples (Turku et al. 2017; Moreno and Saron 2017), Effah et al. (2018) stated that increasing the wood flour content caused a decrease in the mechanical properties of the samples and pure LDPE had higher impact resistance than all composites. Prachayawarakorn et al. (2006) reported that as the amount of wood flour increased, the modulus of elasticity in tensile and modulus of elasticity increased, and the tensile, flexural, and impact resistance decreased significantly. The use of coupling agents (Turku et al. 2017) and minerals (Prachayawarakorn et al. 2006) has been reported to improve the mechanical properties of composites.

In the studies available in the literature on predicting the mechanical properties of wood polymer composites using ANN, a back-propagation multilayer perceptron ANN model was used to predict the mechanical properties of samples made from palm wood fiber and recycled LDPE. Increasing filler content decreased tensile strength, increased flexural and notched impact strength, and achieved a coefficient of determination over 0.95 between ANN predictions and experimental results in all tests (Atuanya et al. 2014).

This study aimed to predict the mechanical properties of the material to be obtained as a function of the production parameters in the production of wood polymer composites using the ANN model, without destructive testing, much faster and at a low cost.

Because the test specimens produced in small numbers and under the same conditions are mostly used in the studies modeling with ANN, it is assumed that the models tend to be memorized rather than learned, and therefore R² values are high. However, due to the different machinery, equipment, and working environment conditions used in the production of wood polymer composites, and the variations in the structure of the polymers used depending on the manufacturer, the variation in the results obtained is greater. It is predicted that the use of a larger database obtained from different sources in the training of the model will provide more realistic results. For this reason, this study aims to make more realistic predictions by using studies in the literature on the mechanical properties of wood polymer composites in the creation of the ANN model. It is aimed to determine the reliability of the model by comparing the mechanical test results of the samples produced with the input parameters not included in the training of the model with the values predicted with the same input parameters.

EXPERIMENTAL

Materials

To determine the reliability of the ANN model, the wastage parts were evaluated in species with high market value that are frequently used as filler in composites. Scots pine (Pinus sylvestris) wood chips were obtained from Bağdatlı Kereste in Safranbolu Industrial Zone. Polymer HDPE (density: 0.964 g/cm³, melting index: 190 °C, 8 to 22 g/10 min, pellet form, white and odorless) and coupling agent PE-g-MA (density: 0.97 to 0.99 g/cm³, pour point: 120 to 126 °C, viscosity 80 to 200 mPa-s, white, fine-grained powder) were used.

Methods

The study consisted of two stages. In the first stage, a literature search was conducted on the mechanical properties of wood polymer composites, and an ANN model was developed using data obtained from 56 different studies directly related to the mechanical properties of wood polymer composites. In the second stage, to determine the reliability of the model, output values were estimated using input parameters that were not previously used in the training and testing of the model. Based on the same input parameters, test specimens were produced, and mechanical tests were performed. The results obtained from the experiments and the ANN model were compared by considering the MAPE value.

Creating the ANN Model

The literature showed that the values of tensile strength, flexural strength, modulus of elasticity in tensile and modulus of elasticity of wood-polymer composites were the most commonly investigated, as well as the effect of filler content, particle size, and coupling agent use on these mechanical properties was also investigated. The data used to create ANN models are given in Table 1, and accordingly, the input parameters and output values of the ANN model were determined. Because the mechanical properties determined in the studies in the literature were not the same, a separate model was created for each of the above mechanical properties. Polymer type, filler type, filler blend ratio, filler particle size, and coupling agent use were determined as input parameters, and the corresponding mechanical property was determined as an output parameter.

When modeling the ANN, the “feed-forward backpropagation” ANN algorithm was used because it can reduce errors backward and from output to input (Gönül et al. 2015). Eighty percent of the data obtained from the literature were used to train the model, and 20% were used to test the model. ANN models were created using MATLAB® R2017b software. The ANN architecture proposed in this study consists of an input layer, two hidden layers, and an output layer (5-n1-n2-1) (Fig. 1). The input layer represents the input parameters, and the output layer represents the mechanical property, which is the dependent parameter of the model. Although the structure of an ANN and the number of neural cells vary, there is no accepted rule for the formation of an ANN. Neural networks with fewer than the required number of hidden layers are inadequate for solving complex functions, while neural networks with too many hidden layers encounter undesirable instabilities. There is no mathematical test for how many neurons should be placed in the hidden layer in the most efficient way. These questions should be decided by trial and error (Ataseven 2013). Because of the complexity of the problem, two hidden layers were used instead of a single layer in this study. In the hidden layers, different numbers of neurons ranging from 5 to 20 were used experimentally to select the number of neurons with the highest accuracy of the model.

