Effect of Water Absorption on the Behavior of Jute and Sisal Fiber Biocomposites at Different Lengths: ANN and RSM Modeling

ABSTRACT

Artificial neural networks (ANN) and response surface method (RSM) modeling were used to study the effect of water absorption on the behavior of jute (JF/epoxy) and sisal (SF/epoxy) biocomposite. The specimens were laminated with short fibers of different lengths (5 mm, 10 mm, and 15 mm) of jute and sisal until saturation lasted about 25 days. The kinetic behavior of water uptake was recorded and then compared to the curves predicted by the two models. The results showed that the water absorption of the jute and sisal biocomposites was fairly lower for the 5 mm length compared to the 10 mm and 15 mm lengths. The optimized solution is attained with 1.00 desirability for 14.73 mm fiber length and 482.85 h immersion time. The RSM model error percentages were higher than the ANN model indicating better accuracy in predicting the water uptake of biocomposites by the ANN model compared to RSM. The results of this investigation offer benefits for the application of jute and sisal fiber biocomposites at the elaboration and application stage; it is easy for engineers to identify the swelling factor of these biocomposites without the need for experiments, thus avoiding costs and time.

摘要

采用人工神经网络（ANN）和响应面法（RSM）模型研究了吸水率对黄麻（JF/环氧树脂）和剑麻（SF/环氧）生物复合材料性能的影响. 用不同长度（5 mm、10 mm和15 mm）的黄麻和剑麻短纤维层压试样，直至饱和持续约25天. 记录吸水的动力学行为，然后将其与两个模型预测的曲线进行比较. 结果表明，与10 mm和15 mm长度相比，5 mm长度的黄麻和剑麻生物复合材料的吸水率相当低. 对于14.73 mm的纤维长度和482.85 h的浸泡时间，最佳溶液的期望值为1.00. RSM模型误差百分比高于ANN模型，表明与RSM相比. 本研究的结果为黄麻和剑麻纤维生物复合材料在制备和应用阶段的应用提供了益处; 工程师很容易确定这些生物复合材料的溶胀因子，而无需实验，从而避免了成本和时间.

KEYWORDS:

• Plant fibers
• biocomposites
• water absorption
• optimization
• ANN
• RSM

关键词:

• 植物纤维
• 生物复合材料
• 吸水率
• 优化

1. Introduction

Composite materials markets are rising in virtually every region of the world and in every application sector (Boumaaza, Belaadi, and Bourchak 2022a). Developing these materials based on natural fibers has many advantages, such as abundant availability, lower price, non-toxicity, biodegradability, eco-friendliness, and renewability (Boumaaza, Belaadi, and Bourchak 2022b; Sathiyamoorthy and Senthilkumar 2020). Plant fibers used as reinforcement in composite materials may be considered as an interesting alternative to replace synthetic fibers, particularly glass fibers, because of their low cost. On the other hand, these fibers show non-uniformity in their original characteristics (varying diameter of the fiber and variable length), usually resulting in a dispersion of their mechanical properties (Dembri et al. 2022; Lekrine et al. 2022). In addition to being biodegradable and ecological, they have several mechanical properties that meet industrial requirements. However, the main shortcoming is their sensitivity to humidity, which affects their properties during their use in aggressive environments. The phenomena of water diffusion within composite materials have shown that polymers are subject to different aging types, reflected by changes in specific material properties. By definition, the aging of a polymer results in a slow alteration of material properties, reversible or irreversible, resulting from their own instability or the environmental effect (Belaadi et al. 2022). For more severe conditions (immersion, high temperature), defects may be created within the matrix and/or the fiber–matrix interface, as well as material losses (washing). These damages are irreversible after drying. The moisture influence study on the degradation of mechanical properties is essential to increase the use of natural fibers of plant origin as reinforcement in composites. These polyester or epoxy matrix composites reinforced by plant fibers, notably sisal and jute, raise more and more the health issues level and environmental (Ramesh, Palanikumar, and Reddy 2013). This change reduces the effect of our growing needs on the environment and improves the environmental performance of materials (Summerscales et al. 2010). Chianelli-Junior et al. (2013) discussed sisal fibers mechanical performance and its possibilities as a reinforcement for Recycled High-density Polyethylene (RHDPE). It has been observed that there is no change in tensile and bending test properties of RHDPE with the addition of sisal fibers. However, the ductility decreases in the presence of agave sisalana. The addition of banana fibers to agave sisalana with different mass fractions ranging from 25% to 75% improved mechanical properties and water absorption resistance (Venkateshwaran et al. 2011).

