Investigation of Tensile Behavior of Carbon Nanotube/Coir Fiber/Fly Ash Reinforced Epoxy Polymer Matrix Composite

ABSTRACT Fiber-reinforced composite materials are lightweight and can withstand heavy loading conditions. Reinforcement augments the strength of the composite material, which is assessed by its elastic modulus. An attempt is made to reinforce epoxy with Coir Fiber, Carbon Nanotube (CNT) and Fly-ash. Central Composite Design (CCD), a Response Surface Methodology (RSM) tool, which is a Design of Experiment (DOE) technique, is used to fabricate the experimental samples to study their tensile behavior. Analysis of variance (ANOVA) is employed to investigate the effect of reinforcement percentage of CNT, coir-fiber, and fly-ash on tensile behavior of composite. The ANOVA results follow the trend of the experimental values with a deviation of less than 10% in yield strength, tensile strength, and Young’s modulus. Two models (Artificial Neural Network and Multiple Linear Regression Model) originated resting on the regression equation to speculate the elastic modulus for various reinforcement parameters using the experimental data. The main objective is to optimize reinforcement parameters using both the models having a maximum elastic modulus of 2.602 GPa and 2.682 GPa, respectively, which is achieved by the teaching learning-based optimization technique. Furthermore, confirmation experiments validate the optimization process with an error of less than 4%.


Introduction
The excellent performance of fiber-reinforced polymer composites, in terms of chemical inertness and high specific strength, caters to a wide range of industrial applications. Adding to the strength However, in the optimization of the microwave-assisted fluidized bed drying process done by Srinivas et al. (2020), the higher regression coefficient and lower SSE values of MLR have better predictions than ANN.
TLBO (teaching-learning-based optimization algorithm) is a subset of Artificial Neural Network (ANN) that has been widely tried to optimize the experimental parameters for the intended outcome. In the work of Abhishek et al. (2017), authors used TLBO for parametric evaluation/optimization of machining of carbon fiber reinforced polymer CFRP composites. Gao et al. (2021) also observed the effectiveness of TLBO algorithm in optimizing the input parameters on sound absorption characteristics of hybrid polymer composites. Ornaghi, Neves, and Monticeli (2021) studied the kinematic parameters on heat treatment of lignocellulosic fiber using the Artificial Neural Network (ANN) curve fitting approach, following to that Surface Response Methodology (SRM) was implemented to find their heating characteristics. Monticeli, Neves, and Júnior (2021) used ANN to predict the thermal degradation of vegetable fibers by training the network with experimental values using 12 hidden layers. After training, the predicted outcomes were used as the reference to analyze the intended quantity of interest.
The present work is the combination of experimental and statistical analyses of epoxy composite with coir, CNT and fly ash as reinforcements aiming for higher tensile properties. Design of Experiment (DOE) approach i.e. Response Surface Methodology (RSM) is utilized for the specimen preparation and analysis of variance (ANOVA) is employed to investigate the impact of reinforcement percentage on tensile properties. In addition, ANN algorithm-based MLR and TLBO Model are also employed to analyze the reformed unification of coir, CNT, and fly ash. The combination of triple reinforcements Coir, CNT and fly ash in epoxy composite and parameters optimization opens a new avenue in the design of composite structures that is not tried much.

