AMSE JOURNALS –2014-Series: Modelling A; Vol.

87; N° 2; pp 70-88
Submitted April 2014; Revised May 17, 2014; Accepted June 20, 2014
 
“ANFIS Prediction of the Polymer and Polymer Composite properties
and its Optimization Technique”

*Vikrant Gupta, ** Singh Rama, ***Manoj Kumar Jha, **** M. F. Qureshi

*Department of Computer Science, Batmool Ashram College, Raigarh, India

**Department of Chemistry, Rungta College of Engg. and technology, Raipur, India
***Department of Computer Science, Rungta Engg. College, Raipur, India
****Department of Electrical Engg., Govt. Polytechnic, Dhamtari, India
([email protected]; [email protected]; [email protected];
[email protected])

Abstract.
Prediction and optimization of polymer properties and polymer composite properties are a
complex and highly non-linear problem with no any easy method to predict polymer properties
directly and accurately. The effect of modifying a monomer (polymer repeat unit) on
polymerization and the resulting polymer properties is not an easy task to investigate
experimentally, given the large number of possible changes. We utilize a database of polymer
properties to train the ANFIS, which accurately predict specific polymer properties. In polymer
composites, a certain amount of experimental results is required to train a well-designed ANFIS.
The ANFIS approach for predicting certain properties of polymer composite materials are discussed
here. These include fatigue life; wear performance, response under combined loading situations,
and dynamic mechanical properties. Prediction of effective thermal conductivity (ETC) of different
fillers filled in polymer matrixes is proposed. The finding shows that ANFIS demonstrates high
prediction accuracy as reflected by the small root mean square error (RMSE) value and high
correlation coefficient (r) and coefficient of determination (R2) values. ANFIS prediction results are
found to be compatible to linear regression estimations. The goal of this paper is to promote more
consideration of using ANFIS in the field of polymer composite property prediction and design.
The predicted results by ANFIS are in good agreements with experimental values. The predicted
results also show the supremacy of ANFIS in comparison with other earlier developed models.
Keywords: ANFIS, Prediction, Polymer properties, Polymer Composites
70  
 

 
Properties, Effective Thermal Conductivity (ETC)

1. Introduction
Experiments on the production of different characteristics of polymer composites are
normally conducted in the labs. Lab research can be very costly and time consuming. Alternatively,
researchers are looking into other methods of studying the properties of polymer composites
produced by using computer application models. In our study presented in this paper, the physical
properties of polymer composites modeled using ANFIS (Adaptive Neuro-Fuzzy Inference
System). Identifying the suitable composition of polymer with other agents and filler in the
production of polymer composites is essential in producing engineering products. The objectives of
this study are:
(i) to develop a computer application model ANFIS that can be used to find the suitable
combination of polymer with other agents and filler in the production of polymer composites with
different physical characteristics.
(ii) to assess the ability of ANFIS in predicting the properties of polymer composites by
comparison with Linear Regression prediction results.
The proposed computer application prediction tool ANFIS is not to replace the conventional
lab experiments or substitute the traditional statistical modeling techniques; instead it is to
strengthen the present system by providing a simple simulation tool which can be useful in studying
the input-output relationship in prediction of properties of polymer composites Besides being highly
non-linear, there are a large number of parameters that need to be accurately defined if such systems
are to be properly characterized. The application of polymer composites as engineering materials
has become state of the art. To design the characteristics of polymer composites is the most
important advantage. In order to meet a special target of engineering application, e.g. concerning
one or several measurable material properties, polymer composites can be designed by selecting the
correct composition and choosing the appropriate manufacturing process, as schematically
illustrated in Fig. 1. Property investigation plays a key role in materials science to evaluate
composites designed for special engineering applications. All three stages shown in Fig. 1 are not
separated, but interconnected, and the integration can be summarized as composite design,
processing optimization and property relationships. The first two fields correspond to the interaction
between the selected compositions or the manufacturing process and the properties investigated,
whereas the last relates to possible correlations between some simple measured parameters (e.g.
modulus, strength and failure strain) and more complex properties (e.g. fatigue, wear, combined
loading and creep). The understanding of all these relationships is important in composite
71  
 

 
materials science, in order to meet the requirements for particular engineering applications.
Modeling of these relationships generally involves the development of a mathematical tool derived
from experimental data; once established it can significantly reduce the experimental work involved
in designing new polymer composites. For this reason, ANFIS has recently been introduced into the
field of polymer composites.