Fig. 1. ANN architecture developed to predict mechanical properties of wood polymer composites

In ANN, the activation function decides whether to activate a node or not. If each layer’s activation functions were linear, they would be equivalent to a single layer, called linear transformations. Therefore, nonlinear activation functions are needed for the neural network to have multiple layers in a meaningful way. Among the nonlinear activation functions, sigmoid and tangent hyperbolic activation functions are widely preferred in the literature. The sigmoid function tends to vanish gradients when inputs are large or small (with gradients approaching zero). This can cause the gradients to “jam” and disappear during back-propagation. The tanh function preserves the gradients even when the inputs are large, resulting in a more stable training process. Since the tangent hyperbolic activation function derivative is steeper than the sigmoid function derivative, it can take more values, and this means that when using the tangent hyperbolic activation function, the ANN will be more efficient, as it will have a broader range for faster learning and classification (Szandała 2021). This property also helps to accelerate convergence during training.

In this study, ANN was used to predict the effects of production inputs on mechanical properties, which are considered material quality criteria in the wood polymer composite production process. Due to the diversity of inputs in ANN training and the variability in the predicted mechanical properties due to this input diversity, it was deemed appropriate to use the tangent hyperbolic activation function (Eq.1), which promises more stable training due to the advantages mentioned above.

(1)

Calculation of the change in bias and weight values used in the ANN model-learning algorithm (Duman et al. 2018) is given in Eq. 2 and 3.

(2)

(3)

The momentum and learning coefficients used in the ANN model were chosen in the range of 0 to 1 (Duman et al. 2018). As a result of the experiments conducted to obtain the highest R² value in the ANN model, the momentum and learning coefficients were applied as 0.0001.

Table 1. Studies Used in the ANN Model

The scaling of input parameters and output values in the back-propagation model significantly affects the performance of the ANN (Çay et al. 2012). For this reason, the input and output data sets used in the ANN model were normalized prior to the training and testing process (Eq. 4), and the data were ensured to take values between 0 and 1.

(4)

where V_Nis the normalized data (input or output), V_i is the input value, V_min is the smallest value in the input set, and V_max is the largest value in the input set.

Root mean square error (RMSE) given in Eq. 5 and R² given in Eq. 6 (Duman et al. 2018) were used to evaluate the performances in trial and test data of the ANN model.

(5)

(6)

where X_i is the observed value, X_p is the predicted value and N is the total number of samples.

To determine the reliability of the ANN model, the MAPE value (Eq. 8) was taken into account when comparing the test results of the samples produced using input parameters not included in the training and testing of the model and the prediction values of the model. Before calculating the MAPE the ANN output data were converted back to real values using Eq. 7 and the prediction values were obtained in this way.

(7)

(8)

where X_pois the predicted value, X_imax is the largest observed value in the training data set, X_imin is the smallest observed value in the training data set, X_ie is the set of experimental values, and n is the number of samples used.

Preparation of Mechanical Test Specimens

The composition of the wood polymer composite samples prepared to determine the reliability of the ANN model is shown in Table 2. For this purpose, the Scots pine chips used as filler were sieved on a shaker sieve (LOYKA ESM-200) at 30-35-100-120 mesh. The wood flour remaining between 30-35 and wood flour remaining between 100-120 was collected and dried in the drying oven (NÜVE KD 400) at 103±2°C until it reached a constant weight. The HDPE supplied in granular form was pulverized by grinding in a grinding machine (Elmas Makine, Turkey) and sieved to obtain the same dimensions as sawdust. The obtained polymer and powdered coupling agent were dried in drying ovens (NÜVE FN 055) at 60 °C until constant weight.

The filler, polymer, and coupling agent, which were obtained in the same particle size and dried, were homogeneously mixed using a multidirectional mechanical mixer. The mixture was then mixed in a twin screw extruder (screw diameter: 18 mm; L/D: 44; speed: 20 rpm) (Polartek Polymer Research Technologies Ind. Trade Co. Ltd., Istanbul, Turkey) according to the melting temperature of the polymer (190 °C) and crushed into granules in the crushing machine. Mechanical test specimens were obtained by injection molding method according to the relevant standard, with ten specimens for each group.

