Experimental, modeling and optimization study on the mechanical properties of epoxy/high-impact polystyrene/multi-walled carbon nanotube ternary nanocomposite using artificial neural network and genetic algorithm

Abdolhosein Fereidoon; Amin Hamed Mashhadzadeh; Yasser Rostamiyan

doi:10.1515/secm-2012-0178

Article Open Access

Experimental, modeling and optimization study on the mechanical properties of epoxy/high-impact polystyrene/multi-walled carbon nanotube ternary nanocomposite using artificial neural network and genetic algorithm

Abdolhosein Fereidoon , Amin Hamed Mashhadzadeh and Yasser Rostamiyan

Published/Copyright: March 15, 2013

Published by

Become an author with De Gruyter Brill

Submit Manuscript Author Information

From the journal Science and Engineering of Composite Materials Volume 20 Issue 3

Abstract

In spite of Epoxy resin’s good tensile strength, are brittle in nature and have poor resistance at the front of crack propagation. In enhancing simultaneously the mechanical strength and fracture toughness of epoxy-based nanocomposites, high-impact polystyrene (HIPS) as thermoplastic phase and multi-walled carbon nanotubes (MWCNT) as nanofiller phases are used incorporately to obtain ternary epoxy-based nanocomposites. Tensile, flexural, compression and impact are the four different mechanical properties. Artificial neural network was used to present models for predicting the mechanical behavior of epoxy/HIPS/MWCNT nanocomposites. Also, this model used as a fitness function of genetic algorithm as a powerful optimization method to find the optimum value of the above-mentioned mechanical properties. The effective parameters investigated were HIPS, MWCNT and hardener. From the result, it was found that the combination of HIPS and MWCNT nanofillers significantly increases tensile, compression and impact strength of neat resin by up to 52%, 43% and 334%, respectively, but flexural strength did not change positively. Also, elongation at break for tensile, flexural and compression rose to 223%, 36% and 26% of neat epoxy, respectively. The morphology of fracture surface was studied by scanning electron microscopy.

Keywords: epoxy; high-impact polystyrene; mechanical properties; multi-walled carbon nanotubes; toughness

1 Introduction

Epoxy-based composite materials have a low-weight, high-adherence and good chemical resistance; this is a good reason to increase using these types of materials as a structural ingredient in the aerospace and automobile industry [1–3]. But comparatively speaking epoxy has weak mechanical properties. This reason prevents the use of epoxy in components that need high mechanical strength and stability. Attempts at reinforcing this material have generally focused on the use of various types of fillers [4]. Carbon nanotubes (CNT) have the potential to improve the mechanical, electrical and thermal properties of epoxy as a filler [5–11]. For example, addition of 0.2–10 wt% CNT into epoxy improves modules [12] and mechanical strength [13]. Montazeri et al. [14] showed that tensile strength and Young’s modulus improved by the addition of 0.5 wt% multi-walled carbon nanotubes. Allaoui et al. [15] found that the addition of 4 wt% MWCNT increases Young’s modulus and strength. Gojny et al. [16] observed that double-walled carbon nanotubes (DWCNT) increase tensile strength and fracture toughness. Gojny et al. [17] studied the effect of CNT on mechanical properties in order to enhance fracture toughness. Zhou et al. [18] reported that the addition of 3 wt% CNT fibers increases by 19.4% Young’s modulus. Also, other researchers showed that the use of CNTs as a reinforcement of epoxy increases by 15–36% compressive strength [19–21] and flexural strength [18, 20, 22–25].

Because of its tight three-dimensional molecular network structure, epoxy has a weak strength in front of crack growth and has a brittle fracture behavior. This reason has limited its use in loading conditions. In the last decade, many works have been done to increase toughness [26]. For many years, incorporation of micro-phase dispersed rubber [27] or different types of thermoplastic polymers [28] is one of the well-known ways to increase the impact strength of epoxy thermoset polymers. Sultan and McGarry [29] used rubber in epoxy matrix for improved toughness. They used carboxyl terminated acrylonitrile (CTBN) rubber to toughen diglycidyl ether of bisphenol A (DGEBA) epoxy resin and observed that fracture energy value is increased. Pearson and Yee [30] showed that the use of rubber particle in DGEBA epoxy resin as a matrix increases fracture toughness by a particle-bridging/crack-deflection mechanism. Frounchi et al. [31] added a rubber particle in DGEBA resin as a matrix and a 40% increase in impact resistance was observed. Chen and Jan [32] increased fracture toughness by up to 171% by adding CTBN as rubber particle in epoxy. Also, other researchers showed that using thermoplastic particle as reinforcement in epoxy resin improves the mechanical properties [33, 34]. Between using rubber and thermoplastic phase in order to enhance impact strength, because of minor dramatic effects on other desired mechanical properties, toughening with engineered thermoplastics such as polysulfone [35], polyether amide [36], a crylonitrile butadiene styrene [37] and polyethersulfone [38] is more usable. In general cases, it should be noted that, although the presence of rubber or thermoplastic phase in epoxy resin can usually play a great role in fracture toughness improvement, it can also enhance viscosity and dramatically decrease toughness strength, modulus and stiffness [39, 40].