Various water aging studies of biocomposites based on natural fibers have been conducted (Aggarwal et al. 2013; Alamri and Low 2013; Dittenber and Gangarao 2012; Salleh et al. 2012) to infer the water absorption process and the moisture effect on mechanical and chemical properties. Hossain et al. (2011) performed bending tests as well as absorption tests after water aging of jute fiber-reinforced biocomposite. The tests were carried out in distilled water for up to 60 days, and the result showed a remarkable decrease in the rupture stress as well as the bending module as a function of immersion time increase. A sisal/polyester biocomposite was aged in distilled water and then characterized in tensile and impact resistance (Athijayamani et al. 2009). The study deduced that hydrogen bonds between water molecules and fiber cellulose cause a decrease in the material static and dynamic mechanical properties with delamination at the fiber–matrix interface. A recent work conducted by Lekrine et al. (2022) to study the Washingtonia filifera (WF) fiber content effect on the diffusion kinetics of HDPE/WF biocomposites by immersion in distilled water for 21 days (500 h). Their study showed that the absorption phenomenon was rapid in the first phase of tests until reaching the saturation level after 408 h of immersion. The water absorption increases with the increase in WF fiber content and the immersion time. A study of moisture effect on the mechanical properties of the epoxy matrix and flax fiber biocomposites was conducted by Scida et al. (2013) to compare two types of architectures: unidirectional and twill. The authors noted that the twill reinforcement had greater absorption kinetics than the unidirectional and showed that the water absorption at saturation was more important for this architecture type.

Other research work has shown that biocomposites exposed in a wet environment have diminished elastic and rupture properties when water diffuses into the material. This loss related to water absorption has been observed in various biocomposites, such as those based on agave sisalana (Chow, Xing, and Li 2007; Marsh 2003), bamboo (Chen, Miao, and Ding 2009), or flax (Duigou, Davies, and Baley 2009) fibers. Recently, researchers have been active in overcoming these disadvantages by studying biodegradable fibers, such as jute (Ramakrishnan et al. 2018), sisal (Ibrahim et al. 2017) and bamboo (Prasanna, Ramanathan, and Raman 2016), to minimize environmental problems. The analysis of these environmental aging and moisture absorption problems is crucial for the development of biocomposites. Water affects plant fibers as a plasticizing agent that causes swelling and degradation. It is clear from the literature that determining the best conditions for water absorption of natural fiber biocomposites is a difficult task. Consequently, the present investigation focuses on both determination and the optimization of the fiber length of FJ/epoxy and SF/epoxy biocomposites of the water absorption kinetics. These biocomposites will be subjected to immersion tests in distilled water to study the effects of water absorption kinetics as a function of fiber length. The statistical treatment of the results was performed by the application of the analysis of variance ANOVA. This can be accomplished with the RSM-based desirability function to find the best multi-response results for water uptake, and the results can be predicted using RSM modeling.

2. Materials and methods

2.1. Material preparation

In this work, we used an epoxy resin with a density of 1.110 kg/m³; this resin is classified as MEDAPOXY STR supplied by the company GRANITEX (Algeria). The average tensile mechanical properties of this resin used in this paper were described previously (Cherief et al. 2020). The reinforcement materials investigated in this study were JF and SF 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. Three fiber lengths were chosen (5 mm, 10 mm, and 15 mm), and cut as short fibers. The biocomposites are prepared in laminate plates of 300 × 300 mm in size, with a thickness of 6 mm at room temperature (25°C) by contact molding and kept in the mold for 24 h to complete the resin polymerization. The plates obtained are left in the open air for 15 days, and then cut into samples of dimensions 40 × 40 mm using a diamond disc. Finally, the samples obtained were dried in an oven at a temperature of 65°C for 5 h.