Table 2. Composition Ratios of HDPE Samples Produced with Scots Pine Filler

Mechanical Tests Applied

Type 1 specimens were used for tensile testing in accordance with ASTM D638-14 (2014). The tensile test was performed on a Shimadzu AG-IS testing machine at a tensile speed of 5 mm/min, at room temperature (23 °C) and constant relative humidity (50%). Modulus of elasticity in tensile values were calculated during the tensile test. Three-point flexure test was performed in accordance with ASTM D790-03 (2003) using a Zwick/Roell Z50 testing machine with a support spacing of 54 mm, a testing speed of 2 mm/min, room temperature (23 °C), and constant relative humidity (50%). Modulus of elasticity values were calculated during the flexural test.

RESULTS AND DISCUSSION

ANN Model

During the training and testing of the model in ANN, an R² value over 0.90 was achieved. Considering the heterogeneous nature of the wood material, the fact that the studies were conducted in different physical environments, and the expectation that the materials used have different properties, these values are considered reasonable. Table 3 shows the number of data used, the number of neurons in the hidden layers where the model achieved the highest R² value, and the R² values in the training and testing phases.

Table 3. Number of Data Used During Modeling and R² Value Obtained

Results showed that the number of hidden layer neurons used in the ANN and the R² value do not always have a linear relationship, and the highest R² value can be obtained with different numbers of hidden layer neurons. The R² value obtained by the model were higher than 0.90, and thus it is seen that the model can successfully predict the mechanical properties of wood polymer composites.

Atuanya et al. (2014) modeled the mechanical properties of a composite material made of palm wood and recycled LDPE with ANN; similarly, Nasri and Toubal (2024) used ANN to model the mechanical properties of composites made of pine and PP, and R² values of 0.99 were reported for both these studies. Also, Tosun and Sofuoğlu (2023) reported that they obtained mean R² values of 0.92 for training, validation, and testing in modeling the surface roughness of densified wood material with ANN.

Fig. 2. Training and test values for tensile strength

Fig. 3. Training and test values for modulus of elasticity in tensile mode

Fig. 4. Training and test values for flexural strength

Fig. 5. Training and test values for modulus of elasticity

The high R² values in these studies could be achieved due to the use of a small number of samples and constant sample production and test conditions. In this study, although relatively lower R² values were obtained since the production conditions, which are the input parameters, were taken from the literature and varied, it can be said that this value is within acceptable limits. The experimental and ANN-predicted values for training and test of the model are given in Figs. 2 to 5.

It can be inferred from the learning curves presented in Figs. 6 and 7 that the RMSE values had reached a plateau, indicating that learning had occurred adequately.

Fig. 6. Learning curve analysis for tensile strength and modulus of elasticity in tensile mode

Fig. 7. Learning curve analysis for flexural strength and modulus of elasticity

Values Obtained from the Mechanical Test

The results of the mechanical tests performed to determine the reliability of the ANN model are shown in Table 4. The tensile strength and flexural strength values decreased as the filler mixture ratio increased, while the modulus of elasticity in tensile and modulus of elasticity values increased. Balasuriya et al. (2001) found that the flexural and tensile strengths decreased with the increase in the wood flour ratio. Gosselin et al. (2006) reported that increased use of wood flour increased the modulus of elasticity values in composites made with HDPE. The use of a coupling agent had a positive effect on the final values in all cases and, as expected, had an increasing effect on the values. Similar results were found in other studies, and significant increases in mechanical values were observed with the addition of coupling agents and additives such as PE-g-MA (Adhikary et al. 2008; Nourbakhsh and Ashori 2009; Mbarek et al. 2011; Koohestani et al. 2017). Considering the effect of the particle size of the samples on the results, larger particle size had a positive effect on all mechanical properties. Rafighi et al. (2014), Chaudemanche et al. (2018), and Bouafif et al. (2009) found that the values of modulus of elasticity, flexural strength, and tensile strength increased with increasing particle size of wood flour. The mechanical tests showed results in agreement with the literature.

Table 4. Mechanical Test Results

Reliability of the ANN Model

The comparison of mechanical test results with the output values of the ANN model with the same input parameters is shown separately for each mechanical property in Figs. 8 and 9. The MAPE and R² values calculated to determine the reliability of the ANN model are given in Table 5. It is an expected result that the difference between the prediction and the test value was relatively high in some groups due to the diversity of production conditions from which the data used to train the ANN model were obtained. According to similar studies in the literature (Atuanya et al. 2014; İlçe and Singer 2019; Özşahin and Singer 2019; Özbay and Kökten 2020), these values are considered more realistic.

