An Unconventional Approach for Analyzing the Mechanical Properties of Natural Fiber Composite Using Convolutional Neural Network

Department of Electronics and Communication Engineering, Saveetha School of Engineering,SIMATS, Chennai 602105, Tamil Nadu, India Department of Electronics and Communication Engineering,Vignan’s Foundation for Science, Technology and Research (Deemed to be University), Vadlamudi, Guntur, Andhra Pradesh 522213, India Department of Electronics and Communication Engineering, Sri Shakthi Institute of Engineering and Technology, Coimbatore-641062, Tamil Nadu, India Department of Mechatronics Engineering, T.S. Srinivasan Centre For Polytechnic College and Advanced Training, Chennai, Tamil Nadu, India Department of Mechanical Engineering, Gokaraju Rangaraju Institute of Engineering and Technology, Nizampet, Hyderabad, India Department of Mechanical Engineering, Faculty of Manufacturing, Institute of Technology, Hawassa University, Awasa, Ethiopia


Introduction
In the current situation, industries were mostly focused on the sustainable idea of manufacture to minimize the nonrenewable resources procedure and adopt the process of ecofriendly materials by waste reuse or recycling. To prepare composite materials, extensive work is done by the researcher in several areas of the world [1]. Polymer and natural fiber composites have been selected as ideal materials due to necessary characteristics including reusability, recyclability, long-term constancy, and widespread accessibility at a reasonable cost. When compared to synthetic fibers, natural fibers have inferior qualities such as significant water absorption, weak bonding, low durability, and low mechanical and thermal properties on industrial applications. e optimization and modeling computational engineers employ some approaches to deal with the mechanical difficulties that arise in science and engineering research [2].
Researchers in the field of materials engineering are investigating these computational methods that may be used to model and optimize the many properties of composite materials reinforced with natural fibers. Due to various environmental concerns, the goal is possibly replacing synthetic fibers and environmental considerations [3]. Fiber treatment, nanofilter addition, and hybridization are a few technologies that have been developed to overcome these restrictions. e use of these technologies to manufacture natural fiber composite (NFC) materials has broadened their range of applications such as household, structural, sporting, automobile, aerospace, and other manufacturing applications in recent decades [4]. e process of analyzing the natural fiber's quality and the composites that arise is complicated, which makes it challenging to establish generic methods for modeling and optimizing composite qualities [5]. e use of current computational approaches for studying the characteristics of NFCs has proven to be helpful in the modeling and optimization of composite materials. e shielding process of mechanical properties on natural fiber composites is done by using the electromagnetic interference (EMI). e shielding of EMI efficiency, microstructure, flexibility, tensile, and inner bonding strength of the composites are examined. e EMI excellent shielding performance and mechanical properties use the composite materials in the EMI protection fields. Researchers and producers can easily determine the best mix of constituent materials to balance the strength and cost of the resultant materials by using computational or mathematical models. e analysis of the relationship between the input and output parameters in materials processing is part of composite material modeling.
A deep neural network is inspired by the biological neural network architectures processing the learning data obtained from the user.
is network usually has input, hidden, and output layers [6]. e DNN is used to succeed in finite element analysis while solving the optimization problem of graded functionality plates. A DNN for forecasting foamed concrete compressive strength models was trained on a huge number of photos, and then their learning ability was transferred to pavement fracture identification using digitized pavement surface images.
A convolutional neuronal network (CNN) model makes it possible to quantitatively predict the mechanical properties of the compound through maximum fraction space volume [7]. CNNs are a type of DNN, and they were chosen because of their performance in image recognition tasks. In the following sections, taking into account all the volume fractions considerably expands the sample space. Also, from a practical perspective, having a model that forecasts composite performance of actual standards instead of classes list derived from an organization model is more helpful [8]. In addition, using an optimization approach depends on GA with the model CNN created for optimizing mechanical properties in terms of stiff and soft material volume fraction and spatial distribution in the microstructure. e remaining part of this paper is as follows. Section 2 describes the previous effort that was done by the scholars in this domain with the various experimental tasks; Section 3 offers the materials and methods, Section 4 represents the evaluation and performance of the result and discussion; and Section 5 represents the work achieved in conclusion. Munde and Ingle (2015) evaluates interest in minimum environmental and inexpensive pollution. is research is focused on biocomposite mechanical properties using the models of available theory and investigational authentication of mechanical properties. Parallel and series, modified Halpin-Tsai, Hirsch model, and Halpin-Tsai are the theoretical models used for mechanical tensile properties evaluation. To prepare the sample specimen by using the compression molding machine, the characterization of the mechanics is performed ASTM D638-01 through the universal test machine.