Figure 1. (A) Sisal and jute long fibers, (b) Sisal and jute short fibers, (c) epoxy resin used in this work and (d) Biocomposites sample developed.

2.2. Water absorption tests

To measure the water absorption kinetics, tests were carried out on three series containing three samples of each fiber type. The samples have been taken from different biocomposite plates (JF/epoxy and SF/epoxy), and each plate has a different length of 5 mm, 10 mm, and 15 mm, having a volume fraction of 40%. The absorption kinetics has been characterized by immersing the samples in distilled water at room temperature (23°C) until full samples saturation. The time required to reach saturation is about 600 h, either about 16 days. The water absorption monitoring process of biocomposites consists of following over time samples mass evolution, measured at regular intervals throughout the period of immersion. At the time of mass measurement, the samples are taken out of the bath, lightly wiping them out using absorbent paper in order to remove the film of water present on their surface. Then, they are weighed before re-immersing them again. Each measurement during this process has been carried out in less than 5 min in order to limit any changes in the moisture content of the samples as much as possible. The tests were carried out according to the ASTM D1037-99 standard method at room temperature (23°C) using a precision scale of 0.0001 mg KERN PFB. During the immersion time, the gain in mass of a biocomposite sample at time t is characterized by the percentage of water absorbed (Equation 1(1) $W_t=\frac{m_t-m_0}{m_0}\times 100\left(\mathrm{\%}\right)$(1) ) in terms of mass (W_t) defined as the ratio of the difference between the mass of the original sample (that has never been immersed in water, that is, at t = 0 and noted m₀) and mass of the samples after a given immersion time (m_t) to the original mass sample (m₀).

(1) $W_t=\frac{m_t-m_0}{m_0}\times 100\left(\mathrm{\%}\right)$(1)

2.3. ANN modeling

The ANN modeling function is generally applied to replace the polynomial regression functions and provides a complex relationship between input and output variables. The ANN architecture includes a layer comprising the input, hidden, and output layers. These layers contain the neurons that receive the input data (Adda et al. 2021; Boumaaza et al. 2022). The number of neurons is directly related to the number of input variables, the output layer gives the results of the problem for the input values and between these two cells is the hidden layer, containing one or more layers, which is made up of neurons defining the internal activity of the network (Ighalo et al. 2021; Wang et al. 2020). Depending on the interconnection chosen, neurons are classified into two large families; the first illustrates static networks (not looped), and the second represents so-called looped dynamic networks. Neurons are interconnected by synapses that also allow for distributed parallel processing. These synapses have an essential advantage which is the ability to establish complex relationships between numerical variables without having a preconceived model (Wang et al. 2020). It is the training phase that offers this advantage. When the total sum of squared errors is close to zero, the ANN learning algorithm will be more efficient (Boumaaza, Belaadi, and Bourchak 2021; DeMuth et al. 2013). A multilayer feedforward neural network consisting of an input layer, hidden layers, and an output layer is used. The accuracy of the models is determined by ANN, checked by the values of correlation coefficient (R2) and the mean squared error (MSE). The values of these last two parameters the lower and positive they are, the more precise the ANN model is in the prediction. The two parameters are given by Equations (2)(2) $\mathit{MSE}=\frac{1}{n}{\sum }_{i=1}^n\left|{\left({\mathit{X}}_{\mathit{Predicted}}-X_{\mathit{Experiment}}\right)}^2\right|$(2) and (3(3) $R^2=1-\frac{{\sum }_{i=1}^n{\left({\mathit{X}}_{\mathit{Predicted}}-X_{\mathit{Experiment}}\right)}^2}{{\sum }_{i=1}^n{\left({\mathit{X}}_{\mathit{Predicted}}-X_{\mathit{Averqge}}\right)}^2}$(3) ) (Adeniyi, Ighalo, and Marques 2021).