Fig. 8. Comparison of test results and ANN prediction values for TS and MET

Fig. 9. Comparison of test results and ANN prediction values for FS and MOE

After conducting mechanical tests, the R² values for tensile strength were found to be 0.97 or higher, except for the 55H45S-30 and 52H45S3M-30 groups (R²=0.94). The MAPE values varied between 2% and 18.8%. R² values for modulus of elasticity in tensile were 0.97 or higher in all groups. The MAPE values were found to be between 6.2% and 13.3% in all composite groups. For flexural strength, R² values between test results and ANN model predictions were 0.97 or higher in all groups; MAPE values were less than 10% except for two composite groups (67H30S3M-120 and 55H45S-120). For modulus of elasticity, R² values were 0.96 or higher, and MAPE values were 12.9% or below, except for the 70H30S-120 (17.1%) group. R² values close to 1 indicate a strong correlation between predicted and actual values (Bewick et al. 2003). Models with MAPE values higher than 50% are considered faulty, 20 to 50% acceptable, 10 to 20% good, and below 10% very good (Lewis 1983). Based on the R² and MAPE values obtained in this study, it was determined that the ANN model generated good predictions and can be efficiently used to anticipate the mechanical properties of wood polymer composites. Similar results were reported in studies on the prediction of surface roughness and adhesion resistance of wood using ANN (İlçe and Singer 2019; Özşahin and Singer 2019).

Table 5. R² and MAPE Values for Mechanical Properties

CONCLUSIONS

With increasing filler content in wood polymer composites, a decrease in tensile and flexural strength and an increase in modulus of elasticity in tensile and modulus of elasticity were observed.
When coupling agents were added to the mixture forming wood polymer composites, an increase in all mechanical properties was observed.
As the particle size of the filler used in wood polymer composites increased, an increase in all mechanical properties was observed.
Considering the values obtained in the training and testing phases of the model, the established ANN model can successfully predict the TS, TEM, FS, and FEM values of the wood polymer composites.
Because most of the R² values obtained as a result of the mechanical tests were 0.99 or higher and the MAPE values were below 10%, it can be said that the reliability of the model was relatively high.

ACKNOWLEDGMENTS

The authors are grateful to Professor Dr. Uğur Güvenç and Associate Professor Dr. Serhat Duman for help in development of the ANN model. The authors are grateful for the support of the Karabuk University Scientific Research Project Coordination Office Under Grant No: KBUBAP-22-DS-133.

REFERENCES CITED

Adhikary, K. B., Pang, S., and Staiger, M. P. (2008). “Dimensional stability and mechanical behaviour of wood-plastic composites based on recycled and virgin high-density polyethylene (HDPE),” Composites Part B: Engineering 39(5), 807-815. DOI: 10.1016/j.compositesb.2007.10.005

Arwinfar, F., Hosseinihashemi, S. K., Latibari, A. J., Lashgari, A., and Ayrilmis, N. (2016). “Mechanical properties and morphology of wood plastic composites produced with thermally treated beech wood,” BioResources 11(1), 1494-1504. DOI: 10.15376/biores.11.1.1494

ASTM D638-14 (2014). “Standard test method for tensile properties of plastics,” ASTM International, West Conshohocken, PA.

ASTM D790-03 (2003). “Standard test methods for flexural properties of unreinforced and reinforced plastics and electrical insulating materials,” ASTM International, West Conshohocken, PA.

Ataseven, B. (2013). “Yapay sinir ağları ile öngörü modellemesi” İstanbul Kültür Üniversitesi, Öneri.C.10.S.39.101-115.

Atuanya, C. U., Government, M. R., Nwobi-Okoye, C. C., and Onukwuli, O. D. (2014). “Predicting the mechanical properties of date palm wood fibre-recycled low density polyethylene composite using artificial neural network,” International Journal of Mechanical and Materials Engineering, 9, Article no. 7. DOI: 10.1186/s40712-014-0007-6.

Ayrilmis, N., Kaymakci A., and Gülec¸ T. (2015). “Potential use of decayed wood in production of wood plastic composite,” Industrial Crops and Products 74, 279-284. DOI: 10.1016/j.indcrop.2015.04.024.