To increase the benefits of adding fillers to matrices, researchers have studied the probability of a synergistic effect in composites with two or more kinds of fillers. The definition of synergy is achieved to improve the combination which cannot be achieved by each phase working alone. Fu et al. [41] showed the effects of synergy on toughness. Maazouz et al. [42] observed the effects of synergy on fracture toughness of hybrid micro-particulate composites. Kinloch et al. [43] added nano-silica and micro-rubber particle to epoxy resin. They showed a significant increase in toughness. Their results indicated that the toughness increased slightly when nanosilica particles were added to epoxy. Geisler and Kelley [44] used a combination of alumina (AL₂O₃) and rubber particle in the epoxy matrix. This composition had fracture toughness values 25% higher than those of epoxy systems having only alumina (AL₂O₃) or rubber particle. They also showed toughness improvement without significantly diminishing the other properties such as elastic modulus and glass transition temperature. Mirmohseni and Zavareh [37, 45] filled epoxy resin with nano-clay and thermoplastic particle and a synergistic effect was observed on tensile and impact strength. In the preparation of hybrid nanocomposite samples, various factors may be affected significantly and control of these parameters including optimizing them using artificial intelligence algorithm seems to be necessary.

With the development of artificial intelligence techniques, many researchers have used these methods in engineering fields. Neural network and genetic algorithm are two types of intelligent algorithms for modeling and optimization. Artificial neural network (ANN) is a non-linear algorithm which is used for modeling process parameters [46, 47]. This algorithm can learn and predict experiential result with a satisfying weight and bias in each neuron. When no analytical relationships exist between process parameters, this algorithm could well be used for modeling of process parameters. Masri et al. [48] showed that neural network is a powerful method for the identification of systems typically encountered in structural dynamic fields. Mirmohseni et al. [45] used a neural network to model the behaviors of concrete in the state of plane stress under monotonic biaxial loading. Rhim and Lee [49] used a neural network to identify the characteristics of damage in composite structures. But this algorithm cannot find the optimum conditions and might be hampered in the local minimum. Genetic algorithm (GA) solves this shortcoming. In the computer science field of artificial intelligence, GA is a search heuristic that mimics the process of natural evolution. GA is a powerful method for solving complex problems. Many researchers have used this method to optimize procedures [50–52]. This optimization method use genes and chromosomes as a code design variable. This algorithm uses a population of designs instead of a single design. A new generation of designs has evolved from previous generations by applying reproduction, crossover and mutation operations to the genetic strings. Moreover, an individual design that has high fitness (in this work measured by an ANN model) also has a high probability of producing a new generation. Mousavi et al. [53] used an ANN and GA to obtain a relatively high flow stress in compression tests for 304 stainless steel.

The main purpose of this research was to study and optimize the mechanical properties of epoxy composite materials reinforced with MWCNT as a nanofiber and with high-impact polystyrene as a thermoplastic phase. Another important factor in epoxy-based nanocomposite sample preparation is the weight percentage of the hardener. Although determination of the appropriate amount of hardener content is based on stoichiometric ratio, but with the presence of thermoplastic phase as toughening agent and also nanofiller in epoxy resin, probability of complete mixture of epoxy monomers and hardener will dramatically decrease and hence prevent complete polymerization. Mirmohseni and Zavareh [37, 45] determined the optimum amount of hardener according to maximum tensile and impact strength of the prepared epoxy samples. Samples were prepared in 27 levels to consider the effect of three parameters – MWCNT, HIPS and hardener – on mechanical properties. In the present study, the weight percentages of MWCNT were 0.5, 2 and 4; the weight percentages of HIPS were 2, 6 and 10; and, finally, the parts per hundred resin (phr) for the hardener were 23, 27and 31. Tensile, compression, flexural and Izod impact tests were performed on the nanocomposites samples. Each test solution was given as input to the neural network. This model was used as a fitness function for the GA for optimization and for obtaining the best mechanical result.

2 Experimental

2.1 Material details

The multi-walled carbon nanotubes used in this study as a nano reinforcement in epoxy matrix were purchased from Research Institute of Petroleum Industry (RIPI) of Iran with an outer diameter 10–20 nm, purity of more than 95% and maximum length of <30 μm. Liquid diglycidyl ether of bisphenol A as the type of epoxy resin, Epon 828, with an epoxy equivalent weight of 185–192 g/equiv, was provided by Shell Chemicals Co. (TX, USA). The curing agent, cycloaliphatic polyamine (Aradur^® 42, Huntsman Chemical Products, Basel, Switzerland), was used nominally. The high-impact polystyrene in the current study as a thermoplastic in matrix was purchased from Tabriz Petrochemical Company, Tabriz, Iran. The solvent used was tetrahydrofuran (THF; Applichem Co., Gatersleben, Saxony-Anhalt, Germany) with a purity (GC) of more than 99%.