Fig. 1. Schematic presentation of composition selection, manufacturing process, and property

investigation in composite materials science.

Polymer composites with high thermal conductivity and low dielectric constant are highly
desirable for use in various applications, such as electric stress control, electromagnetic shielding,
and higher storage capability of the electric energy. Polymer matrixes are commonly used such as
polyethylene, polypropylene, polyurethane, polyvinyl chloride etc. which are good thermally and
electrically insulators. Due to the increasing use of composite materials in many industrials sectors,
including transformation, electronic, and energy supply and production, there is a renewed interest
in simulation techniques to estimate the ETC of fiber and particle filled polymer composites.
Dependence of the ETC of these materials on porosity, shape factor and packing of the particles is a
matter of concern to engineers, mathematicians, and physicists. Thermal conductivity of boron
nitride (BN) reinforced high density polyethylene (HDPE) composites was investigated under a
special dispersion state of BN particles in HDPE, and together with the influence on thermal
conductivity of particle sizes of filler used by Zhou et al. Xu et al. investigated the use of
aluminum nitride (AlN) and poly-vinylidene fluoride (PVDF) as the matrix. Gu et al. investigated
the content of AlN influencing the thermal conductivity and ultimate mechanical properties of AlN/
linear low-density polyethylene (LLDPE) composites. Adaptive neuro-fuzzy inference system
(ANFIS) has recently been introduced to predict the effective thermal conductivity of metal/non-
metal filled polymer composites. The fillers used most frequently are particles of carbon, aluminum,
copper, brass, graphite and magnetite. By the addition of fillers to polymer matrix the thermal
72  
 

 
conductivity of polymers can be increased remarkably. In this study, high-density polyethylene
(HDPE), low-density polyethylene (LDPE), linear low-density polyethylene (LLDPE), and
polyvinylidene fluoride (PVDF) with different metals/non-metals such as boron nitride (BN),
copper (Cu) and aluminum nitride (AIN) are used as inclusions, because of its superior mechanical
and physical properties. Here the variation of ETC of HDPE/BN composites with volume fraction
of filler, the variation of ETC of LDPE/Cu composites with volume fraction of filler, the variation
of ETC of LLDPE/Cu composites with volume fraction of filler, the variation of ETC of
PVDF/AlN composites with volume fraction of filler and the variation of ETC of LLDPE/AlN
composites with volume fraction of filler have been studied. HDPE is one of the most widely used
commercial polymers. However, its toughness, weather resistance, and environmental stress
cracking resistance are not good enough which limits its applications in many high-technology
areas. Reinforcing HDPE with fillers (viz., aluminum and copper particles, short carbon fibers,
carbon, graphite, aluminum nitrides and magnetic particles) has been found to improve its
properties. Low-density polyethylene (LDPE) is a thermoplastic made from petroleum. Compared
with LDPE, LLDPE possesses better strength, toughness, heat-resistance, cold resistance,
environmental stress cracking resistance, and tearing resistance properties. Rule-based modeling,
specifically using fuzzy logic rule is a soft-computing tool-based approach to construct a model for
the systems that are highly complex and exhibit non-linear behavior in nature, for which no well-
defined mathematical expression(s) exist. The effectiveness of the ANFIS approach is extensively
tested by comparing its results with those obtained in real experimentations as well as with those of
various existing empirical/semi-empirical models re-ported in literature.