Badji, C., Soccalingame, L., Garay, H., Bergeret, A., and Benezet, J. C. (2017). “Influence of weathering on visual and surface aspect of wood plastic composites: Correlation approach with mechanical properties and microstructure,” Polymer Degradation and Stability 137, 162-172. DOI:10.1016/j.polymdegradstab.2017.01.010

Balasuriya, P.W., Ye, L., and Mai, Y. W. (2001). “Mechanical properties of wood flake-polyethylene composites. Part 1: Effects of processing methods and matrix melt flow behavior,” Composites Part A: Applied Science and Manufacturing 32(5), 619-629. DOI: 10.1016/S1359-835X(00)00160-3

Bengtsson, M., Oksman, K., and Stark, N. M. (2006). “Profile extrusion and mechanical properties of crosslinked wood-thermoplastic composites,” Polymer Composites 27, 184-194. DOI: 10.1002/pc.20177

Bewick, V., Cheek, L. and Ball, J. (2003). “Statistics review 7: Correlation and regression,” Critical Care 7(6), 451-459. DOI:10.1186/cc2401

Bledzki, A. K., Faruk, O., and Huque, M. (2002). “Physico-mechanical studies of wood fiber reinforced composites,” Polymer-Plastics Technology and Engineering 41(3), 435-451. DOI: 10.1081/PPT-120004361

Bledzkı, A. K., and Faruk, O. (2003). “Wood fibre reinforced polypropylene composites: effect of fibre geometry and coupling agent on physico-mechanical properties,” Applied Composite Materials 10, 365-379. DOI: 10.1023/A:1025741100628

Bouafif, H., Koubaa, A., Perré, P., and Cloutier, A. (2009). “Effects of fiber characteristics on the physical and mechanical properties of wood plastic composites,” Composites Part A: Applied Science and Manufacturing 40(12), 1975-1981. DOI: 10.1016/j.compositesa.2009.06.003

Cay, Y., Cicek, A., Kara, F., and Sagiroglu S. (2012). “Prediction of engine performance for an alternative fuel using artificial neural network,” Applied Thermal Engineering 37, 217-225. DOI: 10.1016/j.applthermaleng.2011.11.019

Chaudemanche, S., Perrot, A., Pimbert, S., Lecompte, T., and Faure, F. (2018). “Properties of an industrial extruded HDPE-WPC: The effect of the size distribution of wood flour particles,” Construction and Building Materials 162, 543-552. DOI: 10.1016/j.conbuildmat.2017.12.061

Delviawan, A., Kojima, Y., Kobori, H., Suzuki, S., Aoki, K., and Shinji, O. (2019). “The effect of wood particle size distribution on the mechanical properties of wood-plastic composite,” Journal of Wood Science 65, article no. 67. DOI: 10.1186/s10086-019-1846-9

Duman, S., Yorukeren, N., and Altas, İ. H. (2018), “A novel MPPT algorithm based on optimized artificial neural network by using FPSOGSA for standalone photovoltaic energy systems,” Neural Comput & Applic 29, 257-278. DOI: 10.1007/s00521-016-2447-9

Effah, B., Reenen, A. V., and Meincken, M. (2018). “Mechanical properties of wood-plastic composites made from various wood species with different compatibilisers,” Eur. J. Wood Prod. 76, 57-68. DOI: 10.1007/s00107-017-1186-7

Ergezer, H., Dikmen, M., and Özdemir, E. (2003). “Yapay sinir ağları ve tanıma sistemleri,” Pivolka 2(6), 14-17.

Fabiyi, J. S., and McDonald, A. G. (2010). “Effect of wood species on property and weathering performance of wood plastic composites,” Composites Part A: Applied Science and Manufacturing 41(10), 1434-1440. DOI: 10.1016/j.compositesa.2010.06.004

Ge, S., Gu, H., Ma, J., Yang, H., Jiang, S., Liu, Z., and Peng, W. (2018). “Potential use of different kinds of carbon in production of decayed wood plastic composite,” Arabian Journal of Chemistry 11, 838-843. DOI: 10.1016/j.arabjc.2017.12.026

Gezer, E. D., Akbas, S., Tufan, M., and Temiz, A. (2016). “Properties of wood plastic composites made of recycled HDPE and remediated wood flour from CCA/CCB treated wood removed from service,” in: The International Research Group on Wood Protection 47^th Annual Meeting (IRG/WP 16-40747), Lisbon, Portugal.