2.2 Measurements of mechanical properties

Tensile tests were performed according to the ASTM-D638 specification with a loading rate of 5 mm/min. Tensile testing was carried out on dumbbell-shaped samples, with 15 mm as a gauge length. The properties of ultimate tensile strength, tensile modulus and elongation at break were calculated after testing. The flexural properties of the prepared composites were determined using the three-point bending tests according to the ASTM-D790 method for obtaining flexural strength. The length-to-depth ratio of the samples was kept at 1:16. Flexural strength was measured by specimen dimensions equal to 3±0.5×14±0.5 mm in cross section and 80±1 mm in length. The crosshead speed used for the flexural samples was set at 2 mm/min. The compression tests were conducted according to ASTM D695 with specimen dimensions of 12.7×12.7×25.4 mm. The rate of motion of the driven grip when the testing machine was running was 1 mm/min. All the results of these mechanical properties were measured by an STM-150 universal testing machine from Santam Company (Iran). Notched Izod impact properties were evaluated using an SIT-50 Izod impact machine from Santam Company. The impact test followed the ASTM D-256 method. In each case, at least five sample specimens were used to calculate the mean values and standard deviation. All experiments were performed at room temperature.

2.3 Scanning electron microscopy analysis

A scanning electron microscope (SEM 1530) from TECNAN (Los Arcos, Navarra, Spain) was utilized to observe the dispersion of fracture surfaces of the cured composites. The fracture surface was gold coated prior to scanning electron microscopy (SEM) studies to avoid charging and was examined at 15 kV of accelerating voltage.

2.4 Preparation of neat resin specimen

Resin specimen was prepared by mixing the epoxy and hardener thoroughly with magnetic stirrer. Air bubbles were removed by putting the mixture under a vacuum. After that, the mixture was poured into silicone molds and cured at room temperature for 24 h. This process was followed by curing the mixture at 50°C, 80°C, 120°C and 150°C each for 2 h.

2.5 Preparation of nanocomposites

THF was selected as a solvent to achieve homogenous mechanism because of the ability of this solution to solve epoxy/HIPS/MWCNT components, it can be easily eliminated at 50°C under a vacuum and decrease the viscosity of DGEBA for better dispersion of MWCNT and HIPS in the matrix. At first, HIPS was solved in THF as solvent. The weight ratio of HIPS to THF was kept at 1:20. Then epoxy and MWCNT were added into the mixture. The mixing time, rotational speed and temperature of the magnetic heater stirrer were set at 2 h, 50 rpm and 50°C, respectively. The mixture was poured into an erlen vacuum; the solvent should have evaporated completely under the vacuum condition created by the vacuum pump. To prepare the homogenous mixture, ultrasonication (ultrasonic SONOPULS-HD 3200, product with Banselin Co., Germany, 20 kHz, amplitude of 80% of the maximum and pulsation; on for 10 s and off for 3 s) was performed for 30 min. After the ultrasonication, a hardener was added to the mixture and then poured into the silicone molds. This process was followed by placing the mixture under vacuum to remove air bubbles using a vacuum pump. The previous protocol must be followed respectively for neat epoxy for curing.

3 Modeling and optimization

3.1 Artificial neural network approach

Artificial neural network is a model for processing of information patterned after the human brain by mimicking the biological neural networks. This method has been developed in a wide variety of structures. Elman networks, radial basis functions, Hopfield networks and feed forward are some examples of the different forms of neural networks. But they have many common features. Neurons are the smallest unit of a constructive ANN. In all of them, neurons are connected to each other by weighted links that pass signals. One of the simplest models of the structure of a neural network is multi-layer perceptron with a layer as input, one or more layer as a hidden layer and an output layer. In this structure, all the neurons in one layer are connected to all neurons of the next layer by a series of weights that are shown in Figure 1.

Figure 1

Multi-layer perceptron with one hidden layer.

Many researchers have shown that networks with only one hidden layer are sufficient to solve complex problems [54–56]. Among the different types of perceptron, the most commonly applied is the feed forward architecture trained using the back-propagation algorithm [56]. During the training process, the weights are changed to obtain the exact value. Output u(x) for each node is calculated from the following formula:

where w_ji is the weight that links the ith node in the input layer to the jth node in the hidden layer, x_i is the ith input and w₀ is the weight that links the bias node to the j+1 node in hidden layer. Obviously, the summation function is not the network response. So to receive the final neuron output, u(x) is passed through the activation function f(u), typically a sigmoid step function and tan h. Before using the network, it should be trained in order to give it randomly initialized weights as input data. This should produce the target output as the network’s response (t_l). After that, the training set, which has many patterns, could be run through the network and the result recorded (o_l). After processing of the samples, the results are used to modify the weights of the layers. It begins with the output layer and spreading backward through the network. To determine the error of each output pair, the difference in target output is evaluated. So the result is multiplied to the first derivative of the activation function to estimate an error signal (δ_l) as presented in the following equation:

The new path weight can be estimated from the following formula:

where η is the rate of learning, Δw_lj is the change in the previous weight and α is the coefficient of momentum. The last one, α, must be added to prevent stalling the training process at the local instead of the global error minima.

A similar process must be followed for the input to the hidden layer path weights, but the signal error (δ_j) is calculated in this step from the following formula:

is the first derivative of the activation function and Σ_iδ_lw_ji is the sum of the product of the output layer δ_l and the relevant weight w_ji for that path. After this, the process of changing the weights is as described by Eq. (3), but with the parameters being the hidden layer outputs, weights and previous weight changes.