2. Application of ANFIS in composite materials science

For materials research, a certain amount of experimental results is always needed first to
develop a well performing ANFIS, After the ANFIS has learned to solve the problems based on
these datasets, new data from the same knowledge domain can then be put into the trained neural
network, in order to output realistic solutions. The process of creating ANFIS for materials research
can, therefore, be summarized in terms of the following stages:
1. Database collection: analysis and pre-processing of the data.
2. Training of the neural network: this includes the choice of its architecture, training
functions, training algorithms and parameters of the network.
3. Test of the trained network: to evaluate the network performance.
4. Use of the trained ANFIS for simulation and prediction.
73  
 

 
 The greatest advantage of ANFIS is its ability to model complex non-linear, multi-
dimensional functional relationships without any prior assumptions about the nature of the
relationships, and the network is built directly from experimental data by its self-organizing
capabilities.
Evaluation of the ANFIS method
A dataset of measurement results will usually be divided into a training dataset and a test
dataset. The training dataset is used to adjust the weights of all the connecting nodes until the
desired error level is reached. Thereafter, the network performance is evaluated by using the test
dataset. The quality of the prediction can normally be characterized by the root mean square error
(RMSE) of the predicted values from the real measured data. The smaller the RMSE of the test
dataset is the higher, is the predictive quality.

3. ANFIS for polymer composites

Fatigue life
Fatigue is one of the most complicated problems for fiber composites, and failure
mechanisms are still not well understood. Extensive tests must be carried out because of the absence
of a well-defined failure criterion that can be used to predict fatigue failure in polymer composites.
ANFIS offer the possibility of developing models that will predict the behavior of composites
without being linked to mechanistic arguments.
Unidirectional (UD) composites
It seems that the fiber orientation in UD composites plays a key role in fatigue performance.
Applying fiber orientation as an input improves the ANFIS predictive quality significantly, even
with a relatively smaller dataset. In order to improve the prediction accuracy, ANFIS was
considered here using the same database employed in previous literatures.
Laminate composites
Three fatigue parameters, peak stress, minimum stress and probability of failure, and four
monotonic mechanical properties, tensile strength, compression strength, tensile failure strain and
tensile modulus, were selected as the ANFIS inputs, which were applied to predict the fatigue life of
the composites as the output. ANFIS was finally optimized by evaluating the changes in RMSE of
ANFIS output with the number of neurons in the intermediate layers. Once a well-trained ANFIS
was obtained, the possibility for predicting fatigue life of new materials could be analyzed. Further,
samples of two other materials, a fifth CFRP (HTA/982) laminate and a GFRP (E-glass/913)
laminate of similar structure, were tested with regard to their mechanical and fatigue properties.
74  
 

 
Wear of composites
The dataset are obtained from fretting tests with various material compositions under
different wear measuring conditions. An ANFIS is proposed with the output of wear volume; the
inputs were mechanical properties and test conditions, i.e. compressive strength, compression
modulus, compressive strain to failure, tensile strength, tensile strain to failure, impact strength,
environmental testing temperature, initial load, average load and average velocity. The design and
the training of the ANFIS were performed using the ‘ANFIS Toolbox’ of MATLAB.
Dynamic mechanical properties
An ANFIS approach has been proposed for the complex problems of fatigue, wear and
combined loading failure discussed earlier. Nevertheless, it is also interesting for dealing with some
relatively simple material problems, which will be of help to understand the characteristics of
ANFIS for polymer composite applications. Large training data are needed to reach a predictive
quality to that in a one-output ANFIS.
Processing optimization
In this study, a Kohonen self-organizing map type of neural network was applied to classify
the measured dataset. Another approach in a similar direction was performed for the optimization of
the polymerization process of polyamide 6.6 by Nascimento and Giudici. It was shown in literatures
that the optimal cure cycles of the ANFIS prediction were reasonably accurate in comparison of the
mean-square-error to the results based on the numerical process models. The use of the ANFIS in
lieu of the numerical models reduced the computational time for process simulations by several
orders of magnitude.

4. Data Set and ANFIS

We concentrate on polymer composites properties (mechanical, thermal, magnetic, optical,
electrical, environmental and deteriorative) and the relationships with their structures (microscopic,
mesoscopic and macroscopic). The prediction of polymer properties from just the structure of the
monomer is somewhat unreliable. But trained ANFIS that are given optimized input data do an
excellent job of characterizing a new modified polymer.
Polymer Database
The polymer database we used was selected carefully so as to include all types of polymers
available. Within each category, several different types of polymers were included to provide a
comprehensive set of data. The total data set consisted of 400 individual polymers. For each
polymer, information stored included its molecular weight (poly-dispersity), mechanical and
75  
 