Ghahri, S., Najafi, S.K., Mohebby, B., and Tajvidi, M. (2011). “Impact strength improvement of wood flour-recycled polypropylene composites,” Journal of Applied Polymer Science 124, 1074-1080. DOI: 10.1002/app.34015

Ghahri, S., Najafi, S. K., and Mohebby, B. (2014). “Influence of impact modifier and coupling agent on impact strength of wood flour/recycled plastic composites,” Pro Ligno 10(1) 3-9.

Gönül, Y., Ulu, S., Bucak, A., and Bilir, A. (2015). “Yapay sinir ağları ve klinik araştırmalarda kullanımı,” Genel Tıp Dergisi 25, 104-111.

Gosselin, R., Rodrigue, D., and Riedl, B. (2006). “Injection molding of postconsumer wood-plastic composites II: Mechanical properties,” Journal of Thermoplastic Composite Materials 19, 659-669. DOI: 10.1177/0892705706067486

Gozdecki, C., Wilczynski, A., Kociszewski, M., Tomaszewska J., and Zajchowski, S. (2012). “Mechanical properties of wood-polypropylene composites with industrial wood particles of different sizes,” Wood and Fiber Science 44(1), 14-21.

Haque, M. M. U., Goda, K., Ito, H., Ogoe, S., Okamot, M., Ema, T., Kagawa, K., and Nogami, H. (2019). “Melt-viscosity and mechanical behaviour of polypropylene (PP)/wood flour composites: Effect of pulverization of wood flour with and without water,” Advanced Industrial and Engineering Polymer Research 2, 42-50. DOI: 10.1016/j.aiepr.2018.11.001

Hyvärinen, M., Ronkanen, M., and Kärki, T. (2019). “The effect of the use of construction and demolition waste on the mechanical and moisture properties of a wood-plastic composite,” Composite Structures 210, 321-326. DOI: 10.1016/j.compstruct.2018.11.063

Homkhiew, C., Ratanawilai, T., and Thongruang, W. (2014). “Effects of natural weathering on the properties of recycled polypropylene composites reinforced with rubberwood flour,” Industrial Crops and Products 56, 52-59. DOI: 10.1016/j.indcrop.2014.02.034.

Homkhiew, C., Rawangwong, S., Boonchouytan, W., Thongruang, W., and Ratanawilai, T. (2018). “Composites from thermoplastic natural rubber reinforced rubberwood sawdust: Effects of sawdust size and content on thermal, physical, and mechanical properties,” Hindawi International Journal of Polymer Science 2018. DOI: 10.1155/2018/7179527

İlce, A. C., and Singer, H. (2019). “Doğu kayını ahşabının yüzey pürüzlülüğünün bir yapay sinir ağı ile modellenmesi,” Düzce Üniversitesi Bilim ve Teknoloji Dergisi, 7, 1867-1878. DOI: 10.29130/dubited.564000

Kamdem, D. P., Jiang, H., Cui, W., Freed, J., and Matuana L. M. (2004). “Properties of wood plastic composites made of recycled HDPE and wood flour from CCA-treated wood removed from service,” Composites Part A: Applied Science and Manufacturing 35(3), 347-355. DOI: 10.1016/j.compositesa.2003.09.013

Karmarkar, A., Chauhan, S. S., Modak, J., M., and Chanda, M. (2007). “Mechanical properties of wood-fiber reinforced polypropylene composites: Effect of a novel compatibilizer with isocyanate functional group,” Composites Part A: Applied Science and Manufacturing 38(2), 227-233. DOI: 10.1016/j.compositesa.2006.05.005

Kaymakcı, A. (2015). Çeşitli Güçlendirici Dolgularla Üretilen Ahşap Plastik Nanokompozitlerin Karakterizasyonu, Ph.D. Dissertation, İstanbul Üniversitesi Fen Bilimleri Enstitüsü, İstanbul.