3.2 Genetic algorithm approach

In the 19th century, research by Charles Darwin led to the discovery of the basic rules in genetic science. The result of his research showed that change helps a person in his/her struggle for survival to have more chances to remain alive. Assuming that change in a person transfers to the next generation by inheritance, Darwin considered the evolution as a natural selection from an inheritable change. Holland in the early 1970s developed an optimization method based on the principles of genetic science. Genetic algorithm belongs to a large class of evolutionary algorithm. Genetic algorithm applies the best conservation rules to obtain a better response. Genetic algorithms obtain a response from each generation by the selection process which mimics natural genetic rules. The target of optimization is to find the general minimum or maximum.

3.2.1 Characterization of genetic algorithm

Genetic algorithm uses chromosomes as a code instead of applying it to their variables. So the result has low associations to its own problem. So we can guess that this algorithm finds the answer to a wide range of problems.
Genetic algorithm can govern a large number of answer spaces at the same time. This trait reduces the high possibility that the algorithm is being trapped in the local optimum points.
This algorithm is easily applied for solving problems that have a large number of variables.
Genetic algorithm is simple and it does not need auxiliary information like the derivative of an objective function. So this algorithm can be used for optimization of complicated objective functions, as discontinuous or non-differentiable functions or systems have no specific mathematical definition.
Genetic algorithm can provide a set of optimal responses instead of just one answer. This characterization is important for finding answers for multi-objective optimization.

3.2.2 Applying a genetic algorithm

At first, GA starts working with a population of chromosomes that were selected randomly. To obtain a better generation, genetic operators are applied on the population. Reproduction, mutation and crossover are the most common genetic operators. Reproduction operator selects coupling people who produce the next generation. People who are more fit have greater chances of producing an offspring, but people who are less fit still have a chance of getting selected because they may have valuable genes. Then in the next step two strings were selected for marriage as selected randomly in the previous step. Crossover operator allows the two selected chromosomes to share their structure based on specific probabilities. Which chromosomes are passed and where discrete happen according to a specific probability. This operation creates a pair of new chromosomes that include the characteristics of their parents. After reproduction and crossover, the mutation operator is applied to each produced chromosomes. A randomly changing value of a bit in each string is called a mutation. By applying the mutation operator on a bit, if the value of the gene was 1, it is changed to 0 and if its value was 0, then it is changed to 1.

4 Results of optimization

In this work, the effect of MWCNT and HIPS on the mechanical properties of a DGEBA type of epoxy resin has been studied. Table 1 contains the results of the ultimate tensile strength (UTS), ultimate bending strength (UBS), ultimate compression strength (UCS) and impact strength.

Table 1

Values of ultimate tensile strength, ultimate bending strength, ultimate compression strength and impact strength.

Run no.	Experimental factors			Result of mechanical tests
Run no.	MWCNT content (wt%)	HIPS content (wt%)	Hardener content (phr)	Tensile strength (MPa)	Flexural strength (MPa)	Compression strength (MPa)	Impact strength (kJ/m²)
1	0.5	2	23	43.9±4	3.00±0.55	193±17	32.5±5
2	0.5	2	27	57.2±6	4.20±0.26	202±19	35.8±8
3	0.5	2	31	52.5±5	4.40±0.34	176±16	34.8±7
4	0.5	6	23	43.1±3	3.05±0.32	183±14	33.5±6
5	0.5	6	27	57.8±4	4.10±0.51	198±18	37.8±8
6	0.5	6	31	43.0±3	4.69±0.42	166±14	36.1±7
7	0.5	10	23	36.5±3	2.29±0.48	124±15	23.9±4
8	0.5	10	27	50.1±4	3.46±0.71	133±18	26.7±3
9	0.5	10	31	45.7±3	3.68±0.56	103±16	25.2±5
10	2	2	23	44.8±4	3.06±0.64	196±18	32.9±6
11	2	2	27	58.2±5	4.16±0.23	209±20	36.6±5
12	2	2	31	54.2±5	4.41±0.46	176±14	35.1±4
13	2	6	23	44.2±3	2.99±0.29	195±15	34.9±6
14	2	6	27	59.1±4	4.26±0.78	204±21	37.0±7
15	2	6	31	52.8±2	4.45±0.69	173±19	36.6±5
16	2	10	23	37.9±3	1.99±0.28	129±12	23.4±3
17	2	10	27	51.3±3	3.29±0.21	136±11	27.0±4
18	2	10	31	47.9±3	3.56±0.19	108±10	26.9±2
19	4	2	23	26.9±2	2.18±0.45	114±12	22.6±4
20	4	2	27	41.1±4	3.46±0.51	126±14	25.2±3
21	4	2	31	36.6±3	3.66±0.36	96±08	24.3±6
22	4	6	23	26.9±2	2.26±0.39	108±13	23.7±5
23	4	6	27	39.8±4	3.44±0.22	119±14	27.6±4
24	4	6	31	34.2±3	3.76±0.54	86±08	26.1±3
25	4	10	23	19.9±1	1.29±0.19	44±06	13.9±2
26	4	10	27	31.2±2	2.56±0.12	53±09	16.1±2
27	4	10	31	27.4±3	2.78±0.34	22±05	15.1±1