 
thermal properties, chemical structure, and data reliability number.
Polymer Properties Selected
Impact resistance is very likely the most desired property of an engineering plastic. One
indicator of good impact resistance is the Tα / Tγ ratio and the dynamic elastic modulus. The higher
this ratio is, the better the impact resistance is. However, high impact resistance with almost no
elastic properties results in a brittle polymer that has no commercial use. Therefore, in the complete
description of the overall mechanical properties of a polymer these properties must be included.
ANFIS and Linear Regression Models
Fuzzy systems and Artificial Neural Networks are computer application approaches that
have been widely applied in various domains. The expressiveness of fuzzy if-then rules using
linguistic variables can be combined with the learning capability of neural networks to produce
Fuzzy Neural Network models. The input attributes of the developed ANFIS system are the
ingredients needed to produce polymer. These imprecise attributes are called fuzzy linguistic
variables and expressed as fuzzy linguistic labels such as Low (A1), Medium (A2) and High
(A3).The research methodology undertaken is summarized in Fig.2.

Data Collected from Laboratory

Data Resampling using Bootstrapping

Training Data Set Checking Data Set

 
ANFIS & Linear Regression Models ANFIS & Linear Regression Models

RMSE Evaluation

Model Comparison

Fig.2 Summary of the research methodology.

The ANFIS model under consideration is a multi-input single-output (MISO) system with
four inputs and one output. ANFIS and linear regression prediction accuracies are measured using
the Root Mean Square Error (RMSE). The ANFIS structure generated in this study utilizes fuzzy

76  
 

 
clustering of the input and output data sets as well as the bell-shape membership function. Thus the
number of rules is equal to the number of output clusters. In order to minimize the over fitting of the
model developed, the complete data set was split into a training (50%) and testing data set (50%).
The ANFIS model was first trained using the training data set followed by validation process using
the remaining data. The errors associated with the training and checking processes are recorded.
ANFIS training was found to converge after training with 95 epochs as shown in Fig.3. RMSE for
both the training and testing of ANFIS are very small which reflects the ability of ANFIS to capture
the essential components of underlying dynamics governing the relationships between the input and
the output variables. Fig.4 shows the architecture of 4-input one-output ANFIS structure. The
computation of membership functions (MFs) parameters is facilitated by a gradient descent vector.

Fig.3 ANFIS training converges after 95 epochs.

Fig.4 ANFIS architecture for a four input single-output Sugeno-fuzzy model.

ANFIS parameters are adjusted as to reduce the error measure defined by the sum of the
77  
 

 
squared difference between the actual and desired output. The root mean square error (RMSE) is
calculated using

Where At and Ft are actual and fitted values, respectively and N is the number of training or
testing sample. The parameters associated with MF’s will change through the learning process of
ANN. The output of the nth node is given in Eq. (2).
Layer 1: Every node in this layer i is an adaptive node with a node function

Where m and n are the inputs to node i and A1,….i are the linguistic labels such as average,
good, excellent associated with this node. O1Ai is the membership grade of fuzzy set A1,….i and it
denotes the degree to which the given inputs m or n satisfies the quantifier At. The membership
grade can be calculated using Eq. (3)

Where ai, bi, ci is the parameter set of a bell-shape figure. Parameters in this layer are
referred as premise parameters.
Layer 2: Every node in this layer is a fixed node and the output is the product of all the incoming
signals presented by Eq. (4).

Each node of output represents the firing strength of a rule.

Layer 3: Every node in this layer is fixed. The nodes in this layer normalizes the weight functions
by calculating the ratio of the ith rule’s firing strength to the sum of all rules’ firing strengths using
Eq.(5).

Layer 4: The nodes in this layer are adaptive nodes. The output of this layer are represented as

78  
 

 
 Where wi is a normalized firing strengths from Layer 3 and ( Pi, qi,… mi, ri ) are the
parameter sets referred as consequent parameters.
Layer 5: The single node in this layer labelled Σ computes the overall output. The output is
calculated
using Eq.(7).