Keskisaari, A., Butylina, S., and Kärki T. (2016). “Use of construction and demolition wastes as mineral fillers in hybrid wood-polymer composites,” J. Appl. Polym. Sci. 133(19), article 43412. DOI: 10.1002/app.43412

Keskisaari, A., and Kärki, T. (2018). “Utilization of industrial wastes from mining and packaging industries in wood-plastic composites,” J. Polym. Environ. 26, 1504-1510. DOI: 10.1007/s10924-017-1052-z

Khonsari, A., Taghiyari, H. R., Karimi, A., and Tajvidi, M. (2015). “Study on the effects of wood flour geometry on physical and mechanical properties of wood-plastic composites,” Maderas. Ciencia y Tecnología 17(3), 545-558. DOI: 10.4067/S0718-221X2015005000049

Koohestani, B., Ganetri, I., and Yılmaz, E. (2017). “Effects of silane modified minerals on mechanical, microstructural, thermal, and rheological properties of wood plastic composites,” Composites Part B: Engineering 111, 103-111. DOI: 10.1016/j.compositesb.2016.12.021

Leu, S. Y., Yang, T. H., Lo, S., F., and Yang, T. H. (2012). “Optimized material composition to improve the physical and mechanical properties of extruded wood-plastic composites (WPCs),” Construction and Building Materials 29, 120-127. DOI: 10.1016/j.conbuildmat.2011.09.013

Lewis, C. D. (1982). Industrial and Business Forecasting Methods, Butterworths, London.

Martins, G., Antunes, F., Mateus, A., and Malça, C. (2017). “Optimization of a wood plastic composite for architectural applications,” Procedia Manufacturing 12, 203-220. DOI: 10.1016/j.promfg.2017.08.025

Mbarek, T. B., Robert, L., Hugot, F., and Orteu, J. (2011). “Mechanical behavior of wood-plastic composites investigated by 3D digital image correlation,” Journal of Composite Materials 45(26), 2751-2764. DOI: 10.1177/0021998311410466

Moreno, D. D. P., and Saron, C. (2017). “Low-density polyethylene waste/recycled wood composites,” Composite Structures 176, 1152-1157. DOI: 10.1016/j.compstruct.2017.05.076

Murayama, K., Ueno, T., Kobori, H., Kojima, Y., Suzuki, S., Aoki, K., Ito, H., Ogoe S., and Okamoto, M. (2019). “Mechanical properties of wood/plastic composites formed using wood flour produced by wet ball‑milling under various milling times and drying methods,” Journal of Wood Science 65, article 5. DOI: 10.1186/s10086-019-1788-2

Najafi, S. K., Hamidinia, E., and Tajvidi, M. (2006). “Mechanical properties of composites from sawdust and recycled plastics,” Journal of Applied Polymer Science 100, 3641-3645. DOI: 10.1002/app.23159

Najafi S. K., and Englund, K. R. (2013). “Effect of highly degraded high-density polyethylene (HDPE) on processing and mechanical properties of wood flour-HDPE composites,” J. Appl. Polym. Sci. DOI: 10.1002/app.39021

Nasri, K., and Toubal, L. (2024). “Artificial neural network approach for assessing mechanical properties and impact performance of natural-fiber composites exposed to UV radiation” Polymers 16, article 538. DOI: 10.3390/polym16040538

Ndiaye, D., Matuana, L. M., Therias, S. M., Vidal, L., Tidjani, A., and Gardette, J. L. (2010). “Thermal and mechanical properties of polypropylene/wood-flour composites,” Journal of Applied Polymer Science 119, 3321-3328. DOI: 10.1002/app.32985

Nitz, H., Reichert, P., Römling, H., and Mülhaupt R. (2000). “Influence of compatibilizers on the surface hardness, water uptake and the mechanical properties of poly (propylene) wood flour composites prepared by reactive extrusion,” Macromol. Mater. Eng. 276/277, 51-58.

Nourbakhsh, A., and Ashori, A. (2009). “Preparation and properties of wood plastic composites made of recycled high-density polyethylene,” Journal of Composite Materıals 43(8), 877-883. DOI: 10.1177/0021998309103089

Nourbakhsh, A., and Ashori, A. (2010). “Effects of particle size and coupling agent concentration on mechanical properties of particulate-filled polymer composites,” Journal of Thermoplastic Composite Materials 23(2), 169-174. DOI: 10.1177/0892705709340962

Oksman, K., and Clemons, C. (1997). “Mechanical properties and morphology of impact modified polypropylene-wood flour composites,” Journal of Applied Polymer Science 67, 1503-1513.