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

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

2.4. RSM modeling

The response surface method (RSM) allows the analysis of multiple regressions by determining solutions for various linear and non-linear regression problems based on experimental data (Krishnaiah et al. 2015). It is a set of mathematical and statistical techniques useful for analyzing problems where several independent variables influence a dependent variable or response (Onu et al. 2021). This method is based on an empirical modeling process and aims at finding the relationship between input and output parameters, including water absorption and the changes in fiber length as a function of biocomposite immersion time. The response parameter values, WAS and WAJ, obtained from several experimental combinations are given in Table 1. The experimental design adopted in the present research is an orthogonal array of central composite designs (CCDs). This plan has been adopted to reduce the number of experiments. This would eventually reduce the time and cost of experiments. Two groups level is envisaged as the length of fibers (L) and the immersion time (W), with four levels in each group (Table 2).

Table 1. Experimental results for absorption of water of sisal and jute fiber biocomposite.

Number of experiment	Input variables		Output variables
	A-times (h)	B-Length (mm)	WAS (%)	WAJ (%)
1	1	5	0.939529	0.551954
2	1	10	1.19108	0.627721
3	1	15	2.48197	0.902231
4	8	5	1.14594	0.595929
5	8	10	1.62752	0.750246
6	8	15	2.74434	0.951718
7	24	5	1.53647	0.703909
8	24	10	2.07015	0.914398
9	24	15	3.26889	1.18949
10	72	5	1.97223	0.835819
11	72	10	2.56338	1.04676
12	72	15	3.82672	1.48734
13	120	5	2.24966	0.876092
14	120	10	2.85177	1.09358
15	120	15	4.09028	1.44796
16	192	5	2.5467	0.882513
17	192	10	2.97227	1.1219
18	192	15	4.26044	1.79087
19	240	5	2.69899	0.884847
20	240	10	3.06144	1.13693
21	240	15	4.27323	1.82629
22	360	5	2.78419	0.888349
23	360	10	3.14338	1.17219
24	360	15	4.27963	1.83008
25	480	5	2.78796	0.8899
26	480	10	3.15462	1.17854
27	480	15	4.28602	1.83324
28	600	5	2.78459	0.8902
29	600	10	3.15673	1.1791
30	600	15	4.2876	1.83577

Table 2. Experiments design adopted in this work.

N°	Factors	Notation	Units	Levels
				L1	L2	L3
1	Immersion time	W	hour	1	240	600
2	Fiber length	L	mm	5	10	15

3. Results interpretation

3.1. Water absorption

Figure 2 illustrates the water absorption results in weight percentage (wt.%) for the JF/epoxy and SF/epoxy biocomposites as a function of immersion time. The values shown in this figure are mean values with standard deviations based on measurements of three samples for each type of material considered. Figure 2a,b represent the progression of water uptake in JF/epoxy and SF/epoxy biocomposites, respectively, at different fiber lengths of 5 mm, 10 mm, and 15 mm. It is clear that the water uptake curves of the different biocomposites are similar and have two different water diffusion zones. In the first zone, water uptake increases proportionally as a function of the immersion time. This zone is quasi-linear for absorption values Wt less than 1% and 3% of JF/epoxy and SF/epoxy biocomposites, respectively. In the linear part of the curve, the water molecules diffusivity depends on the penetration speed of the water into the material. The second zone is non-linear until saturation, meaning that the mass gain gradually increases until the saturation level appears. The analysis also shows that JF/epoxy biocomposites are the least sensitive to water absorption compared to SF/epoxy biocomposites. The high water uptake is mainly due to the high hydrophilicity and internal structure of fibers (Célino et al. 2014; Komuraiah, Kumar, and Prasad 2014). The results obtained indicate that water absorption varies as a function of fiber length withJFL5 and SFL5 having an absorption content of 0.88% and 2.78%, which is relatively low compared to those of JFL10, JFL15, SFL10, and SFL15 (1.17%, 1.82%, 3.12%, and 4.26%, respectively). It is clear that biocomposites reinforced with sisal fibers absorb more water than those reinforced with jute fibers. The SFL15 biocomposite showed a water saturation of 4.4%, while for the HDPE/jute biocomposite, the saturation was 1.72% in the recent work of (Makhlouf et al. 2022). On the other hand, in the studies of (Ramesh, Palanikumar, and Reddy 2016a, 2016b), immersed sisal and jute composites and hybrid composites reinforced with sisal, jute, and glass fibers in distilled water for 5 h and 20 min. The authors found that jute and jute/glass fiber composites absorbed more water than sisal; sisal/glass fiber and sisal/jute/glass fiber composites.