Literature Survey
e results compared the assumed modulus of tensile and strength of tensile with the characteristics of mechanical properties [9]. De Araujo Alves Lima et al. (2020) analyzed alkalization influence. By combining the behavior of alkalization and silanization on the adhesion interfacial and intralaminar hybrid composite's mechanical properties, the single tensile fiber examined fragmentation and also performed the tests of the short beam. Finally, the analysis of SEM is used to research the surface chemical influence treatment used on the natural fibers morphology, to improve the fiber tensile strength by the research of chemical treatment and alkaline treatment on fibers such as ramie, curaua, and jute. After the fusion treatment, there are better results from the presented sisal fibers. e best treatment of being superficial for individual types of composites is the alkaline treatment for fibers obtaining better results for the untreated samples [10].
Sridharan and Muthukrishnan (2013) evaluate, during the machining and production, whether biocomposites require less energy. e quality of fiber is enhanced by using different chemical treatments. e process of mercerization is improving the composite mechanical properties significantly.
is research compares the machinability of reinforced composite jute fiber organized using processed and untreated alkaline fiber. e value of the hole is depending on a comparison of the delamination factor achieved through digital image processing techniques. e experiments are based on the factorial design conducted at the various levels of feed rate and velocity. e ANOVA (analysis of variance) is used to research the chemical treatment influence on delamination. To find the condition of optimum drilling, the Grey Relational Analysis (GRA) is employed [11]. Sultana  Composites (JFRCC). e water-cement relationship (W/ C), the capacity of jute fibers, and the length are the most dependent properties. e data for JFRCC properties are matrix-design-based and these data are used to plan, compare, and estimate the models. us, the act of the SVR model is better than RSM and ANN models with different performance and parameters measuring to predict both tensile and compressive strengths [12]. Wang, Wu, and Wang (2010) evaluate the fiber aspect ratio effect on steel fiber reinforced concrete (SFRC) mechanical characteristics.
is analysis exposes the aspect ratio of the best parameter for strength in every batch. e steel fibers addition can relatively increase the effect of ductility. e erosion-based algorithm is used for numerically simulated steel projects on two SFRC targets such as perforation and crater. Due to the postyield injury, the behavior of softening model has tabulated the effective plastic strain and effective stress. ese results show that the hydrodynamic model described the SFRC responses under effect loading. In addition, the higher appearance ratio of the fiber engages high energy causing reduced craters in the SFRC goal and less remaining rate of the projectile [13]. Table 1 shows the various mechanical properties of natural fiber composites.

Materials and Method
e composite material of natural fibers was modeled using certain computation methods in this research. e materials include mechanical properties such as jute, sugarcane, ramie, pineapple, and banana fibers. e various parameters are observed from the mechanical characteristics of natural fiber composites. e natural hybrid composites are fabricated through the process of vacuum-assisted rein transfer molding (VARTM). e composites presented better electromagnetic interference (EMI) shielding performance with better mechanical properties. e properties are inherited from the aluminum sheets and natural fiber composites. Currently, the natural fiber composites have mechanical properties of inferiority when compared to the parts of the synthetic counter [21]. e two or more fibers that are dissimilar are get mixed in a common matrix to create a composite panel in the hybrid composite. e hybridization is achieved by merging the different lengths and dissimilar diameters for small fibers. is method aims to evaluate the mechanical properties and shielding performance of natural fiber composites. Next, the content of the nanofiller is used to overcome the limitations of the composite material that is reinforced in natural fibers to enhance dimensional stability. Figure 1 shows the manufacturing of man-made fibers from raw materials.

Modeling of Mechanical Properties of Natural Fiber
Composites. Modeling the mechanical properties of natural fibers composite materials is considered a complex process due to different parameters including the matrix type, used fiber, overall composition, the process of manufacturing, the process of fabrication, and planned applications. e composites offered electromagnetic interference (EMI) for shielding performance with better mechanical properties.
Researchers have been able to build these improved modeling approaches with better precision due to recent advances in processing capacity. e goal is to review some computational modeling methodologies that have been employed in the study of fiber-reinforced composite materials.