4.1 Result and discussion

4.1.1 Tensile properties

According to the mentioned algorithm, for maximum tensile properties the optimum conditions found were as follows: for ultimate tensile strength, MWCNT content of 1.3%, HIPS content of 3.6% and for hardener content of 28 phr. The maximum value of the ultimate tensile strength predicted by the model at optimum values of variables was 62.5 MPa. After preparing the samples for validation of the predicted results at optimum conditions, the average value obtained in the experiment for ultimate tensile strength for five samples was 61 MPa. By preparing the samples according to optimum conditions and testing them, verification of the predicted results has been performed. Figure 2 shows the stress-strain plot for the tensile property. This plot shows the synergic effects of the combination of HIPS and MWCNT at optimum conditions on the ultimate tensile strength and elongation at break. In Figure 2, note that ultimate tensile strength increased by up to 52% of the neat epoxy value, 35% of the epoxy/HIPS value (at 3.6 wt% of HIPS) and 16% of the epoxy/MWCNT value (at 1.3 wt% of MWCNT). One obvious note here is that the tensile modulus was reduced a little in comparison with other properties. From the plot, it can be seen that the elongation at break for a composite reinforcement with HIPS and MWCNT improved by up to 223% of the neat epoxy value, 293% of the epoxy/HIPS value and 57% of the epoxy/MWCNT value.

Figure 2

Stress-strain plot of the tensile property.

Figure 3 shows three-dimensional response surface plots of the tensile strength of an epoxy/HIPS/MWCNT ternary nanocomposite according to the modeling method mentioned in Section 3. These plots have been employed in order to show the dependence of the mechanical properties of epoxy-based hybrid nanocomposite on effective parameters as design factors. Figure 3A shows the effect of hardener and MWCNT loading on ultimate tensile strength while the HIPS factor was fixed at an optimum value (3.6 wt%). As can be seen from the results, MWCNT content and hardener loading have significant effects on the measured ultimate tensile strength. The ultimate tensile strength improved with increasing MWCNT content to some extent and decreased with higher loadings, whereas hardener loading has a similar behavior. Well-dispersed nanofillers at low concentrations increase the epoxy ultimate tensile strength. Figure 3B shows the effect of MWCNT and HIPS as a thermoplastic phase while the hardener was fixed at an optimum value (28 phr). The result showed that HIPS content has a poor effect on tensile strength. In other words, increasing or decreasing amounts of HIPS have a weak effect on ultimate tensile strength compared to those of the hardener and MWCNT. Figure 3C also proves this claim. Based on this figure, the effect of HIPS and hardener on tensile strength has been observed while MWCNT was fixed at an optimum value (1.3 wt%)

Figure 3

Three-dimensional plots of UTS for (A) HIPS (3.6 wt%), (B) hardener (28 phr) and (C) MWCNT (1.3 wt%).

4.1.2 Flexural properties

The value of ultimate flexural strength is shown in Table 1. The optimization algorithm mentioned in Section 3 found the following optimum conditions for maximum flexural strength: for ultimate flexural strength, MWCNT content of 0.8%, HIPS content of 4.2% and hardener content of 30 phr. The optimization method predicted that the maximum value of ultimate flexural strength at optimum values of variables was 5.2 MPa. After preparing the samples for validating the predicted results at optimum conditions, the average value obtained in the experiment for ultimate flexural strength for five samples was 5 MPa. Then the samples were prepared to verify the predicted results according to optimum conditions. Figure 4 shows the stress-strain plot for the flexural property. This plot indicates a synergistic effect only for elongation at break, but no improvement was seen for ultimate flexural strength of the epoxy/MWCNT/HIPS mechanism. This combination reduced ultimate flexural strength by 17% of the neat epoxy value, decreased it by 27% of the epoxy/MWCNT value (at 0.8 wt% of MWCNT) and improved it by up to 57% of the epoxy/HIPS value (at 4.2 wt% of HIPS). Figure 4 shows that elongation at break increased by up to 36% of the neat epoxy value, 294% of the epoxy/HIPS value (at 4.2 wt% of HIPS) and 201% of the epoxy/MWCNT value (at 0.8 wt% of MWCNT). One obvious point here is that the flexural modulus was reduced a little in comparison with the epoxy/MWCNT/HIPS ternary nanocomposite and epoxy/MWCNT value.

Figure 4

Stress-strain plot of the flexural property.

Figure 5 shows three-dimensional response surface plots of flexural strength for a combination of epoxy, MWCNT and HIPS according to the modeling method mentioned in Section 3. In fact, these plots represent the effect of effective parameters as design factors on ultimate flexural strength. Figure 5A shows the effect of MWCNT and hardener on flexural strength while HIPS was fixed at an optimum value (4.2 wt%). Figure 5A shows that the best flexural result happened at low concentrations of MWCNT as with the tensile properties. Figure 5B proves this claim. Based on this figure, the effect of MWCNT and HIP on flexural strength has been observed while hardener content was fixed at an optimum value (30 phr). Figure 5B,C shows that increasing or decreasing amounts of HIPS at low concentrations have a low effect on flexural strength, but increasing amounts of HIPS content at higher concentrations reduced flexural strength.

Figure 5

Three-dimensional plots of UBS for (A) HIPS (4.2 wt%), (B) hardener (30 phr) and (C) MWCNT (0.8 wt%).