Figure 2. Absorption curve of biocomposites test specimens.

3.2. ANN modeling

Figure 3 illustrates the ANN architecture that consists of three layers, including the input, the hidden layer, and the output. The purpose of this section is to present numerical modeling using the ANN method for the water diffusion process of the two JF/epoxy and SF/epoxy biocomposite materials. The distribution of the samples that have been used for ANN modeling is as follows: 22 for learning, 3 for validation, and 5 for testing. During this study, the Levenberg-Marquardt algorithm was chosen for ANN model training to predict the water uptake of the developed biocomposites. The network was repeatedly configured for learning once the collection of input and target data is performed. Figure 4a,b shows that the learning process stopped after 80 and 18 iterations, when the gradient recorded an optimized validation performance of 0.0014631 and 0.0038486 at the fifth iteration for the jute and sisal biocomposites, respectively (Figure 4c,d). ANN considers the stopping point as an optimal point, beyond which the validation error exceeds the tolerated thresholds. The errors generated in the network throughout training, validation, and testing have been represented using an error histogram, which is distributed over a 20-cell interval (Figure 4e,f and Table 3). ANN classifies the dotted lines in the diagrams as functions that are essential to achieving the objectives or as the best possible choices for constituting the data, so that the best fit is possible for each process in the network as well as for the whole network.

Figure 3. ANN architecture used in this work for (a) SF/epoxy and (b) JF/epoxy composites.

Figure 4. ANN validation, distribution of performance, training conditions, and error results for (a,c,e) SF/epoxy and (b,d,f) JF/epoxy biocomposites.

Table 3. Allocation of samples used in this study for ANN modeling.

		Samples	MSE	R
WAS	Training	22	0.01737	0.9932
	Validation	3	0.00384	0.9981
	Testing	5	0.05138	0.9801
WAJ	Training	22	0.00009	0.9998
	Validation	3	0.00146	0.9997
	Testing	5	0.00802	0.9659

Figure 5 shows ANN modeling results compared to the experimental data for the SF/epoxy and JF/epoxy biocomposites during training, validation, and testing. The variability of the experimental data is illustrated by the coefficient of determination R², which indicates the degree of the model predictions (Belaadi et al. 2020; Boumaaza, Belaadi, and Bourchak 2021). The accuracy level of the correlation for training, testing, and all the data is (0.99328 and 0.99997), (0.98011 and 0.96594), and (0.99077 and 99591) as shown in Table 3, ensuring the reliability of the model developed. Furthermore, the results of the values predicted by ANN are close to the parity threshold (Figure 5); they perfectly agree with those obtained experimentally. Thus, the ANN method can reliably predict the water absorption of biocomposites.

Figure 5. ANN modeling results for (a-c) SF/epoxy and (d-f) JF/epoxy biocomposites.