Rule of Hybrid Mixture (RoHM).
e Rule of Hybrid Mixture (RoHM) model considers that the structure is made up of more than one fiber embedded in one matrix material as a hybrid composite [22]. Applying the isostrain state condition to a reinforced hybrid composite with the two fibers yields equation (1), where ε co , ε m1, ε m2 denote the total material strain, initially fiber strain, and next fiber strain. By assuming no interplay between each type of fiber, the equation of RoHM is modeled to compute the hybrid composite modulus as shown in equation (2). en L co , U r1, and U r2 represent the hybrid composite elasticity and qualified volume fraction of hybrid composite for initial and next fibers.
Equations (1) and (2) are under the following conditions: , where U f is the complete reinforcement fraction volume upon the (U m1 + U m2 ), which is basically used as the volume fraction of reinforcement for the elasticity computation (L r1 and L r2 ) of the both exclusive composites. Tensile strength and density fraction of banana module/sisal fiber in variable ratios using experiments and applying RoHM are considered a predictor tool. e comparison between the exact experimental results and the RoHM tensile strength results is shown in Figures 2 and 3 which allows for concluding the methods.

Halpin-Tsai Model.
e model of Halpin-Tsai is employed to expect the composite material elastic modulus using fiber elasticity, materials of the matrix, and geometry as input parameters [23]. e empirical approach is considered, the field approach of self-consistency from the basic model. e model offers that the modulus of tensile and strength of tensile composite material is given in (3) and (4), where S a, S d, I a, I d represent the composite elastic moduli, matrix elastic moduli, matrix tensile strength, and composite tensile strength, respectively. U m represents the volumetric fraction of fibers in composites. e parameter value of F � J − 1 and J � 1 + 2s/t, where the diameters of fiber and length are denoted as t and s, respectively. e parameter c is given by equation (5) and (6). e parameter of F measures the geometry of fiber, distribution of fiber, and loading of fiber condition. I m and S m represent the moduli of elastic and tensile strength of the fiber, respectively. e

Advances in Materials Science and Engineering
Halpin-Tsai model's essential equations are used to explain the issues analytically to determine the natural occurrence of the structure's fluctuation. e results obtained after using the optimization revealed a higher frequency attained by the hybrid nanocomposites compared with nonhybrid composites.

Model of Hirsch.
e model is concerned with the sequence stacking and fiber orientation in a matrix material [24]. A diagram of the Hirsch model is given in Figure 4 which shows the parallel and series mixture models. us, equation (5) shows the composite characteristics, where I a, I d , and I m represent the composite strength of tensile, matrix material strength of tensile, and fibers tensile strength. U d denotes the matrix volume fraction and U m denotes the fraction of fiber volume.
z in the equation indicates a variable amount which gives the load transfer between the fiber and the array. e Hirsch model is used to determine the best characteristics of tensile hybrid composite, which reveals that the mechanical performance is modifiable between the values 0.37 and 0.55. Figure 5 shows that A2, A3, A2H11, and A3H12 are investigated using the micro-A1 sheets.

Electromagnetic Interference for Shielding Process.
e EMI shielding occurs through the radiation reflection. e EMI shielding is the most popular method by using metal sheets and coating. e metal sheets are regularly used, but some suffer from drawback on the bad mechanical properties and resistance corrosion. e metal sheets are converted into the natural fiber composites. e EMI shielding process results from using metal sheets and thus having better mechanical properties given by the natural fiber composites. e metal sheets are protected to avoid the electrochemical corrosion by natural fiber composites. e microstructures of EMI shielding performance and the mechanical properties of the composites are examined.