Materials and methods
The use of eco-friendly materials used in this work, to prepare the composites, is motivated by varied literature and environmental sustainability. The epoxy resin [LY556]/hardener [HY951] (purchased from M/s. Herenba Instruments and Engineers, Chennai, India) is used as a matrix and the coir, fly ash (purchased from M/s. Go Green Products, Chennai, India) and Carbon Nano-Tubes (CNT) (purchased from M/s. Nanoshel, Punjab, India) are used as reinforcements to prepare the composites. The dimensions of CNT used for the present investigation are OD: 10-20 nm, Length: 3-8 µm, 90% purity. A Design of Experiments (DOE) is used to evaluate the parameters, which are grouped for distinct combinations using Central Composite Design (CCD) featuring Response Surface Methodology (RSM). The fillers, namely, coir fiber, CNT, and fly ash are the input variables used in different weight percentages at five levels as shown in Table 1. Table 2 gives the actual value and coded value of input parameters and obtained a set of 20 test runs from the CCD method.
ASTM D3039/D3039-17 standard is used to prepare the tensile samples using the mold with dimensions 250 mm × 25 mm × 3 mm for testing ( Figure 1). Based on the combination of fillers as shown in Table 2, the CNT is first added with epoxy resin and stirred well to avoid agglomeration using a mechanical stirrer (Remi Mechanical Stirrer, Model -RQ − 121/D, with a Stirring shaft of 8 mm dia, 250 mm length, the impeller of 36 mm dia, speed of stirrer 200 to 2000 rpm, 10-l capacity). Then, coir is added to the mixture and finally, the fly ash is added to prepare the epoxy composite mixture. The test samples are subjected to tensile tests (INSTRON 8801,Norwood,MA,USA) at the strain rate of 1 mm/minute till failure to extract tensile properties as given in Table 3. The proposed ANN model is generated using "nntool," a neural network toolbox available in MATLAB 2017 version. The following steps are followed to generate an ANN model for the specific response.
Step 1. A new Feed-forward backpropagation network is created using the "newff" function.
Step 2. Then, initialization of network parameters is done using the "init" function to initiate the weights & bias values and parameter values of the network.
Step 3. Training of network is carried out using "train" function and based on the training function parameter, this returns a new network along with output and error.
Step 4. The "sim" function is used to simulate the neural network to get return outputs, errors, and network performance.
Step 5. The stopping criteria are to be reached after repeating steps 3 and 4 till an acceptable performance of the network is reached.   Triplicate samples are prepared for each combination, tested for their tensile characteristics and the results are averaged for the entire analysis. The selection of levels for the input parameters is based on the wt % of fillers added to the epoxy resin. The maximum weight percentage of filler addition in the epoxy system is found to be 7% and beyond which leads to agglomeration and fails at lower load. Hence, the maximum overall percentage of filler addition in the present work is limited to 5 wt %.

Results and discussions
The experimental results and the corresponding statistical analysis of the tensile behaviors of composite laminate are presented in the following sections.

Experimental and statistical analyses of tensile behavior
Tensile properties of the test samples are recorded for a different combination of input parameters shown in Table 3. From Table 3, it is observed that sample number 11 exhibits the maximum yield strength of 14.926 MPa with 1% CNT than other combinations. The mechanical strength of epoxy increases with an increase in CNT (Mohan and Rajmohan 2017). The samples, reinforced with coir 1%, CNT 0.5%, and fly ash 1%, exhibit the maximum ultimate tensile strength of 29.574 MPa (Ashok et al. 2021). Also, samples with 1.5% coir, fly ash, and 0.25% of CNT has given a maximum Young's modulus of 2.595 GPa. The samples, reinforced with CNT alone, exhibit good bonding strength with epoxy. However, samples reinforced with coir, CNT, and fly ash show better mechanical properties than only with CNT reinforced epoxy samples. The test results reveal that the combined influence of Coir, CNT and fly ash is appreciable at a specific wt % exhibiting their threshold limits of individuals in a composite (Rajkumar et al. 2021). Consequent to the experimental outcome, the ANOVA approach is implemented to investigate the level of impact on the wt. % of coir, CNT and fly ash on the aforementioned composite materials' characteristics (Gopalan and Pragasam 2018).
Regression Equation (1)- (3) are obtained from the ANOVA analysis for the yield strength, ultimate tensile strength and Young's modulus of the test samples, respectively. In the derived regression models, A, B and C refer to the wt. % of coir, wt. % of CNT and wt. % of fly ash respectively. It is evident from Equation 1 that wt. % of coir fiber has a negligible impact on yield stress after 1%. On the contrary, wt. % of CNT has a negative influence and it decreases initially. However, the maximum yield stress is obtained at 1%. Resting on the coefficient of parameter C i.e. fly ash in Equation (1) , it has the least influence of all. Table 4 compares the experimental yield strength values with regression equations and the deviation between them is less than 10%, excluding 5 samples.
The underlying limitation of the regression model in arriving at optimum wt% combination for maximum tensile parameters drives the authors to implement ANN and MLR models. The objective of the present work is to investigate the effect of filler addition on the mechanical properties of composite materials aiming at property enhancement. It is observed from the regression analysis that, the combined presence of coir and fly ash in the composite is not appreciable in attaining yield stress and ultimate tensile stress, whereas with the addition of 1 wt% of CNT along with coir and fly ash, the composite attains higher yield and ultimate stresses. However, the contribution of CNT, with coir and fly ash in obtaining Young's modulus seems to be invariant exhibiting their amalgamated behavior within the elastic limit. This exhibits the limitations of regression analysis. The listed tensile parametersin Table 4  Predictive Models namely Artificial Neural Network (ANN) and Multi Linear Regression Model (MLR) are utilized to discover the desirable combination of the input parameters (Lal et al. 2021;Richa et al. 2021). Furthermore, the Teaching Learning-Based Optimization (TLBO) algorithm is utilized to get the input parameter combination for maximum Young's modulus value and then compared with experimental results.

