Modeling Moisture Absorption of Flax/Sisal Reinforced Hybrid Biocomposites Using Fick’s and ANN Methods

ABSTRACT

Natural fiber composites are being increasingly used in several fields, owing to their considerable cost, weight, and environmental benefits. The objective of this research is to study the effect of water absorption on the behavior of laminated biocomposites using flax (F) and sisal (S) fiber biocomposites with different stacking sequences (3S and 3F) and fiber orientation at 90° (F090s and S090s), as well as hybridizations (F/4S90/F and S/4F90/S) to reinforce an epoxy matrix. The kinetics of water diffusion within the proposed biocomposites was monitored and analyzed through a model-based optimization approach using artificial neural network (ANN) method. The results of this study showed that the amount of water absorbed by different types of biocomposites increased with time according to Fick’s law. Non-hybrid flax and sisal fiber composites (UD and (0/90°)s oriented) had lower water absorption rates than the hybrid orientation. It was also observed that sisal fiber was most sensitive to water absorption than flax. A high correlation of experimental data with the model was revealed which denoted that the ANN learning process was perfectly performed. Therefore, it was inferred that the ANN approach can accurately predict biocomposites water absorption.

摘要

天然纤维复合材料由于其相当大的成本、重量和环境效益，越来越多地应用于多个领域. 本研究的目的是研究吸水性对使用不同堆叠顺序（3S和3F）的亚麻（F）和剑麻（S）纤维生物复合材料（纤维取向为90°（F090和S090））以及杂交（F/4S90/F和S/4F90/S）增强环氧基料的层压生物复合材料性能的影响. 通过使用人工神经网络（ANN）方法的基于模型的优化方法，监测和分析所提议的生物复合材料中的水扩散动力学. 本研究结果表明，根据菲克定律，不同类型的生物复合材料吸收的水量随时间增加. 非杂交亚麻和剑麻纤维复合材料（UD和（0/90°）s取向）的吸水率低于杂交取向. 还观察到剑麻纤维比亚麻对吸水最敏感. 实验数据与模型具有很高的相关性，这表明ANN学习过程是完美的. 因此，可以推断ANN方法可以准确预测生物复合材料的吸水率.

KEYWORDS:

• Sisal/Flax fiber
• Hybrid biocomposites
• Water absorption
• Fick’s law
• Artificial neural network

关键词:

• 剑麻/亚麻纤维
• 杂交生物复合材料
• 吸水率
• 菲克定律
• 人工神经网络

Introduction

In recent years, innovations in the usage of biocomposite materials have been on the rise (El Hajj et al. 2012), as these materials offer considerable environmental benefits in addition to their remarkable low cost and weight. They are annually renewable, biodegradable, and recyclable (Das et al. 2022; Sekar et al. 2022). The development of natural fibers as reinforcing materials in biocomposites has been a matter of high interest to various industries and created a new durable biocomposite material for industry applications (Sanjay et al. 2019). Plant fiber applications are increasingly being developed mainly in the transport and leisure industries (Boumaaza et al. 2022; Rangappa et al. 2022). These fibers can be derived from wood stems such as flax, jute, and hemp (Boumaaza, Belaadi, and Bourchak 2021a), leaves such as sisal, pineapple, and banana (Chen, Miao, and Ding 2009), and fruits such as cotton, and coconut (Kazi, Eljack, and Mahdi 2020). Moreover, vegetable fibers are increasingly being used in place of glass fibers as reinforcement for composite materials. These fibers offer a judicious alternative from an environmental and mechanical point of view (Belaadi, Laouici, and Bourchak 2020; Benzannache et al. 2021; Tabet et al. 2021; Vigneshwaran et al. 2020).

Owing to the hydrophilic nature of plant fiber, the use of plant fiber biocomposites in exterior structures requires knowledge about their behavior in a humid environment (Liu et al. 2019). Prolonged exposure of plant fiber biocomposites to moisture and temperature can adversely affect their mechanical properties (Sekar et al. 2021). Indeed, these fibers are particularly sensitive to climatic conditions and the environment to which they are operating at Moisture is an important parameter to be taken into consideration, as it can cause serious structural failures and call into question the choice of plant fibers over synthetic fibers (Coniglio et al. 2013).