Convolutional Neural Network Model.
e convolutional neural network model is designed to predict the rigidity, toughness, and strength of two material fibers, one rigid and brittle and the other soft and ductile [25]. e CNN technique model frequently needs important data to provide the exact prediction of fibers. During the training process, it examines new data that is invisible to the model following the formation of the CNN model. e CNN model is based on the data of an FEA (Finite Element Analysis). e FEA is performed to estimate the interest assets. e replication of the finite elements is motionless when the von Mises (ε vs ) deformation at the crack point reaches the corresponding fracture deformation. en, the von Mises strain is explained as Young's modulus value for stiff fiber is V � 1 GPa, and the value of soft fiber material is V � 0.1 GPa. e Poisson relationship for both flexible and rigid fiber materials is u � 1/3. e break deformation of brittle and rigid fiber materials is 10%, although the break deformation of soft and ductile fiber materials is expected to be 100%. Consider three active qualities of fibers: modulus, strength, and toughness. In FEA, four-node elements were used when the individual section has one point of four quadratures and the individual part has one degree of the two freedoms [26]. e cracked fibers have two dissimilar sizes assumed as 8 × 8 and 16 × 16 size systems as examples for the input.
Finite Element Analysis (FEA) is conducted to create data critical to this research. Consider both grid sizes 8 × 8 and 16 × 16 as illustrated in Figure 6. e FE approach is used to study a large amount of alternative complex   Advances in Materials Science and Engineering arrangements corresponding to the two grid sizes, and the resultant characteristics within each structure are recorded. e microstructure, containing information on volume fractions and phase spatial distributions, is called configuration. Some preprocessing processes are required after obtaining data from the FEA to make the information for feeding into the CNN model [27]. A matrix is a representation of data in its most basic form from the acquired FEA. e training instances are represented in the rows of that matrix, while the binary distribution of the materials is represented in the columns as 0 for stiffness and 1 for softness. Every learning algorithm has a labeled vector that contains the three material properties: strength, toughness, and strength. Each function is translated into an image by specifying the spatial model for every component to be supplied into the CNN model, whose convolutions could be conducted to develop the CNN model.
e key strategy of a CNN is that the input data are or may be viewed as a picture. As a result, the set of variables is reduced greatly, resulting in speedier processing. In Figure 7, the convolutional layer, pooling layers, fully connected layers, and beginning purposes are the layers that make up a typical CNN. A filter is connected to the input image by an activation function in the convolutional layer, which maintains the interconnections between pixel values. A pooling layer combines the input layer as maxpooling that yields the best accuracy division from every session. Layered and simulation functions in fully linked DNN systems are comparable to those in simple DNN structures [28]. e next layer, such as stall layers, which are the regulation strategies to minimize model overflow, may also be added to CNN. Two CNN architectures were developed in each dimension grid with six fused layers, one exclusion layer, and one fully connected layer in the respective architecture. Each composite layer consists of a 2D convolution layer, a 2D batch stabilization layer, and a corrected linear part activation (ReLU) function. e convolutional layers kernel scopes are varied between the two architectures.

Evaluation of CNN Model.
e mean square error (MSE) cost purpose is used during the formation phase to reduce the residual error among the output data model and the ground-truth data. e average absolute percentage error (MAPE) is used to calculate the model precision. Equations (9) and (10) are used to calculate the mean square and the mean absolute percent error to measure the difference between the actual and projected values.
MAPE � 1 a a n�1 k n − k n k n , (10) where kn is the prediction found based on the model and k n is the real value.

Genetic Algorithm for
Optimization. e various bioinspired operators such as crossover, mutation, and selection are frequently used in GA optimizers to obtain the optimal Hard Fiber Soft Fiber e GA optimizer selects random individuals from the last generation as parents generating offspring for subsequent generations through each optimization process [29]. e roulette wheel mechanism is utilized as the selection operator, with greater fitness values yielding a larger selection probability. e frequency of the optimal solution in an optimization system is called fitness.
A genetic algorithm is a type of computer-based research and optimization algorithm that depends on natural genetics and natural selection mechanics. Other conventional optimization methods are less likely to provide optimal global coverage than the GA. e reason for this is that GA is based on probabilistic criteria and research based on a population of the points in inherited space. Traditional optimization strategies are usually based on deterministic research which uses the idea of the gradient to localize local optima. Each genetic operation cycle is referred to as a developing process. e fitness of every member of the population was assessed in each generation and subsequent generations were formed using the chromosomes of the current population. is entails the mixing and recombination of genes from both parents to produce offspring in the following generation. Reproduction, mutation, and crossover are all parts of the evolution process. e optimization problem as is follows: a i are decoded with this linear mapping rule: where a i � p 1 , p 2 ; (U) � minimum limitation; (V) � maximum limitation; ui � binary bit count; and t i � equivalent of decimal values. A chromosome relates to some microstructures that store the material around it to form a specific microstructure element. A uniform distribution is used to choose the first generation at random. e suitability of a microstructure is determined independently of the break of the generational chromosomes. e technique for using the genetic algorithm is shown in Figure 8.

