Experimental investigation and prediction of wear behavior of cotton fiber polyester composites

The cotton fiber reinforced polyester composites were fabricated with varying amount of graphite fillers (0, 3, 5 wt.%) with a hand lay-up technique. Wear tests were planned by using a response surface (Box Behnken method) design of experiments and conducted on a pin-on-disc machine (POD) test setup. The effect of the weight percentage of graphite content on the dry sliding wear behavior of cotton fiber polyester composite (CFPC) was examined by considering the effect of operating parameters like load, speed, and sliding distance. The wear test results showed the inclusion of 5 wt.% of graphite as fillers in CFPC increase wear resistance compared to 3 wt.% of graphite fillers. The graphite fillers were recommended for CFPC to increase the wear resistance of the material. A scanning electron microscope (SEM) was used to study the wear mechanism. To predict the wear behavior of the composite material, comparisons were made between the general regression technique and an artificial neural network (ANN). The conformation test results revealed the predicted wear with the ANN was acceptable when compared with the actual experimental results and the regression mathematical models.


Introduction
Within the last few decades natural fiber reinforced polymer matrix composites (NFRPCs) were in boom due to their low cost, low weight, easy availability, and biodegradability. The natural fibers were in demand in automobile and structural sectors [1]. In the present work focus was placed on the cotton fiber due to its high strength, durability, biodegradability, and ease of blending with other fibers and resin. Failure due to wear was more common in automobile and failure of small parts leaded to the shut down in the industry. In this present investigation emphasis was placed on the wear behavior of the cotton fiber reinforced polyester composite (CFPC). Many researchers have worked on different types of NFRPCs and have analyzed the effect of operating parameters (like load, speed, sliding distance, and temperature) and material parameters (like fiber length, fiber volume fraction, fiber orientation, and fiber treatment) on the wear behavior of NFRPCs. The work done of a few researchers is shown in the Table 1. Table 1 reveals that most of the researchers have analyzed the wear rate of NFRPCs by varying different material parameters and different operating parameters. Very few have added fillers with the natural fiber and see the effect of fillers on the wear behavior of materials. Few studies show that use of proper wt.% of fillers with synthetic fiber helps to increase the wear resistance of the materials [16,17].
In the present investigation graphite fillers are used with the CFPC to analyze the effect of different wt.% of fillers on the wear behavior of CFPC.
The response behavior of material is implicit in the experiments. Usually, the behavior of materials is modeled analytically using mathematical expressions. However, it may not always be possible to have a simple expression. To create a complicated expression is difficult. The artificial feedforward neural network may be extremely helpful in terms of Refs. [18,19]:  The ability to implicitly detect complex nonlinear relationships between dependent and independent variables.
 The ability to detect all possible interactions between variables.
 The neural network is effective in terms of predicting the behavior of a new material before the material is produced. This may reduce the experiment cost and time. An exhaustive literature review has been completed on the artificial neural network (ANN) to predict mechanical and tribological behavior of fiber reinforced polymer composites. Table 2 provides the summary of various studies for predicting the mechanical and tribological behavior of fiber reinforced polymer (FRP) composites with the ANN. In the present investigation

Specimen preparation
Single wound 7 count cotton yarn procured from PBM Polytex Limited, Petlad, Gujarat was used as reinforcement. Unsaturated polyester resin, accelerator cobalt naphthenate and hardener were supplied by S K Enterprise, Surat, India. Matrix was prepared with a resin to hardener ratio 10:1. Graphite particles obtained from Heny Chemicals, Vadodara, India was used as filler with the average particle size 11.91 μm.
Hand lay-up technique was used to prepare the composite plates of size 300 mm × 300 mm × 10 mm. The composition of fabricated composites with hand lay-up technique was listed in Table 3.
To determine graphite particle size distribution in the composites, the tensile test was carried out on the composites at Advance Metallurgical Services (AMS) Laboratory, Vadodara, Gujarat. The tensile test was performed according to the ASTM D 3039-M14 standard. A 10 kN load cell was used and three tests were repeated for each set of specimen and the average values were calculated and plotted in Fig. 1.
It was clear from the results that the addition of 3 wt.% graphite reduced the tensile strength of the cotton fiber polyester composites. The strength of the composite was even more deteriorated by increased graphite wt.%. This result was in agreement with Shalwan and Yousif [29] who concluded from their study that the addition of the graphite was highly recommended for the natural fiber composites which could enhance the wear characteristics of the polymer composites. However, the high content of the graphite deteriorated the mechanical properties. The micrograph    Fig. 2 good bonding between the resin and the fiber could be seen. This adhesion might be responsible for higher tensile strength. Figure 3 revealed that by adding 3% graphite in the CFPC, the first failure process was initiated from the matrix crack and then it was followed by the fiber failure in the direction of loading. Fiber pull out and fiber failure revealed from Figs. 4(b) and 4(a) for 5GCFPC that might be responsible for the low tensile strength of the material.