Response prediction by ANN Model
Feed Forward Back Propagation has established itself as the universal tool for the ANN Model to predict the response values (Mulenga, Ude, and Vivekanandhan 2021). Table 5 represents all the input parameters' combinations (CNT, Coir Fiber and Fly Ash) and their corresponding measured response values (Young's Modulus). The above-mentioned data is utilized to originate an ANN Model where the MATLAB R2017 software is used to execute, investigate, and legitimize the data. The ANN Model supervises the interaction between the input parameters and the response values. Several features like learning, training and performance, number of hidden layers and neurons, each hidden layer's transfer function and initial weights are the parameters that determine the potential of the ANN model. Figure 2 portrays the blueprint of the network architecture developed in this work along with various functions and training parameter values utilized in the present work. Figure 3 unveil the training performance and regression analysis respectively. The filler wt parameters used in the optimization model are "0" as the Lower limit and 0.364 for Coir and flyash and 0.182 for CNT. The accuracy of the ANN model is 97.8% and the same for the MLR model is 95.3%.

Execution of TLBO algorithm
TLBO algorithm is based on the trainee and trainer model using a combination of parameters namely, wt. % of coir fiber, CNT and fly ash affects the composite response and is considered as student in this work. The gradual course of action to conduct the TLBO algorithm is given below.