Results and Discussion
e efficiency of the linear model is associated with the CNN model. To develop a linear model quantitatively based on the composite properties, the equation in the linear model is given as follows: where M is the property of the material and C is the material distribution of the compounds in vector form. X and Y are determined as weights, and X and Y weights are regulated by the training process to suit every physical property. Since composite gratings are of 8 × 8 and 16 × 16 sizes, they standardize for every model for all 3 properties of fiber material. e determination of the resulting coefficient (D 2 ) for the adjusted line is 0.916. en, the mean absolute error ratio represents the limitations and generalization issues by the linear model; then, the modulus exceeds the mean absolute error percentage (25%), and the toughness and strength are as high as 200% and 40%. As 16 × 16 grid size, the outcomes of the linear model are determination coefficient (D 2 ) of 0.928; then, the values of MAPE are maximum with 21%, 32%, and 127% for modulus, strength, and toughness. High error standards are generated from the linear model to explain the CNN for the prediction problem. Figure 9 and 10 show that the ranked weights are determined by the dark color near to the tip crack, the light color is determined by the straight one at the crack tip as positive of soft fiber materials, and then colorless materials are determined by the negative as stiff fiber materials.
ese composite mechanical properties have high accuracy. e CNN is used in signal processing and recognition of images and is also utilized to extract geographies from datasets. Before the inference process, CNN models must be developed. e training process is a problem of optimization; then, the MSE is minimalized by the optimal collection of the weight of the model CNN. To begin the training process, it required many examples of dataset; then, the input and output data determine the approximate relationship by optimal mapping. en, the CNN model uses e CNN model developed and produced the best problematic results and predicted the three properties of fiber materials with great precision.
After the process of training, estimate the CNN model performance using the 3 parameters, the mean absolute error ratio, data points percentage, and maximum error with an error maximum of 5%. ese parameters are measured by testing the data set that is nongenerated by the CNN during the process of model creation. e performance of the CNN model is illustrated in Tables 2 and 3. Resistance, toughness, and MAPE modulus are less than 5%. A stronger limit for estimating the activity of the CNN is the extreme error. e extreme strength and error in the module are greater than 5%. But the data points' number larger than 5% error is meager, 8 × 8 grid has 1.7%, and 16 × 16 has 0.082%.

Shielding Performance.
e EMI samples results A2, A3, A2H11, and A3H12 in a range of 10 GHz are compared and examined as given in Figure 13. e EMI shielding was intensely improved after the A1 sheets are presented because of the good shielding property. e EMI samples A2, A3, A2H11, and A3H12 are 1.0-4.8 dB, 1.4-6.3 dB, 30.7-46.8 dB, and 28.5-53.5 dB as in 8 Advances in Materials Science and Engineering Figure 13(a). Comparing the control samples, EMI shielding improves the samples A2H11, and A3H12 are important. e samples EMI shielding performance A2H11 and A3H12 are on the needed values of EMI shielding in applications. Figure 13 Figure 13(c). e natural fiber composites have low EMI, while the composites with A1 sheets presented much more reflection.
e EMI reflections of A2H11 and A1H12 minimized the A1 sheet measured as 83.7-93.9%. During the reflection and incidence, materials absorb EMI waves; it will reduce the shielding efficiency.   e objective aggregate function (AOF), which combines the various objects into the scalar function,   depends on the simple biased objective amount that works poorly in the nonvexatious case. One solution is used for programming compromise: T m � modulus maximum modulus , where x is the objective exponent function, and it is selected to be x � 4. en E m, E s, and E t are the masses of modulus, strength, and toughness. When optimizing two properties, the weight matching to the properties and then optimized for the assigned value is 0.5. en the third property of the attributed weight value is 0.0. For three properties, the value of each of the three weightings is 0.333. In addition, the various goals have various scales which can cause bias to higher values of the goals. is issue is determined by each normalizing target by its maximum value as shown in (16). Figure 15 and Table 4 show the outcome result derived from the optimizer GA when the multiobjective is measured with the checkerboard material. From the optimization of single-objective, the material is not soft, so the modulus is maximum. But when optimizing toughness or strength in addition to the module, the optimizer allows for equilibrium of the various purposes. e modulus determined from optimization of multiobjective is importantly minimized due to the soft materials, and the soft fiber material volume segment ranges from 23% to 33% depending on the materials that are optimized. en, the configuration of strength high values and the toughness of fiber materials have a good synchronization. e optimizer inclines to  bundle the flexible fiber material at the location next to the tip of the crack. erefore, less compromise is required.

Conclusion
An approach to analyze the mechanical properties of natural fiber composites for modeling and optimization methods was developed, such as RoHM, CNN, and GA methods to improve properties and usage. e techniques for application modeling in NFC allowing to identify and isolate the various parameters that have an important effect on the outcoming materials are carried out to the optimization. e convolution neural network (CNN) is established to predict the mechanical properties of NFC composites by considering the checkboard of two sizes of 8 × 8 and 16 × 16 gird. e grid model is used to determine the module, strength, and tenacity of the material. It has shown the potential for using CNN in hardware and structural study. e composites represent the shielding performance of EMI with better mechanical properties as the structure of EMI shielding materials. e development of the model CNN is combined with a GA optimizer to find the fraction of volume of the composite material configuration to enhance performance. e optimization of GA results from the mechanical property of fiber length and content. ere is better agreement between values of experimental and predicted values. CNN models take the accelerating potential of present optimization methods, and they develop the field of material and structural design.

Data Availability
e data used to support the findings of this study are included within the article.

Conflicts of Interest
e authors declare that there are no conflicts of interest regarding the publication of this article.