Several recent studies have investigated the water absorption kinetics of plant fiber biocomposites and its consequences. Scida et al. (2013) studied the effect of moisture on the mechanical properties of epoxy matrix and flax fiber biocomposites by comparing two different types of architectures: unidirectional and twill. The results showed that both materials follow a Fickian absorption behavior. However, the authors noted that the biocomposite with twill reinforcement showed greater absorption kinetics than the one with unidirectional reinforcement. Their study also revealed that the saturation water absorption was greater for this type of architecture. Some studies investigated the effect of glass fiber-natural fiber hybridization on water absorption of composite materials to infer the influence of hybridization on water absorption. Panthapulakkal and Sain (2007) evaluated the effect of hemp-glass hybridization on water absorption, thermal and mechanical properties of polypropylene (PP) matrix short-fiber hybrid composites. They found that hemp-glass hybridization significantly increased the resistance of the composites to water absorption. Yorseng et al. (2020) developed a fully biobased hybrid bioepoxy composite reinforced with kenaf/sisal fibers to fabricate biobased hybrid composites and composite laminates. Their findings showed that the all-biobased hybrid bioepoxy composites reinforced with kenaf/sisal fibers exhibited moderate properties in comparison to the pure bioepoxy composite before and after environmental conditions. Vinod et al. conducted another study on hybrid hemp/jute/hemp composite (Vinod et al. 2022). The authors revealed that hybridization with hemp fiber as a skin layer and jute fiber as a core layer can improve the tensile and flexural strength, but it decreases the inter laminar shear strength. Another study was conducted by Boopalan, Niranjanaa, and Umapathy (2013) to evaluate the mechanical and thermal properties of hybrid epoxy biocomposites reinforced with jute and banana fibers. For this purpose, jute and banana fibers with different weight ratios (100/0, 75/25, 50/50, 25/75, and 0/100) were prepared and then incorporated into an epoxy matrix using an injection molding technique to form biocomposites. The study results showed that the addition of banana fibers in jute/epoxy composites up to 50% by weight increased the mechanical and thermal properties and decreased the moisture absorption. The water absorption behavior of the hybrid sisal/polyester composite is investigated with the numerical model of Fick’s diffusion law (Vigneshwaran et al. 2020).

The water absorption behavior of the flax/epoxy biocomposites was initially linear (Fickian diffusion), then slowed down until the moisture content approached saturation level; the water absorption rate at this point was about 13.5% for each sample. Watanabe et al. (2013) investigated several artificial neural network (ANN) models to evaluate and predict the final moisture content of individual Sugi (Cryptomeria japonica) samples during air drying. They found that the proposed ANNs could create highly nonlinear relationships between inherent wood properties as well as predict MCF with higher accuracy than the model. Ozsahin and Aydin (2014) also used the ANN method to predict the optimum drying temperature of varnish for achieving good bonding in the manufacture of plywood sheets from beech and spruce varnishes, dried at temperatures of 20°C, 110°C, 150°C, and 180°C, and tested in shear. The researchers concluded that high drying temperatures could alter some physical, mechanical and, chemical characteristics of wood and cause drying-related defects.

In this investigation, the ANN model was used to establish nonlinear dependencies among the process input parameters (distilled water immersion test time and sisal/flax fibers filling content) and the response (moisture content absorbed by the biocomposite). Modeling the water uptake of an epoxy matrix loaded with non-hybrid and hybrid sisal/flax fibres has not previously been conducted in literature, thus justifying the topic selection for this paper. Consequently, the novelty of this study is to develop a numerical tool using the Fick’s model and a numerical model using ANN method, to predict and model the water absorption kinetics of plant fibers reinforced biocomposites. The tests were carried out on laminated biocomposites of flax and sisal fibers on one hand and hybrids on the other hand with a volume fraction of 40%, using an epoxy resin.