Özbay, G., and Kökten, E. S. (2020). “Modeling of bio-oil production by pyrolysis of woody biomass: Artificial neural network approach,” Journal of Polytechnic 23(4), 1255-1264. DOI: 10.2339/politeknik.659136

Özşahin Ş., and Singer, H. (2019). “Prediction of surface roughness and adhesion strength of wood by artificial neural networks,” Journal of Polytechnic 22(4), 889-900. DOI: 10.2339/politeknik.481762

Prachayawarakorn J., Khamsri, J., Chaochanchaikul, K., and Sombatsompop, N. (2006). “Effects of compatibilizer type and rubber-wood sawdust content on the mechanical, morphological, and thermal properties of PVC/LDPE blend,” Journal of Applied Polymer Science 102, 598-606. DOI 10.1002/app.24324

Raj, R. G., Kokta, B. V., and Daneault, C. (1990). “A comparative study on the effect of aging on mechanical properties of LLDPE-glass fiber, mica, and wood fiber composites,” Journal of Applied Polymer Science 40, 645-655.

Rasat, M. S. M., Wahab, R., Shafie, A., Yunus, A. A. M., Yusoff, M., Ramle, S. F. M., and Zulhisyam, A. K. (2013). “Effect of wood-fiber geometry size on mechanical properties of wood-fiber from neolamarckia cadamba species reinforced polypropylene composites,” Journal of Tropical Resources and Sustainable Science 1(1), 42-50.

Rafighi, A., Dorostkar, A., and Madhoushi, M. (2014). “Investigation on mechanical properties of composite made of sawdust and high density polyethylene,” International Journal of Lignocellulosic Products 1(2), 134-141.

Ratanawilai, T., and Taneerat K. (2018). “Alternative polymeric matrices for wood-plastic composites: Effects on mechanical properties and resistance to natural weathering,” Construction and Building Materials 172, 349-357. DOI: 10.1016/j.conbuildmat.2018.03.266

Soccalingame, L., Perrin, D., Benezet, J. C., Mani, S., Coiffier, F., Richaud, E., and Bergeret, A. (2015). “Reprocessing of artificial UV-weathered wood flour reinforced polypropylene composites,” Polymer Degradation and Stability 120, 313-327. DOI: 10.1016/j.polymdegradstab.2015.07.013

Szandała, T. (2021). “Review and Comparison of Commonly Used Activation Functions for Deep Neural Networks,” in: Bio-inspired Neurocomputing. Studies in Computational Intelligence: Bhoi, A., Mallick, P., Liu, CM., Balas, V. (eds), Vol. 903. Springer, Singapore. DOI:10.1007/978-981-15-5495-7_11.

Tajvidi, M., Shekaraby, M. M., Motiee, N., and Najafi, S. K. (2006). “Effect of chemical reagents on the mechanical properties of natural fiber polypropylene composites,” Polymer Composites 27, 563-569, DOI 10.1002/pc.20227

Tosun, M. and Sofuoğlu, S. D. (2023). “The use of an artificial neural network for predicting the machining characterizing of wood materials densified by compressing,” Bilge International Journal of Science and Technology Research 7(1), 55-62. DOI:10.30516/bilgesci.1240583.

Turku, I., Keskisaari, A., Kärki, T., Puurtinen, A., and Marttila, P. (2017). “Characterization of wood plastic composites manufactured from recycled plastic blends,” Composite Structures 161, 469-476. DOI: 10.1016/j.compstruct.2016.11.073

Turku, I., Karki, T., and Puurtinen, A. (2018). “Durability of wood plastic composites manufactured from recycled plastic,” Heliyon 4, article e00559. DOI: 10.1016/j.heliyon.2018.e00559

Stark, N. M., and Berger, M. J. (1997). “Effect of particle size on properties of wood-flour reinforced polypropylene composites,” The Fourth International Conference on Woodfiber-Plastic Composites, Madison, WI, USA, pp. 134-143.

Stark, N. M., and Berger, M. J. (1997). “Effect of species and particle size on properties of wood-flour-filled polypropylene composites,” Functional Fillers for Thermoplastics and Thermosets, InterTech Conferences, San Diego, CA, USA.

Stark, N. M., and Rowlands, R. E. (2002). “Effects of wood fıber characteristics on mechanical properties of wood/polypropylene composites,” Wood and Fiber Science 35(2), 167-174.

Zhang, X., Hao, X., Hao, J., and Wang, Q. (2018). “Heat transfer and mechanical properties of wood-plastic composites filled with flake graphite,” Thermochimica Acta 664, 26-31. DOI: 10.1016/j.tca.2018.04.003

Article submitted: December 28, 2023; Peer review completed: April 30, 2024; Revised version received: April 30, 2024; Accepted: May 3, 2024; Published: May 17, 2024.

DOI: 10.15376/biores.19.3.4468-4485