Fuzzy reasoning which is made up of fuzzy if-then rules together with fuzzy membership
functions is the main feature of fuzzy inference systems (R. Jang, 1993). Fuzzy reasoning derives
conclusions from the set of rules which are either data driven or provided by experts (E. Neilsen,
1991). Fig.5 shows the reasoning procedure for a first order Sugeno fuzzy model. Each rule has a
crisp output and the overall output is a weighted average. For example;
“If Input-1 is High and Input-2 is Low and Input-3 is Medium and Input-4 is Low THEN the output
MF1 will be Medium is a complete rule defining the relations of input and output linguistic
variables”.

Fig.5 (a) Fuzzy reasoning procedure for Sugeno model of physical properties of degradable plastics
(b) If-Then rules derived by ANFIS.

The rule set given below illustrates the reasoning mechanism and the corresponding
equivalent ANFIS architecture where the nodes of the same layer have similar functions.

79  
 

 
5. Result and Discussion
Initially the training data set was used to develop ANFIS models with 2-input, 3-input and
4-input. The models were run for 800 epochs before the best models are identified based on the
smallest RMSE values. Next, the testing data set are fed into the trained ANFIS models. ANFIS
outputs are recorded and the error is calculated by comparing ANFIS predicted values with the
actual lab values. Similar training data sets are used to generate linear regression equations which
are then tested on similar testing data set as used for ANFIS. Linear regression outputs are recorded
and the error is also calculated by comparing linear regression predicted values with the actual lab
values. Table1 shows observed data, predicted data and percentage error in training ANFIS for
prediction of dielectric constants of various polymers.

Table.1. Observed data, Predicted data and Percentage Error in training ANFIS
Die electric Constants
Name of Polymer Observed Data Predicted Data Error %
Poly(tetrafluoroethylene) 2.00 1.87 6.50
Polyisobutylene 2.23 2.32 -4.00
Polyethylene 2.30 2.19 4.80
Polypropylene 2.30 2.41 -4.80
Polyisoprene 2.40 2.37 1.70
Polybutadiene 2.51 2.44 2.80
Polysiloxane 3.04 2.87 5.60
Poly(vinyl acetate) 3.50 3.39 3.10
Poly(methyl methacrylate) 3.60 3.06 15.0
Poly(oxymethylene) 3.70 2.54 31.4
Polyacrylonitrile 6.50 4.12 36.6
Poly(vinyl alcohol) 7.80 3.25 58.3
Poly (vinylidenefluoride) 8.40 3.04 63.8

All ANFIS and linear regression predicted outputs on the physical properties of polymer are
recorded and analyzed. Tables 2 tabulate the RMSE values for the prediction of Melting Point, Melt
Flow Index and Density of polymer. The best predictor set (Optimized) is determined based on the
smallest RMSE values. The results showed that ANFIS model prediction has very low RMSE
values which indicate high prediction accuracies in predicting the physical properties of polymer.
80  
 

 
 Other combinations could also be predicted using the developed ANFIS model to suit the
needs of the polymer industry. ANFIS prediction outputs are found to be compatible to linear
regression estimations.
These must be predicted extremely accurately for predicting miscibility. We have seen that
the ratio Tα / Tγ is extremely sensitive and important to predicting good mechanical properties. The
main advantage of this comparison is that it enables us to select ANFIS that train quickly and
provide acceptable results as compared to the best network we found. In Figure 6, points on the
straight line indicate that the actual and predicted output were identical. If most of the points lie on
this line, there is a danger of memorization and lack of generalization by the network. Similarly,
wide deviations from the straight line indicate a poorly trained ANFIS that may not give accurate
predictions.