Materials and methods

Materials and manufacturing process

The reinforcement materials investigated in this study were flax and sisal fibers (Figure 1). These fibers present a very interesting alternative, as they have the advantages of having highly competitive mechanical characteristics and being easily accessible. The constituents for this study were provided by the natural fiber cable and cord manufacturing plant in Blida, Algeria. The characteristics of these utilized materials are described in Table 1. The epoxy resin type MEDAPOXY STR with a density of 1.110 kg/m³ was used with an HY951 hardener to form the matrix; both these ingredients were provided by a company called GRANITEX Alger (Algeria) (Belaadi et al. 2020; Cherief et al. 2020, 2021). These laminates were prepared at room temperature (25°C) through contact molding and the setup was retained in the mold for 24 hours to ensure complete polymerization of the resin. They were produced in the form of 300 mm by 300 mm plates, and oriented as defined in Table 2 (Belaadi, Amroune, and Bourchak 2019) as follows: 3 unidirectional (UD) plies with cross (0/90°) orientations and 6 plies for hybrid laminates placed as S/4F90/S and F/4S90/F. The obtained plates were left exposed to air for 7 days, and then cut into samples of dimension 30 mm by 30 mm using a diamond disk. Finally, the cutout samples were post-cured in an oven at a temperature of 70°C for 4 hours.

Figure 1. (a) Unidirectional sisal fibers, (b) unidirectional flax fibers, (c) epoxy resin used in this work and (d) example of biocomposites elaborated in this study.

Table 1. Properties of the sisal, flax fibers and matrix used in the present work.

Material	σ (MPa)	ε (%)	E (GPa)	Diameter (µm)	ρ (g/cm^3)	Reference
Sisal	462 ± 71	7.83 ± 1.25	7.47 ± 1.37	240 ± 27	1.42	(Belaadi et al. 20132013)
Flax	1036 ± 621	1.68 ± 0.95	54.52 ± 22.70	18-20	1.5	(Belaadi, Amroune, and Bourchak 2019)
MEDAPOXY	43.13	4.03	2.475	-	1110	(Belaadi et al. 2020)

Table 2. Fiber lay-up and volume fraction of the various considered hybrid biocomposites.

Laminate	Denomination	Lay-up	Vf (%)
Unidirectional	3S	[0]3	25
	3F	[0]3	26
Cross-ply	S090	(0/90)S	30
	F090	(0/90)S	32
	S/4F90/S	(0/902)S	38
	F/4S90/F	(0/902)S	40

Water absorption tests

Several methods can be used to estimate the amount of moisture absorbed by composite materials. The most widely used method is the gravimetric method, which involves testing a mass of the sample before and after aging (Dhakal, Zhang, and Richardson 2007). In this method, the evolution of the sample mass is observed in regular intervals over a total period exceeding 600 hours. For this purpose, a series of sample laminates of different thicknesses were taken from each biocomposite plate while their kinetics of water absorption was measured. The samples were weighed and immersed in distilled water, then removed from the bath and lightly wiped with absorbent paper to remove any water film present on the surface. The weighed samples were then immersed in water again and the process was repeated until complete saturation was reached. These tests were carried out according to ASTM standard methods (D1037–99, ASTM, 1999) at an ambient temperature of 23°C and a relative humidity of about 49%. The mass samples were taken on a KERN PF Bis balance with a precision of 0.0001 mg. The absorption rate of the composite material was calculated by finding the difference between the mass of the sample exposed to water and its mass in the unaged state.

Fick’s model

Moisture sensitivity of natural fiber composites is a major problem in various sectors, including automotive and marine industries (Gourier et al. 2014; Mokhothu and John 2015). This is because the presence of water molecules in these materials can affect their mechanical properties, thereby limiting their use. Many researchers have studied the water absorption kinetics of natural fiber biocomposites extensively (Cherif et al. 2016). Although the kinetics of water diffusion in a material can be modeled using different laws, the most widely used law is Fick’s law (Fick 1995). This law is widely used in the literature mainly due to its simplicity. Thus, after testing different mass samples over time, the characteristics related to the water absorption of the biocomposite were determined by finding out the mass percentage of water absorbed, M_t defined by equation 1,

(1) $\mathit{Water}\mathit{absorbed}\left(M_t\mathrm{\%}\right)=\frac{w_t-w_0}{w_0}\times 100$(1)

where w₀ is the mass of the unmerged sample at t = 0 and w_t is the mass of the samples after a specific immersion time.

The absorption of water within a material is governed by the maximum absorption capacity (M_t) that corresponds to the maximum mass of water absorbable by the material and the diffusion coefficient (D) that expresses the speed of penetration into the environment considered. D is determined from the linear part of the gravimetric water absorption curve by the Equation (2)(2) $\mathit{D}=\mathit{\pi }{\left(\frac{k}{4M_m}\right)}^2$(2) . Where, (k) is the slope of the linear part of the curve representing$M_t=\mathit{f}\left(\sqrt{\mathit{t}}/\mathit{h}\right)$.