3.3. RSM modeling

RSM method has been adopted to analyze the results obtained during the experiments and evaluate the contribution of fiber length and immersion time to the water absorption of the biocomposites. Equations (4)(4) $\mathit{WAS}=1.47944+\mathrm{0.009061.}\mathit{\hspace{0.17em}}\mathit{W}-\mathrm{0.134787.}\mathit{\hspace{0.17em}}\mathit{L}-3.9\mathrm{x}{10}^{-5}.\mathit{W}.\mathit{L}-1.1\mathrm{x}{10}^{-5}.\mathit{\hspace{0.17em}}{W}^2+\mathrm{0.015321.}\mathit{\hspace{0.17em}}{L}^2$(4) and (5(5) $\mathit{WAJ}=0.7548+\mathrm{0.001832.}\mathit{\hspace{0.17em}}\mathit{W}-\mathrm{0.05646.}\mathit{L}+10.2\mathrm{x}{10}^{-5}.\mathit{W}.\mathit{L}-3.54552\mathrm{x}{10}^{-6}.\mathit{\hspace{0.17em}}{W}^2+\mathrm{0.005304.}\mathit{\hspace{0.17em}}{L}^2$(5) ) represent the quadratic models generated by the Design-Expert software. These are second-order mathematical equations based on RSM method for water absorption of the various biocomposites, WAS for sisal and WAJ for jute.

(4) $\mathit{WAS}=1.47944+\mathrm{0.009061.}\mathit{\hspace{0.17em}}\mathit{W}-\mathrm{0.134787.}\mathit{\hspace{0.17em}}\mathit{L}-3.9\mathrm{x}{10}^{-5}.\mathit{W}.\mathit{L}-1.1\mathrm{x}{10}^{-5}.\mathit{\hspace{0.17em}}{W}^2+\mathrm{0.015321.}\mathit{\hspace{0.17em}}{L}^2$(4)

(5) $\mathit{WAJ}=0.7548+\mathrm{0.001832.}\mathit{\hspace{0.17em}}\mathit{W}-\mathrm{0.05646.}\mathit{L}+10.2\mathrm{x}{10}^{-5}.\mathit{W}.\mathit{L}-3.54552\mathrm{x}{10}^{-6}.\mathit{\hspace{0.17em}}{W}^2+\mathrm{0.005304.}\mathit{\hspace{0.17em}}{L}^2$(5)

3.4. ANOVA analysis for water absorption

The ANOVA results in Table 4 show that the quadratic model is statistically significant for the WAS and WAJ biocomposites. The p-values that are lower than 0.05 indicate the significance of the model terms. The model terms in this study, A, B, A², and B², were significant. Therefore, the p-values that are greater than 0.05 indicate that the model terms were significant. For values greater than 0.1000, the model terms are not significant. In the same way, the F-values of the WAS and WAJ models are 81.33 and 69.74, respectively, and the corresponding R² values are 94.43% and 93.56%) indicating that the models are significant. The difference between the predicted and adjusted R² values (91.02% and 88.90%, and 93.27% and 92.22%) is less than 0.2 for the WAS and WAJ models, respectively. The adequate accuracy value is the signal ratio noise. It is recommended to have a ratio higher than four. A ratio of 29.5517 and 29.4839 indicates that the signal is adequate for the two models. Therefore, the model can be used to investigate the design space. Consequently, an adequate fit between the independent factors and the responses is obtained with this regression model. The normal probability plots presented in Figure 6a,b proved the effect of minimum scattering occurred on the tendency line caused by the residual points coinciding with the normal probability. The relationship between the predicted and actual response to the biocomposite water absorption factor is presented in Figure 6c,d. The focus was on the residuals between actual and predicted values for model fit. Uniform variation in residuals was required, and the residual points were symmetrically arranged with grouping near the plot center (Rajabi et al. 2019) (Figure 6e,f).

Table 4. ANOVA for the quadratic response model of water uptake data.

Source	DF		SS		MS	F-value	P-value	Remarks
WAS
Model	5		26.25		5.25	81.33	<0.0001	Significant
A-Time	1		5.32		5.32	82.43	<0.0001	Significant
B-Length	1		10.61		10.61	164.42	<0.0001	Significant
AB	1		0.0296		0.0296	0.4578	0.5051	Insignificant
A^2	1		3.80		3.80	58.79	<0.0001	Significant
B^2	1		0.9781		0.9781	15.15	0.0007	Significant
Error	24		1.55		0.0646
Total	29		27.80
	SD = 0.2541 Mean = 2.83 Coefficient of variation = 8.96%							R2 = 94.43 R2 adjusted = 93.27% R2 predicted = 91.02% Adequate precision = 29.
WAJ
Model	5		4.13		0.8259	69.74	<0.0001	Significant
A-Time	1		0.5839		0.5839	49.31	<0.0001	Significant
B-Length	1		2.66		2.66	224.79	<0.0001	Significant
AB	1		0.2046		0.2046	17.27	0.0004	Insignificant
A^2	1		0.4072		0.4072	34.39	<0.0001	Significant
B^2	1		0.1172		0.1172	9.90	0.0044	Significant
Error	24		0.2842		0.0118
Total	29		4.41
	SD = 0.1088 Mean = 1.11 Coefficient of variation = 9.80%							R2 = 93.56 R2 adjusted = 92.22% R2 predicted = 88.90%Adequate precision = 26.4839