Table 2: RMSE values of ANFIS models for the prediction of Melt Flow Index, Melting Point and
Density.
Input RMSE RMSE RMSE
Melt Flow Index Melting point Density
Train Test Train Test Train Test
Polyethylene 0.0050 0.0049 0.0991 0.1519 0.0056 0.0055

Polypropylene 0.0259 0.0239 0.3581 0.3359 0.0279 0.0266

Polyisoprene 0.0144 0.0157 0.5764 0.5545 0.0117 0.0123
Polybutadiene 0.0348 0.0342 0.2905 0.3109 0.0409 0.0385
Polysiloxane 0.0155 0.0187 0.4799 0.5120 0.0134 0.0139
Poly(vinyl acetate) 0.0197 0.0183 0.4754 0.5168 0.0177 0.0176
Poly(methyl 0.0030   0.0030   0.0820   0.1344   0.0043   0.0044  
methacrylate)
Poly(oxymethylene) 0.0030   0.0030   0.0820   0.1343   0.0043   0.0044  

Polyacrylonitrile 0.0030   0.0030   0.1089   0.1639   0.0043   0.0044  

Poly(vinyl alcohol) 0.0030   0.0030   0.0820   0.1344   0.0043   0.0044  

Poly(vinylidenefluoride) 0.0030   0.0030   0.0820   0.1344   0.0043   0.0044  

Predicted Properties
Table 2 gives the results when the final selected ANFIS model is used. We found that for
these modifications better Tα/Tγ ratios and dynamic mechanical modulus values were
81  
 

 
predicted as compared to the parent polymers. On the basis of these results, we were able to
conclude that both steric factors, and the intra- and intermolecular polarities of the polymer play a
vital role in the final outcome of the prediction of the mechanical properties of the polymers tested.

Fig. 6: Actual versus predicted output when final ANFIS model is applied to the 96 testing set.

Table 2: a) Results of applying the final model to 20 modified bisphenol-A polycarbonates. b)

Results of applying the final model to 15 modified poly (2, 6-dimethyl-1, 4-phenylene oxide)
Monomer Dynamic Modulus
(200C, dynes/cm2)
(a)
Modification PC-1 3.13 5.67x109
Modification PC-2 2.70 5.22x109
Modification PC-3 2.74 5.38x109
Modification PC-4 3.37 6.39x109
Modification PC-5 2.58 5.06x109
(b)
Modification PPO-1 1.98 6.32x109
Modification PPO-2 2.29 6.59x109

The dielectric constants of polymers

The dielectric constants of polymers used in this study were obtained from the literature

82  
 

 
(Brandrup and Immergut, 1975). Out of the available 13 conductive polymers, 12 were used for
training and one was left for testing or to validate the training. In the next run, a different set of 12
polymers was used for training, and the remaining polymer was used for testing. A total of 13 sets
of training were conducted in this manner and the results have been presented in Fig. 7.
However, when the dielectric constant data was large, the percent error was also getting
large. The reason for this, we feel, was the small number of compounds tested. If you don't provide
enough information to train the ANFIS, it won't learn properly.
Polymer Composites
Thermal conductivity of boron nitride reinforced high density polyethylene composites was
investigated under a special dispersion state of boron nitride particles in high density polyethylene,
and together with the influence on thermal conductivity of particle sizes of filler used by Zhou et al.
Xu et al. investigated the use of aluminum nitride (AN) and poly-vinylidene fluoride (PvF) as the
matrix. Gu et al. investigated the content of AN influencing the thermal conductivity and ultimate
mechanical properties of AN/ linear low-density polyethylene (LldP) composites.

Fig.7 Correlation between the actual values and predicted values by the ANFIS of dielectric
constant.

In this study, high-density polyethylene (HdP), low-density polyethylene (LdP), linear low-
density polyethylene (LldP), and polyvinylidene fluoride (PvF) with different metals/non-metals
such as boron nitride (BrN), copper (Cu) and aluminum nitride (AN) are used as inclusions, because
of its superior mechanical and physical properties.
Effective Thermal conductivity of boron nitride reinforced high density polyethylene composites

Fig.8 shows the variation in experimental values of effective thermal conductivity of boron
83  
 

 
nitride reinforced high density polyethylene composites and those predicted by the ANFIS, and
other theoretical models with volume fraction of dispersed phase (filler). It is seen that with the
increase in filler loading the ETC of the composite increases. The ETC of 1.129 W/m K is achieved
by ANFIS for HdP containing 29 % volume fraction of BrN, more than four times of pure HdP.

Effective Thermal Conductivity of Copper reinforced low-density polyethylene and linear low-
density polyethylene Composites
Fig.9 show the experimental values of effective thermal conductivity of LdP/copper
composites and those predicted by the ANFIS and other theoretical models over a wide range of
volume fraction of dispersed phase (filler) between 0% to 24%. It is clear that the effective thermal
conductivities.