Procedure 01: TLBO-ANN strategy
Part 1: Initialization: An arbitrary array of input parameters (v ij ) is originated within their limits, utilizing the function "rand (ns,nv)," where, "ns" and "nv" indicates the number of students and input parameters, respectively.
Part 2: Development of ANN model (net1): The function of the ANN Model is a feed-forward back-propagation evolved utilizing the computed Young's modulus values as well as other input parameters. MATLAB functions, "train" and "newff," are utilized to carry out the task.
Part 3: Calculation of Young's modulus (ts i ): All array of parameters (v ij ) originated in Part 1 are utilized in Equation (4) to evaluate the Young's modulus by applying the ANN model.
Part 4: Ideal Teacher (bv) and Mean Value (mv) Calculation of the intended parameter.
Part 5: Teacher's Phase: In this phase, new parameter value (nv ij ) is computed utilizing Equation (5) which is verified for their limits by Equation (6). The Young's modulus (nts i ) values are evaluated by Equation (7) and it is compared with ts i . Furthermore, the parameters corresponding to the highest Young's modulus are the refined parameter after the teaching phase (tv ij ) and its Young's modulus is served as the improved performance (tts i ) of the student after the teaching phase. Here, Part 6: Learner Phase: Interchanging of skills of the arbitrarily selected students enhances the performance. The selection criteria rest on the fact that the selected students say a and b, must not have the same potential (tts a ≠tts b ). Equations (8) and (9) are utilized to compute the new values (nv ij ).
if tts a <tts b if tts a >tts b As earlier the new values are verified for their limits by Equation (6) and compared with tts i . Similar to the teaching phase, the parameters offering the highest Young's modulus and its magnitude can be served as the refined parameters after the learner's phase (lv ij ) and improved performance (lts i ) respectively. Part 7: Replacement The total values of v ij and ts i are interchanged with lv ij and lts i respectively in this part. Part 8: Stopping Criterion Iteration stops automatically on crossing the target. If not, it experiences the repetition of all the parts from 4 to 7.  Table 7.
Part 3: Evaluation: The Young's modulus values, for all the students, are evaluated utilizing the MATLAB function "x2fx" and the same strategy is applied to acquire the optimal Young's modulus.
The TLBO Algorithm is implemented with the goal to acquire the highest Young's modulus value utilizing MLR and ANN models. Table 6, gives the ANN and MLR-optimized data sets obtained through the TLBO strategy. It is observed that the predicted wt of fillers coir, CNT and fly ash in the ANN for optimized young's modulus value of 2.6015 GPa are in the range of 0.2188-0.2744 gm, 0.0215-0.0397 gm and 0.3521-0.3603 gm respectively. Whereas in the MLR Model, 0.3556 gm of Coir, 0.0021 gm of CNT and 0.3168 gm of fly ash are found to offer a young's modulus value 2.6822 GPa. The predictions made by ANN and MLR are upper bound than CCD regression analysis (2.326 GPa) which signifies their superiority in framing the prediction model.
The Young's modulus obtained from the experiment and ANN, MLR prediction is given in Figure 4. As per Aziz, it is cogent that the ANN model seems more promising in predicting data when compared to the MLR model from the actual data attributed to the higher intended Youngs Modulus with lower values of errors. The ideal composite material parameters having the highest Young's modulus value obtained from MLR-TLBO and ANN-TLBO algorithms are given in Table 7.
In the process of optimization of Young's modulus, the ANN -TLBO technique identifies the changes in wt% of Coir, CNT and Fly ash as follows, 1.5 to 1.2%, 0.25 to 0.118% and 1.5 to 1.98% respectively. This shows a drop only in Coir and CNT and a rise in fly ash when compared with the wt % given by CCD regression (1.5%, 0.25% and 1.5%). Whereas in MLR-TLBO process, the wt% of the Coir and Fly ash is marginally the same around 1.95% and with a least wt% of (0.012%) CNT. The findings of the ANN -MLR Model with the least CNT wt below 0.25 appreciate the economic aspect and minimization of CNT in the composite making to offer higher mechanical properties. Additionally, Figure 5 exhibits the actual and the predicted values of Young's Modulus using the ANN -TLBO process with R = 0.9217. It is evident from Figure 4 that the predicted values are resting on the ANN model and are congruent to the actual values as predicted by Lal et al. (2021). Compared with MLR, the ANN shall be recommended and can be utilized to predict the tensile properties of epoxy composite with reinforcement filler as Coir, CNT, and Fly ash.
To validate the above-mentioned response (maximum Young's Modulus), experimental tests are conducted for the optimized wt % of fillers as obtained in the ANN (Coir −0.22 gm, CNT −0.02 gm and Fly ash − 0.36 gm) and MLR (Coir −0.35 gm, CNT −0.02 gm and Fly ash − 0.36 gm). It is evident that the experimental results (Table 6) are in accordance with the MLR -TLBO and ANN -TLBO algorithms with a deviation of less than 5%.

Conclusions
The exploration of reinforcing coir, CNT, and fly ash in epoxy polymer matrix composite is carried out in the present work. The regression equation is scrutinized to study the impact of wt. % of the reinforcements on their tensile behaviors. It is evident that Young's modulus greatly influences the strength under the elastic region. Therefore, it is intense to investigate the wt. % of parameters to maximize Young's modulus.
To achieve the aforementioned target, ANN and MLR models are employed on the experimental data and assigning response values as Young's modulus. Furthermore, the outcome of the two models is conveyed through the TLBO algorithm to get optimized parameters. ANN model makes it possible to secure the highest Young's modulus value of 2.6015 GPa with the least error. Additionally, to achieve the above-stated target, the coir fiber content is slashed from 0.273 to 0.2188 g (1.5% to 1.2%); CNT content from 0.0455 to 0.0215 g (0.25% to 0.118%) and the fly ash content is increased from 0.273 to 0.3603 g (1.5% to 1.98%) when compared to the experimental data. Likewise, utilizing MLR model, the maximum Young's modulus value is found to be 2.6822 GPa with 0.3556 g (1.95%) of Coir, 0.0021 g (0.012%) of CNT and 0.3618 g (1.99%) of Fly ash. This work can further be extended to enhance filler interaction by incorporating filler sizing and its response to dynamic loading rather than static.

Highlights
• Triple particle (coir, CNT, and fly ash) reinforced epoxy composites were examined for optimum wt% to maximize tensile properties. • ANN and MLR models are employed with TLBO for the wt% optimization of particles. • ANN model is efficient to optimize the wt% of particles to attain the highest young's modulus value with reduced wt% of coir, CNT and increased wt% of fly ash. • The increase in wt% of fly ash in proportion with Coir and CNT by ANN has devised an effective utilization of power plant waste for structural application.