Figure 6. Normal probability, predicted normal plot vs actual, residuals for (a,c,e) SF/epoxy and (b,d,f) JF/epoxy biocomposites.

3.5. Interaction among various parameters of the process RSM

The mapping of the response surface of water absorption of sisal and jute fiber reinforced biocomposites as a function of fiber type and immersion time is illustrated in Figure 7a,b. For the biocomposite constituting of sisal fibers, WAS was not greater than 2 when the fiber length was 5–12 mm and the immersion time was 1–108 h. In addition, WAS was greater than 4 when the fiber length was 13–15 mm and the immersion time was 110–600 h (Figure 7a). In the case of the biocomposite constituting of jute fibers, it was found that WAJ did not exceed 0.8 for a fiber length ranging from 5 to 11.5 mm and immersion time of 1 to 105 h. For the latter, it exceeded 1.8 when the fiber length was 13.2 to 15 mm and immersion time of 360–600 h (Figure 7b).

Figure 7. Water absorption data of residuals for RSM and ANN for (a) sisal/epoxy and (b) jute/epoxy biocomposites.

The 3D plots of the two output parameters WAS and WAJ are combined in Figure 7c,d. Again, a greater slope was found for fiber length and immersion time, revealing a significant effect on water absorption. This trend agrees with the ANOVA results reported in Table 2.

3.6. Comparison between RSM and ANN models

ANN and RSM models predicted results and experimentally observed results are shown in Table 5 and Figure 8. Both models adequately described the experimentally observed results, thereby providing an appropriate representation of the experimentally acquired results. However, RSM model error percentages in predicting water uptake were greater than those of the ANN model, as illustrated in Figure 9, so it must be concluded that the ANN model provided higher prediction precision than the RSM model.

Table 5. Comparison between RSM and ANN approach for elaborate biocomposites.

	WAS (%)			WAJ (%)
Order	EXP	RSM	ANN	EXP	RSM	ANN
1	0.4395	1.1974	0.4397	0.3520	0.6074	0.3940
2	0.5511	1.6724	0.5121	0.4277	0.7234	0.4212
3	1.2820	2.9134	1.2969	0.3522	1.1046	0.3568
4	1.1459	1.2588	1.1454	0.5959	0.6236	0.5544
5	1.6275	1.7324	1.6124	0.7502	0.7431	0.7534
6	2.7443	2.9721	2.7130	0.9517	1.1278	0.9134
7	1.5365	1.3951	1.5293	0.7039	0.6592	0.6833
8	2.0701	1.8657	2.1009	0.9144	0.7869	0.9332
9	3.2689	3.1023	3.2930	1.1895	1.1798	1.2453
10	1.9722	1.7709	2.0083	0.8358	0.7552	0.8132
11	2.5634	2.2321	2.5917	1.0468	0.9073	1.0682
12	3.8267	3.4595	3.8274	1.4873	1.3246	1.5381
13	2.2497	2.0968	2.2845	0.8761	0.8349	0.8634
14	2.8518	2.5487	2.8303	1.0936	1.0114	1.1114
15	4.0903	3.7668	4.0607	1.6480	1.4531	1.6513
16	2.5467	2.4921	2.5454	0.8825	0.9238	0.8996
17	2.9723	2.9301	3.0214	1.1219	1.1369	1.1409
18	4.2604	4.1342	4.2296	1.7909	1.6153	1.7402
19	2.6990	2.6933	2.6561	0.8848	0.9626	0.9121
20	3.0614	3.1221	3.0873	1.1369	1.2002	1.1519
21	4.2732	4.3169	4.2827	1.8263	1.7030	1.7766
22	2.7842	2.9781	2.7940	0.8883	0.9882	0.9236
23	3.1434	3.3836	3.1332	1.1722	1.2868	1.1666
24	4.2796	4.5553	4.3147	1.8301	1.8507	1.8305
25	2.7880	2.9512	2.7909	0.8899	0.9116	0.9142
26	3.1546	3.3335	3.0767	1.1785	1.2714	1.1730
27	4.2860	4.4819	4.2694	1.8332	1.8963	1.8522
28	2.7846	2.6125	2.8135	0.8902	0.7330	0.8655
29	3.1567	2.9716	3.1620	1.1791	1.1538	1.1919
30	4.2876	4.0968	4.2891	1.8358	1.8398	1.8608