(2) $\mathit{D}=\mathit{\pi }{\left(\frac{k}{4M_m}\right)}^2$(2)

ANN modeling

Artificial neural networks (ANNs) provide a model of computation based on biological neural networks. ANNs are composed of many interconnected parallel neurons that can optimally interact with each other, owing to their high precision and optimization. An ANN is generally made up of three layers (input layer, output layer, and hidden layer). Input layers: These layers contain the neurons that receive the input data (Adda et al. 2021; Safa and Samarasinghe 2011) where the number of neurons is directly related to the number of input variables. Then there is the output layer which is the layer that gives the results of the problem and between these two compartments are the hidden layers which are made up of neurons defining the internal activity of the network (Ighalo et al. 2021; Wang et al. 2020). ANNs are classified into two large families and based on the chosen interconnection; the first illustrates static networks (not looped), and the second represents so-called looped dynamic networks. Synapses interconnect neurons and allow distributed parallel processing. These synapses have the ability to establish complex relationships between numerical variables without requiring a preconceived model (Macedo et al. 2015). The ANN learning algorithm is more efficient when the sum of the squared errors is close to zero (Boumaaza, Belaadi, and Bourchak 2021b). The accuracy of the model given by ANN can be checked using the values of the correlation coefficient (R²) and the mean squared error (MSE). The lower the values of the MSE, the more precise the ANN model is in the prediction. The two parameters are given by equations 3 and 4 (Adeniyi, Ighalo, and Marques 2021).

(3) $\mathit{MSE}=\frac{1}{n}{\sum }_{i=1}^n\left|{\left({\mathit{X}}_{\mathit{Predicted}}-X_{\mathit{Experiment}}\right)}^2\right|$(3)

(4) $R^2=1-\frac{{\sum }_{i=1}^n{\left({\mathit{X}}_{\mathit{Predicted}}-X_{\mathit{Experiment}}\right)}^2}{{\sum }_{i=1}^n{\left({\mathit{X}}_{\mathit{Predicted}}-X_{\mathit{Average}}\right)}^2}$(4)

Analysis of variance (ANOVA)

The results of the experiments were assessed using statistical analysis. One-way analysis of variance (ANOVA) was used to determine if there were any significant differences between the groups with respect to the architectures of the investigated biocomposites. The effectiveness of the ANOVA approach can be evaluated using different statistical parameters such as interval of confidence (CI), F-Fisher test, and probability. The F-test was utilized at a 0.05 level of significance, which represents a 95% confidence level.

Results and discussion

Water absorption

The water absorption rates recorded for the different biocomposites that were examined are presented in Figure 2. In this figure, the increased of water absorption rates of flax and sisal fibers at different architectures is shown. It can be observed that the absorption curves of each biocomposite showed two different absorption areas. In the first area, the water absorption increases proportionally to the square root of the immersion time expressed in hours. This area is almost linear for water absorption values below 3%. This linear part of the curve gives information about the diffusivity of the water molecules, which depends on the speed at which the water penetrates the material. The second part is non-linear until saturation, i.e., the mass absorption increases progressively until the appearance of an equilibrium level called the saturation level. This bearing indicates the water mass absorbed by the material when the immersion time was up to 1400 h^0.5 of immersion.The saturation weight gain M_m and the diffusion coefficient D are shown in Table 3. The analysis also shows that the 3S and 3F architectures are the least sensitive to water absorption, as the unidirectional layers exhibit good performance in the fiber direction. The 3S architecture has a lower percentage of absorption than the 3F architecture. This is due to the fact that the chains of the sisal fiber are connected by regularly reproduced hydrogen bonds between the groups, so their absorption is minimal (Célino et al. 2014; Komuraiah, Kumar, and Prasad 2014). The results obtained are very close to those reported by various other authors (Couture, Lebrun, and Laperrière 2016; Monti et al. 2017). It was also found that the orientation of the fibers had a significant impact on the water absorption: the S090s biocomposite had an absorption rate of 3.15%, while for the F090s biocomposite, this rate was 3.56%. Similarly, for hybrid reinforcements, the amount of water absorbed by the F/4S90/F biocomposite was higher than that of the S/4F90/S reinforcement; they were about 4.7% and 4.4%, respectively.