4.1.3 Compression properties

The maximum compression results found based on the previously mentioned optimization method were as follows; for ultimate compression strength, MWCNT content of 1.4%, HIPS content of 3.6% and hardener content of 26 phr. The ultimate compression strength predicted by the optimization method at optimum values of the variables was 220 MPa. Then after preparing epoxy/HIPS/MWCNT ternary nanocomposites at optimum conditions and after testing, a value of 215 MPa was observed for ultimate compression strength. Figure 6 shows the synergic effects of the combination of epoxy, HIPS and MWCNT at optimum conditions on ultimate compression strength and elongation at break and elastic module. Figure 6 shows that the ultimate compression strength and elongation at break of this combination increased by up to 43% and 28% of the neat epoxy value, respectively. Also, epoxy/MWCNT/HIPS ternary nanocomposites improved ultimate compression strength by up to 18% and increased elongation at break by up to 14% of the epoxy/HIPS value (at 3.6 wt% of HIPS). From the plot, it can be seen that ultimate compression strength and elongation at break for a composite reinforcement with HIPS and MWCNT decreased by 4% and improved by up to 16% of the epoxy/MWCNT value (at 1.4 wt% of MWCNT), respectively.

Figure 6

Stress-strain plot of the compression property.

Figure 7 shows three-dimensional response surface plots of the compression strength of epoxy/HIPS/MWCNT ternary nanocomposites according to the modeling method mentioned in Section 3. Figure 7A shows the effect of hardener and MWCNT loading on ultimate compression strength while the HIPS factor was fixed at an optimum value (3.6 wt%). As can be seen from the results that MWCNT content and hardener loading have significant effects on the measured ultimate compression strength. The ultimate compression strength improved with increasing MWCNT content to some extent and decreased with higher loadings, whereas the hardener has a similar behavior. Figure 7B shows the effect of MWCNT and HIPS as a thermoplastic phase while the hardener was fixed at an optimum value (26 phr). From Figure 7A and B, it can be observed that the best ultimate compression strength occurred at low amounts of MWCNT content. Figure 7C shows the effect of HIPS and hardener content on compression strength while the MWCNT factor was fixed at an optimum value (1.4 wt%). From this plot, the poor effect of HIPS content on compressive strength can be seen. In other words, increasing or decreasing amounts of HIPS have a weak effect on ultimate compression strength compared to hardener and MWCNT.

Figure 7

Three-dimensional plots of UCS for (A) HIPS (3.6 wt%), (B) hardener (26 phr) and (C) MWCNT (1.4 wt%).

4.1.4 Impact properties

The maximum results of impact strength found had the following conditions: MWCNT content of 1.3%, HIPS content of 4.5% and hardener content of 28 phr. The best result of impact strength predicted by the modeling method was 41 kJ/m². Then samples were prepared using a combination of epoxy, MWCNT and HIPS at optimum values to validate the predicted results, and after testing, the average value found in the experiment for five samples was 40 kJ/m². Figure 8 shows that the combination of epoxy, HIPS and MWCNT improved impact strength by up to 334% of the neat epoxy value and up to 233% of the epoxy/MWCN value at the optimum value of MWCNT (1.3 wt%) but decreased by 12% that of the epoxy/HIPS value (4.5 wt% of HIPS).

Figure 8

Impact strength of different mechanisms.

Figure 9 shows three-dimensional response surface plots of the impact strength of epoxy/HIPS/MWCNT ternary nanocomposites according to modeling and optimization method mentioned in part 3. Figure 9A shows the effect of MWCNT and hardener on impact strength. From the figure, it can be seen that the best impact strength occurred at low contents of MWCNT as in the previous steps. Figure 9B shows the effect of HIPS and MWCNT on impact properties. By increasing the amount of HIPS, impact strength improved and at higher loadings of HIPS impact properties decreased.

Figure 9

Three-dimensional plots of impact for (A) HIPS (4.5 wt%), (B) hardener (28 phr) and (C) MWCNT (1.3 wt%).

Figure 10 shows a cut surface of a tensile specimen at optimum amounts of HIPS, MWCNT and hardener. As can be seen from the figure, good dispersion of MWCNT nanofillers and HIPS as thermoplastic phase with little agglomeration has occurred. Also, it can be seen that there was a phase separation of nano- and micro particles in the epoxy-rich matrix. This modifier with homogenous dispersion can play an important role as crack stoppers and also as reinforcements in enhancing mechanical strengths [57]). In agreement with SEM investigation of the epoxy/HIPS/MWCNT samples, a homogeneous combination of MWCN nanofillers in the matrix is shown.

$Figure 10 Scanning electron micrographs of the fracture surface of (A) neat epoxy samples having (B) MWCNT (1.3 wt%), (C) HIPS (3.6 wt%) and (D) MWCNT (1.3 wt%)+HIPS (3.6 wt%).$

Figure 10

Scanning electron micrographs of the fracture surface of (A) neat epoxy samples having (B) MWCNT (1.3 wt%), (C) HIPS (3.6 wt%) and (D) MWCNT (1.3 wt%)+HIPS (3.6 wt%).