Effective Thermal Conductivity of polyvinylidene fluoride with aluminum nitride Composite

The effective thermal conductivity of PvF/AN composites with volume fraction of dispersed
phase (filler) over the range between 0% to 75% is shown in Fig.10. It is noticed that the effective
thermal conductivity of the composite increases with the increase in filler loading, except that the
ETC decreased when the AN volume fraction is increased from 70% to 75% (due to increase in
porosity). The highest values of effective thermal conductivity 5.101 W/m K and 3.654 W/m K are
predicted by ANFIS for PvF containing 70% and 75% volume fraction of AN, respectively. It
84  
 

 
is also shown that the calculated results by the Singh et al. equations are in better agreement with
the experimental and ANFIS results.

Effective Thermal Conductivity of linear low-density polyethylene with aluminum nitride

Composite

Fig.11 shows the variation in experimental ETC of LldP/AN composites over a wide range
of volume fraction of dispersed phase (filler) between 0% to 32% and those predicted by the ANFIS
and calculated by various model with volume fraction of dispersed (filler) phase. It is clear that the
effective thermal conductivities of composites are higher than that of pure LldP matrix. The ETC of
composites increases considerably with the increase of volume fractions of inclusions. The results
are satisfactory in agreement with the experimental and ANFIS results.

In Fig.8-11, it is noticed that the ETC of different metal/non-metal filled polymer

composites increases with the increase in volume contents of filler in polymer composites. The
enhancement in the effective thermal conductivity of present composites with increase in volume
content of metal/non-metal is mainly due to more interaction between metal/non-metal particles as
they come in contact with each other, resulting in the ease in transfer of heat and consequent
enhancement of the effective thermal conductivity. Highly conductive different metal/ non-metal
like BrN, Cu, and AN are used as fillers into polyethylene (HdP, LdP, and LldP) and poly-

85  
 

 
vinylidene fluoride (PvF) composites as matrix in this study. All the predictions of the ETC by
ANFIS are in good agreement with the available experimental results and calculated by the Singh et
al. model. Clearly, there are many benefits of using ANFIS for prediction, including the following:
1) It is a general framework that combines two technologies, namely neural networks and fuzzy
systems; 2) By using fuzzy techniques, both numerical and linguistic knowledge can be combined
into a fuzzy rule base; 3) The combined fuzzy rule base represents the knowledge of the network
structure so that structure learning techniques can easily be accomplished; 4) Fuzzy membership
functions can be tuned optimally by using learning methods; 5) The architecture requirements are
fewer and simpler compared to neural networks, which require extensive trails and errors for
optimization of their architecture; and 6) ANFIS does not require extensive initializations through
several random starts before training, as always happens in neural networks. Other advantages of
the two-phase neuro fuzzy hybrid technique in the ANFIS model also include its nonlinear ability,
its capacity for fast learning from numerical and linguistic knowledge, and its adaptation capability.

4. Conclusions
In this paper we had described the development of a data driven ANFIS model using real
data set obtained from the polymer laboratory. The developed ANFIS is a soft computing approach
utilizing a feed-forward multilayer neural network for fuzzy modeling. This study had shown that
ANFIS models are highly robust and compatible. ANFIS models are found to have good prediction
ability for the prediction of physical properties of polymer is recommended. It is noticed that the
ETC of different metal/non-metal filled polymer composites increases with the increase in volume
contents of filler in polymer composites. The enhancement in the effective thermal conductivity of
present composites with increase in volume content of metal/non-metal is mainly due to more
interaction between metal/non-metal particles as they come in contact with each other, resulting in
the ease in transfer of heat and consequent enhancement of the effective thermal conductivity.
Highly conductive different metal/ non-metal like BrN, Cu, and AN are used as fillers into
polyethylene (HdP, LdP, and LldP) and poly-vinylidene fluoride (PvF) composites as matrix in this
study. All the predictions of the ETC by ANFIS are in good agreement with the available
experimental re-sults and calculated by the Singh et al. model. Max-well as well as Hamilton and
Crosser models are calculated fairly well the ETC only for low concentration of present composites.
The predicted results show that using a hybrid intelligent approach, in particular ANFIS, gives good
prediction ac-curacies for the ETC of metal/non-metal filled polymer composites. The resultant
predictions of effective thermal conductivity by the ANFIS agree well with the available
experimental data. The ANFIS exhibit the capability to use for the predictions of effective
86  
 