Figure 2. Water absorption results for biocomposites study in this work.

Table 3. Diffusion parameters for biocomposites laminate elaborated: Slope (k), diffusion coefficient (D) and water content at saturation (M_m).

Material	k	Dx10^‒^6(mm^2/s)^2	Mm (%)
F/4S90/F	0.03772944	9.99124	4.7333
S/4F90/S	0.01702022	2.846043	4.4683
S090	0.00793975	97.5233584	3.5617
F090	0.01057549	2.2009689	3.1579
3S	0.01171553	4.2703864	2.5115
3F	0.01453343	9.9684186	2.0392

ANN modeling

The ANN network was structured as a multilayer network with artificially generated neurons in each layer. The ANN is a multilayer perceptron (MLP) consisting of three layers. The first layer (input layer) consists of one neuron representing the independent variable that was already used in the Fick’s model. The second layer is a hidden layer. The last layer (output layer) is the response layer consisting of six neurons. The number of neurons required in the hidden layer was chosen to minimize the prediction error and also to reduce the number of neurons. A hidden layer with 10 neurons was selected to create robust ANN models where time was considered as an input parameter and the different architectures realized were considered as output parameters. The architecture of the developed ANN model is shown in Figure 3.

Figure 3. ANN architecture used in this work.

The samples were distributed among 28 individuals for training and 4 for testing and validation (Figure 4). During this study, the Levenberg-Marquardt algorithm was selected for ANN model training to predict the water uptake of the elaborated biocomposites. The network was repeatedly configured for training once the input and target data collection were performed. The discrepancies between the experimental data and the outputs generated by the ANNs are illustrated in Figure 5. It can be noted that the MSE values were elevated at the initial stage of the ANN formation, but then got reduced in the later stages of the ANN training phase. The training of the ANN network finished with the MSE value, reaching the least in the fourth time epoch and when the gradient recorded an optimized validation performance of 0.0045858 at the fifth iteration. The graph shows the position where the lines representing the training, validation, and testing phases join the perfect line (best), indicating that the ANN network formation phase ended successfully.

Figure 4. Training, validation and testing MSE and R² values for water absorption data.

Figure 5. ANN validation results for biocomposites data.

The performance data of ANN in the training operation is presented in Figure 6. It was observed that the initial error in the training process of ANN network occurred at the sixth epoch, continued for 6 repetitions, and ended at the tenth epoch. In the validation check that was conducted over a total of 10 epochs, the occurrence and recurrence of errors after the sixth epoch indicated an ideal fit of the data. An important part of the performance analysis of the ANN training process is the analysis of the error percentages of the ANN estimates. The difference in the error rates between the target value and the output values generated in the network throughout the training, validation, and testing phases are shown by means of an error histogram, which is distributed over an interval of 20 cells (Figure 7). The lowest error values were found on and around the zero-error line when looking at the graph. Small error levels are a further indication of successful ANN training phase completion with minimum error levels.

Figure 6. Training condition.

Figure 7. Distribution of errors results for biocomposites data.

Figure 8 shows a comparison between the ANN modeling results and the experimental data for linen/sisal/epoxy biocomposites during formation, validation, and testing. The variability of the experimental data is illustrated by the coefficient of determination, R2, which indicates the degree of prediction of the model (Boumaaza, Belaadi, and Bourchak 2021b). In the formation process, the proximity of the data points to the zero-error line shown by the dashed line and the line of equality drawn in blue, as well as the R-value of 0.99967, suggest that the formation process of the ANN network was perfectly performed. It was noted that the data points were close to the equality line and zero-error line, and that the obtained R-value was 0.99801 in the ANN network validation process. This proves that the ANN network validation phase was performed with minimal error rate. The R-value for the test phase is 0.99942. The data obtained in the test phase was located near both lines. It can be clearly seen from the graph that the test phase of the ANN network was completed with a very low error rate. In addition, it can be observed that the data obtained in all three phases of the ANN were along the fit line and the zero-error line. However, the R-value for all the data was obtained at 0.9995. The results of the ANN predicted values were close to the parity threshold (Figure 9) and they are in perfect agreement with those obtained experimentally (Ighalo et al. 2021). Thus, the ANN method gives a more reliable prediction of the water absorption levels of biocomposites.