5 Conclusion

In the current study, a new combination of thermoplastic nanofiller as a modifier of epoxy-based composites has been proposed. In this article, tensile, flexural, compression and impact strength are the four different mechanical properties that were studied. In order to enhance mechanical properties, ANN was used to present a model and GA was used to optimize the mentioned mechanical properties. Also, the effect of the parameters on the mechanical strength of epoxy/HIPS/MWCNT ternary nanocomposites is represented by a three-dimensional response surface and a two-dimensional contour plot. From the result, it was found that the combination of HIPS and MWCNT nanofillers significantly increased the tensile, compression and impact strength of neat resin by up to 52%, 43% and 334%, respectively, but flexural strength did not change positively. Also, elongation at break of the tensile, flexural and compression properties of neat epoxy rose by up to 223%, 36% and 26%, respectively. A correlation between morphology and mechanical properties was obtained using a SEM technique.

Corresponding author: Amin Hamed Mashhadzadeh, Department of Mechanical Engineering, Semnan Branch, Islamic Azad University, 48164-194 Semnan, Iran

References

[1] Frisch HL, Mark JE. Chem. Mater. 1996, 8, 1735–1738.Search in Google Scholar

[2] Park J, Jana SC. Macromolecules 2003, 36, 8391–8397.10.1021/ma0303465Search in Google Scholar

[3] Guo H, Zhang J, Chen Z, Wei J, Zhao R, Liu X. Journal 2012, 19, 347.10.1515/secm-2012-0021Search in Google Scholar

[4] Krumova M, Klingshirn C, Haupert F, Friedrich K. Compos. Sci. Technol. 2001, 61, 557–563.Search in Google Scholar

[5] Gojny FH, Nastalczyk J, Roslaniec Z, Schulte K. Chem. Phys. Lett. 2003, 370, 820–824.Search in Google Scholar

[6] Bai JB, Allaoui A. Compos Part A: Appl. Sci. Manuf. 2003, 34, 689–694.Search in Google Scholar

[7] Sandler J, Shaffer MSP, Prasse T, Bauhofer W, Schulte K, Windle AH. Polymer 1999, 40, 5967–5971.10.1016/S0032-3861(99)00166-4Search in Google Scholar

[8] Schadler LS, Giannaris SC, Ajayan PM. Appl. Phys. Lett. 1998, 73, 3842–3844.Search in Google Scholar

[9] Cui S, Canet R, Derre A, Couzi M, Delhaes P. Carbon 2003, 41, 797–809.10.1016/S0008-6223(02)00405-0Search in Google Scholar

[10] Ci L, Bai J. Compos. Sci. Technol. 2006, 66, 599–603.Search in Google Scholar

[11] Laborde-Lahoz P, Maser W, Martínez T, Benito A, Seeger T, Cano P, Villoria RGd, Miravete A. Mech. Adv. Mater. Struct. 2005, 12, 13–19.Search in Google Scholar

[12] Liao Y-H, Marietta-Tondin O, Liang Z, Zhang C, Wang B. Mater. Sci. Eng.: A 2004, 385, 175–181.10.1016/S0921-5093(04)00857-3Search in Google Scholar

[13] Zhu J, Peng H, Rodriguez-Macias F, Margrave JL, Khabashesku VN, Imam AM, Lozano K, Barrera EV. Adv. Func. Mater. 2004, 14, 643–648.Search in Google Scholar

[14] Montazeri A, Khavandi A, Javadpour J, Tcharkhtchi A. Mater Des 2010, 31, 3383–3388.10.1016/j.matdes.2010.01.051Search in Google Scholar

[15] Allaoui A, Bai S, Cheng HM, Bai JB. Compos. Sci. Technol. 2002, 62, 1993–1998.Search in Google Scholar

[16] Gojny FH, Wichmann MHG, Köpke U, Fiedler B, Schulte K. Compos. Sci. Technol. 2004, 64, 2363–2371.Search in Google Scholar

[17] Gojny FH, Wichmann MHG, Fiedler B, Schulte K. Compos. Sci. Technol. 2005, 65, 2300–2313.Search in Google Scholar

[18] Zhou Y, Pervin F, Rangari VK, Jeelani S. Mater. Sci. Eng.: A 2006, 426, 2212–228.10.1016/j.msea.2006.04.031Search in Google Scholar

[19] Subramaniyan AK, Sun CT. Compos. Part A: Appl. Sci. Manuf. 2006, 37, 2257–2268.Search in Google Scholar

[20] Iwahori Y, Ishiwata S, Sumizawa T, Ishikawa T. Compos. Part A: Appl. Sci. Manuf. 2005, 36, 1430–1439.Search in Google Scholar

[21] Cho J, Chen JY, Daniel IM. Scripta Mater. 2007, 56, 685–688.Search in Google Scholar

[22] Shahid N, Villate RG, Barron AR. Compos. Sci. Technol. 2005, 65, 2250–2258.Search in Google Scholar

[23] Dean D, Obore AM, Richmond S, Nyairo E. Compos. Sci. Technol. 2006, 66, 2135–2142.Search in Google Scholar

[24] Siddiqui NA, Woo RSC, Kim J-K, Leung CCK, Munir A. Compos. Part A: Appl. Sci. Manuf. 2007, 38, 449–460.Search in Google Scholar