 
thermal conductivity of various types of tailored complex materials.
References
1. G. Seyhan, G. Tayfur, M. Karakurt & M. Tanoglu, (2005) “Artificial neural network (ANN)
prediction of compressive strength of VARTM processed polymer composites”.
Computational Materials Science, No. 34, pp. 99-105.
2. R. Jang, (1993) “ANFIS: adaptive-Network-Based Fuzzy Inference System”, IEEE trans. On
Systems, Man and Cybernetics, Vol. 23, No. 3, pp. 665-685.
3. N. K. Roy, D. P. Landau, and W. D. Potter, (2004) “Designing Polymer Blends Using Neural
Networks, Genetic Algorithms and Markov Chains,” Applied Intelligence, 20(3), pp. 215-
229.
4. D.W. Van Krevelen and P. J. (1976) “Hoftyzer, Properties of Polymers: Their Estimation
and Correlation with Chemical Structure”, Elsevier Science Publishers, Amsterdam.
5. J.M. Schultz, (1974) “Polymer Materials Science”, Prentice Hall, Englewood Cliffs, New
Jersey.
6. I. E. Neilsen, (1991) “Mechanical Properties of Polymers”, Reinhold, New York.
7. T. Murayama, (1978) “Dynamic Mechanical Analysis of Polymeric Materials”, Elsevier,
New York.
8. K. Friedrich, R. Reinicke and Z. Zhang (2002) “Wear of polymer composites. In: Journal
Engineering Tribology”, Proceedings of the Institution of Mechanical Engineers, Vol. 216
part J: Journal of Engineering Tribology, Stachowiak GW, editor. The Professional
Engineering Publishing Ltd; pp. 415–26.
9. Y. Al-Assaf and H. El Kadi (2001) “Fatigue life prediction of unidirectional glass
fiber/epoxy composite laminate using neural networks”. Composite Structures; 53(1):65–71.
10. H. El Kadi and Y. Al-Assaf (2002) “Prediction of the fatigue life of unidirectional glass
fiber/epoxy composite laminate using different neural network paradigms”. Composite
Structures; 55:239–46.
11. K. Velten, R. Reinicke and K. Friedrich (2000) “Wear volume prediction with artificial
neural networks”. Tribology International; 33:731–6.
12. Z. Zhang and P. K. Klein, (2002) “Friedrich Dynamic mechanical properties of PTFE based
short carbon fiber reinforced composites: experiment and artificial neural network
prediction”. Composites Science and Technology; 62(7–8):1001–9.
13. J. Gu, Q. Zhang, J. Dang, J. Zhang and Z. Yang, (2009) “Thermal Conductivity and
Mechanical Properties of Aluminum Nitride Filled Linear Low-Density Polyethylene
87  
 

 
 Composites,” Polymer Engineering & Science, Vol.49, No.5, pp.1030-1034.
doi:10.1002/pen.21336
14. R. Singh and P. Sharma, (2008) “Effective Thermal Conductivity of Polymer Composites,”
Advanced Engineering Materials, Vol. 10, No. 4, pp. 366-370. doi:10.1002/adem.200700336
15. R. Singh, P.K. Sharma and R.S. Bhoopal, (2011) “Prediction of Effective Thermal
Conductivity of Cellular and Polymer Composites,” Indian Journal of Pure Applied Physics,
Vol. 49, No. 5, pp. 344-349.
16. R.S. Bhoopal, P.K. Sharma, S. Kumar, A. Pandey, R.S. Beniwal and R. Singh, (2012)
“Prediction of Effective Thermal Conductivity of Polymer Composites Using Artificial
Neural Network Approach,” Special Topics & Review in Porous Media—An International
Journal, Vol. 3, No. 2, pp. 115-123.  
 

-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐-­‐  

88