Figure 8. ANN modeling results.

Figure 9. Experimental and predicted values of ANN model.

ANOVA analysis

Table 4 presents the water absorption results of flax and sisal fibers biocomposites taken by varying fiber architectures and determining whether significant differences existed between biocomposites consisting of 3-ply unidirectional UD (3S and 3F), cross orientations (F090_s and S090_s), and 6-ply hybrid laminates (S/4F90/S and F/4S90/F) with CI = 95%. The statistical terms used for the ANOVA were between-group (BG), within-group (WG), sum of squares (SS), degree of freedom (DF), mean square (MS), F-test statistic (F), and critical value (F_crit) in order to confirm the effect of biocomposite structure type on the responses. The F-test statistic was used to determine if significant BG differences occurred in the range of the different biocomposites that were tested. If F was found to be superior to F_crit, the null hypothesis would be dismissed, resulting in a significant variation in the mean value. This would indicate that the architectures were not all equally successful. If the value of F was found to be less than F_crit, then the null hypothesis cannot be dismissed, which indicates that the results do not show a significant variation in the mean values. The significance of the P-value = .000 (p < .05) is less than the significance level (0.05); also, F = 24.02 is superior to the critical F = 2.26. Therefore, the null hypothesis that the means are equivalent can be rejected. The ANOVA results showed that there was a significant difference in the mean values between the different biocomposite structures as P (probability) is less than 0.05 and the F-value is above the critical value (Ighalo et al. 2021). The results suggest that the flax and sisal fiber biocomposites with different structures incorporated in an epoxy bio-resin exhibit a statistically significant change in water absorption. Group comparison reveals that the variance for the flax biocomposites is significantly lower than for the other sisal biocomposites. Additionally, the results identified whether the type of fiber as well as their orientations were able to provide a statistical improvement in the water absorption of these biocomposites.

Table 4. Variance analysis of moisture absorption at 95% level of significance.

Source of variation	Sum of squares (SS)	Degree of freedom	Mean square (MS)	F-test statistic (F)	Probability (P)	Critical value (Fcrit)
Between-group	120.84903	5	24.16981	24.02229	4.3813E-19	2.25706
Within-group	211.28953	210	1.00614
Total	332.13856	215

Conclusion

Although the use of plant fibers in place of synthetic fibers can provide several benefits, the hydrophilic nature of plant fibers necessitates one to have knowledge about their performance in a humid environment, before actually putting them to use. The current study investigated the effect of water absorption on the behavior of flax/sisal/epoxy biocomposites built in different architectures. The following are the inferences drawn from the study: (1) The water absorption rates of both non-hybrid flax and sisal fiber composites (UD and (0/90°)_s oriented) was lower when compared to the hybrid orientation. (2) Sisal fibers have a significant influence on the water diffusion process. Their water absorption was found to be lower than that of flax. (3) The water absorption pattern of all developed biocomposites was in accordance with the Fickian model. (4) ANN predicted values were close to the parity threshold; they were in perfect agreement with those obtained experimentally. Therefore, it was concluded to be the most appropriate approach as it provided a reliable prediction of the water absorption kinetics of flax/sisal biocomposites with a tolerable error quotient. The ANOVA results indicate that the flax and sisal fiber composites built into different structures and incorporated into an epoxy bio-resin revealed a statistically significant change in water absorption.

Highlights

• The sisal fibers have a significant influence on the water diffusion process.

• The water absorption pattern of all developed biocomposites was in accordance with the Fickian model.

• The non-hybrid flax and sisal fiber composites exhibited low water absorption over the hybrid composites.

• The ANN model based n the LM algorithm has the highest performance.

Ethics approval

The work contains no libelous or unlawful statements, does not infringe on the rights of others, or contains material or instructions that might cause harm or injury.

Acknowledgment

This project was funded by the Deanship of Scientific Research (DSR), King Abdulaziz University, Jeddah, under grant No. (D-1001-135-1443). The authors, therefore, gratefully acknowledge DSR technical and financial support.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This project was funded by the Deanship of Scientific Research (DSR), King Abdulaziz University, Jeddah, under grant No. (D-1001-135-1443).