Test setup, test conditions, test parameters and design of experiments
To perform the wear test on the CFPCs, a pin-on-disc (POD) test setup (supplied by DUCOM, Bangalore) was used at CHARUSAT, Changa, Gujarat and it was shown in Fig. 5. The specimen was kept stationary against the disk and the counterface rotated. The load was functional through the lever mechanism.
The detailed experimental conditions were listed in Table 4. The specimen surfaces were prepared by rubbing them on different grade emery paper followed by acetone cleaning. The average surface roughness for the specimens before and after the test was measured with a Taylor Hobson Roughness tester and listed in Table 5.
Response surface methodology (Box Behnken method) was used to reduce the number of experiments in an organized way.

Wear data from pin-on-disc machine
The operating parameters were set on the POD machine and experiments were performed. The wear response of composite materials was plotted from the POD machine and listed in Table 6.

Mean effects of operating parameters on response
The results of the mean wear for CFPC and graphite filled composites for different operating parameters were shown in Fig. 6. To analyze the wear mechanism for the composite materials SEM analysis was performed on the wear surfaces. Figure 7(a) showed the SEM image for CFPC. It was clearly visible that good bonding between the fibers and matrix occurred. The fibers were not debonded normally from the matrix due to proper bonding. They were worn out during the wear process. Some fibers were peeled off during the wear process. Figure 7(b) revealed that 3GCFPC had undergone severe damage under the dry sliding conditions. Large and small wear debris was left over from the dry sliding condition on the worn surface of the specimen. The presence of this debris was mainly responsible for the large scale of wear. Matrix cracking was also found on the worn surface. These mechanisms were responsible for the high wear rate for 3GCFPC. Large scale disconnection of the matrix material and fibers produced several deep grooves on the specimen surface. Figure 7(c) indicated the formation of a friction film between two contact surfaces due to this less matrix cracks found and the tribological properties of a material improved in terms of less wear.

Regression mathematical models for wear
An ANOVA statistical tool was used with 95% confidence level with Minitab 17 software to quantify the influence of process parameters. Response surface methodology was used to create a mathematical model for wear of CFPCs. The second order general regression mathematical equations representing the relation between the wear and the input parameters (like load, speed, and sliding distance) for different materials were obtained and listed below. where L= load, S = Speed, and SD = Sliding distance. The coefficient of determination (R 2 ) obtained for different models was listed in Table 7. It represented the ratio of variability explained by the model to the total variability in the actual data. Larger values of adjusted R 2 suggested models of greater predictive ability. Table 7 showed that all models satisfy this condition. All the response predicted R-sq values were in agreement with the adjusted R-sq values. This indicated the capability of the model was used effectively.

Configuration of artificial neural network (ANN)
The process of creating an artificial neural network (ANN) for the present research work was summarized in the following steps: 1. Collect the data and prepare the database: With the help of POD machine wear for the three different materials (CFPC, 3GCFPC, and 5GCFPC) was obtained. Box Behnken gave 15 design of experiments for the each material. Total 45 different wear values with one replica that is total 90 wear data were obtained for the three different materials.
2. Train the network: In this step network architecture, training functions and training algorithms for the network were required to select. The ANN model developed in the present research work was created using a MATLAB R13 software package. This package allowed the modification of the network architecture, such as generalized regression, Hopfield and feedforward backpropogation. Also the design parameters like number of neurons in each layer, number of hidden layers, neurons in the hidden layer, learning rate and momentum could be modified. Although this package offered a mixture of possible modification to the network design, not every modification was investigated. The work mainly focused on developing an ANN model for the wear instead of addressing an optimum network design. In the present work 72 data points (80%) were used for training/calibration and the remaining 18 data points (20%) were used for testing. The calibration/training data set and testing data set were selected randomly from the entire population. The schematic configuration of ANN was shown in the Fig. 8. Table 8 showed the design parameters used to train the network. The network was trained with a      where M i is measured wear and P i is predicted wear.