Figure 8. Comparison between experimental and predicted RSM and ANN models water absorption data.

Figure 9. Desirability graph for optimization.

3.7. Optimization of biocomposites water absorption

In this context, RSM-based optimization is applied to maximize water absorption and ensure that factors are in the appropriate range (Boumaaza, Belaadi, and Bourchak 2021). To this end, many optimal solutions are proposed and the best solution is preferred. The optimal solution is the one that presents the highest desirability (Benzannache et al. 2021). The best-optimized solution is obtained when the desirability is equal to 1.00 for a fiber length of 14.73 mm and an immersion time of 482.85 h, as illustrated by the desirability ramp function (Figure 10 and Table 6).

Figure 10. Desirability graph for optimization.

Table 6. Optimization of response parameters based on RSM.

Number	Time (h)	Length (mm)	WAS (%)	WAJ (%)	Desirability
1	482.852	14.730	4.561	1.883	1.000	Selected
2	507.000	14.750	4.347	1.854	1.000
3	447.891	14.699	4.441	1.850	1.000
4	465.469	14.711	4.420	1.853	1.000
5	360.000	15.000	4.555	1.851	1.000
6	431.691	14.653	4.444	1.840	1.000
7	403.859	14.789	4.500	1.849	1.000
8	516.188	14.984	4.394	1.888	1.000
9	475.758	14.637	4.381	1.842	1.000
10	513.750	14.875	4.367	1.871	1.000

4. Conclusion

In the current investigation, jute and sisal fibers reinforced epoxy composites laminate have been studied to evaluate the effect of fiber lengths and immersion time on water absorption in distilled water. The experimental results led to several conclusions: The water absorption was noticed to increase with the length of the fibers and the immersion time. In contrast, the adsorption process was fast at the initial stage of the experiments to achieve saturation in an immersion time of nearly 192 h. These results revealed that the biocomposites JFL5 and SFL5 showed a percentage uptake of 0.88% and 2.78%, which is comparatively lower than those of JFL10, JFL15, SFL10, and SFL15 (1.17%, 1.82%, 3.12%, and 4.26%), respectively. For biocomposites made of sisal and jute fibers, the obtained R² values were 98.08% and 99.59%, respectively, in all the data and low MSE values (<1% by weight) revealing the developed model accuracy. The predicted ANN and RSM values agree perfectly with those obtained experimentally. However, ANN method can reliably predict the water uptake compared to the RSM of biocomposites made of jute and sisal fibers. Finally, the results of this study have positive implications in terms of cost-effectively designing sisal and jute fibers in biocomposites for future applications. As a continuation of this research, the program of experiments will investigate the fabrication of other biocomposites with different treatments of the fibers using different rates of alkaline treatment and immersion times.

Highlights

• Hybrid biocomposites based on jute and sisal fibers.

• Prediction of water absorption of biocomposites by varying fiber length.

• Optimization using ANN and RSM approaches.

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.

Disclosure statement

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

Additional information

Funding

The authors declare no funding information.