[25] Timmerman JF, Hayes BS, Seferis JC. Compos. Sci. Technol. 2002, 62, 1249–1258.Search in Google Scholar

[26] Garg AC, Mai Y-W. Compos. Sci. Technol. 1988, 31, 179–223.Search in Google Scholar

[27] Kinloch AJ, Shaw SJ, Tod DA, Hunston DL. Polymer 1983, 24, 1341–1354.10.1016/0032-3861(83)90070-8Search in Google Scholar

[28] Brooker RD, Kinloch AJ, Taylor AC. J. Adhes. 2010, 86, 726–741.Search in Google Scholar

[29] Sultan JN, McGarry FJ. Polym. Eng. Sci. 1973, 13, 29–34.Search in Google Scholar

[30] Pearson RA, Yee AF. J. Mater. Sci. 1991, 26, 3828–3844.Search in Google Scholar

[31] Frounchi M, Mehrabzadeh M, Parvary M. Polym. Int. 2000, 49, 163–169.Search in Google Scholar

[32] Chen TK, Jan YH. J. Mater. Sci. 1992, 27, 111–121.Search in Google Scholar

[33] Ramakrishna HV, Priya SP, Rai SK. J. Appl. Polym. Sci. 2007, 104, 171–177.Search in Google Scholar

[34] Torres A, López-de-Ullibarri I, Abad MJ, Barral L, Cano J, García-Garabal S, Díez FJ, López J, Ramírez C. J. Appl. Polym. Sci. 2004, 92, 461–467.Search in Google Scholar

[35] Yun NG, Won YG, Kim SC. Polym. Bull. 2004, 52, 365–372.Search in Google Scholar

[36] Kimoto M, Mizutani K. J. Mater. Sci. 1997, 32, 2479–2483.Search in Google Scholar

[37] Mirmohseni A, Zavareh S. Mater. Des. 2010, 31, 2699–2706.Search in Google Scholar

[38] Mimura K, Ito H, Fujioka H. Polymer 2000, 41, 4451–4459.10.1016/S0032-3861(99)00700-4Search in Google Scholar

[39] Rose LRF. Mech. Mater. 1987, 6, 11–15.Search in Google Scholar

[40] Faber KT, Evans AG. Acta Metall. 1983, 31, 577–584.Search in Google Scholar

[41] Fu S-Y, Mai Y-W, Lauke B, Yue C-Y. Mater. Sci. Eng.: A. 2002, 323, 326–335.Search in Google Scholar

[42] Maazouz A, Sautereau H, Gerard JF. J. Appl. Polym. Sci. 1993, 50, 615–626.Search in Google Scholar

[43] Kinloch AJ, Mohammed RD, Taylor AC, Eger C, Sprenger S, Egan D. J. Mater. Sci. 2005, 40, 5083–5086.Search in Google Scholar

[44] Geisler B, Kelley FN. J. Appl. Polym. Sci. 1994, 54, 177–189.Search in Google Scholar

[45] Mirmohseni A, Zavareh S. J. Polym. Res. 2011, 18, 509–517.Search in Google Scholar

[46] Calcaterra S, Campana G, Tomesani L. J. Mater. Process. Technol. 2000, 104, 74–80.Search in Google Scholar

[47] Yescas MA, Bhadeshia HKDH, MacKay DJ. Mater. Sci. Eng.: A. 2001, 311, 162–173.Search in Google Scholar

[48] Masri SF, Chassiakos AG, Caughey TK. J. Appl. Mech. 1993, 60, 123–133.Search in Google Scholar

[49] Rhim J, Lee SW. Comput. Mech. 1995, 16, 437–443.Search in Google Scholar

[50] Song RG, Zhang QZ. Mater. Sci. Eng.: C. 2001, 17, 133–137.Search in Google Scholar

[51] Chung JS, Hwang SM. J. Mater. Process. Technol. 1997, 72, 69–77.Search in Google Scholar

[52] Song R-G, Zhang Q-Z, Tseng M-K, Zhang B-J. Mater. Sci. Eng.: C. 1995, 3, 39–41.Search in Google Scholar

[53] Mousavi Anijdan SH, Madaah-Hosseini HR, Bahrami A. Mater. Des. 2007, 28, 609–615.Search in Google Scholar

[54] Funahashi K-I. Neural Netw. 1989, 2, 183–192.Search in Google Scholar

[55] Hartman EJ, Keeler JD, Kowalski JM. Neural Comput. 1990, 2, 210–215.Search in Google Scholar

[56] Khandetsky V, Antonyuk I. NDT E Int. 2002, 35, 483–488.Search in Google Scholar

[57] López J, Ramírez C, Abad MJ, Barral L, Cano J, Díez FJ. J. Appl. Polym. Sci. 2002, 85, 1277–1286.Search in Google Scholar

Received: 2012-12-21

Accepted: 2013-1-31

Published Online: 2013-03-15

Published in Print: 2013-08-01

This article is distributed under the terms of the Creative Commons Attribution Non-Commercial License, which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

Articles in the same Issue

https://doi.org/10.1515/secm-2012-0178

Keywords for this article

epoxy; high-impact polystyrene; mechanical properties; multi-walled carbon nanotubes; toughness

Creative Commons

BY-NC-ND 3.0