j m a t e r r e s t e c h n o l .

2 0 2 0;9(6):14036–14050

Available online at www.sciencedirect.com

https://www.journals.elsevier.com/journal-of-materials-research-and-technology

Original Article

Modelling and optimisation of hardness behaviour

of sintered Al/SiC composites using RSM and ANN:
a comparative study

Mohammad Azad Alam a,∗ , Hamdan H. Ya a , Mohammad Azeem a , Patthi Bin Hussain a ,
Mohd Sapuan bin Salit b , Rehan Khan a , Sajjad Arif c , Akhter Husain Ansari c
a Mechanical Engineering Department, Universiti Teknologi PETRONAS, Seri Iskandar 32610, Perak, Malaysia
b Laboratory of Biocomposite Technology, Institute of Tropical Forestry and Forest Products, Universiti Putra Malaysia, 43400 UPM
Serdang, Selangor, Malaysia
c Department of Mechanical Engineering, Aligarh Muslim University, Aligarh 202002, India

a r t i c l e i n f o a b s t r a c t

Article history: In present work, Aluminium matrix composites reinforced with x wt.% SiC (x = 5, 7.5 and
Received 13 July 2020 10) microparticles were synthesised by powder metallurgy route. The microhardness (VHN)
Accepted 24 September 2020 of the Al/SiC composites were investigated using Response Surface Methodology (RSM)
Available online 9 October 2020 and artificial neural network (ANN) approach. Scanning electron microscopy (SEM), Energy-
dispersive X-ray spectroscopy (EDS), Elemental mapping and Optical microscopy were done
Keywords: for the microstructural investigations. The X-ray diffraction (XRD) analysis was done for
Al–SiC composites received powders and composites samples for phase recognition and existence of rein-
Powder metallurgy forcement particles (SiC) in the synthesised sintered composites. The design of experiments
Vickers Microhardness based on RSM was utilised following the central composite design method. Empirical models
Response surface methodology have been developed by considering variance analysis (ANOVA), to establish relationships
Artificial neural networks among the control factors and the response variables. A feed-forward back-propagation neu-
ral network (FF-BPNN) was used to determine the qualitative characteristics of the process,
and the accuracy of the BPNN system was attributed with mathematical models based on
RSM model. The ANN model predicted surface hardness values are near the experimental
findings. It is established that the developed models can be used to predict the hardness of
the surface within the investigation range. The composite with reinforcement 7.5% revealed
higher sintered density and Vickers microhardness due to the uniform distribution of filler
particles in the Al matrix featuring no pores. The results indicate overall higher accuracy in
the ANN method than RSM model.
© 2020 The Author(s). Published by Elsevier B.V. This is an open access article under the
CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Abbreviations: ANN, artificial neural network; RSM, response surface methodology; R2, determination coefficient; ANOVA, Analysis of
Variance; FFBPN, Feed-Forward Backpropagation Network; RMSE, Root Mean Square Error; MAPE, Mean Absolute Percentage Error; XRD,
X-ray diffraction; AMCs, Aluminium matrix composites.
∗
Corresponding author.
E-mail: azadalam.mech3@gmail.com (M.A. Alam).
https://doi.org/10.1016/j.jmrt.2020.09.087
2238-7854/© 2020 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://
creativecommons.org/licenses/by-nc-nd/4.0/).
 j m a t e r r e s t e c h n o l . 2 0 2 0;9(6):14036–14050 14037

and animal). These simulations are the gross simulation ele-

1. Introduction ment by element and neuron by neuron [17]. The application
of artificial intelligence in many fields of engineering has been
The Present demand for structural materials in automotive
enormous in recent years [18–20]. The ANN was employed for
and aircraft industries is to be lightweight, high strength and
better prediction of the response behaviours of newly devel-
hardness. Aluminium is a lightweight metal, but not appropri-
oped composites by various researchers in their investigations
ate for many structural components of automotive and aircraft
[21–23]. ANN offers a complicated correlation between vari-
structure due to its lower strength and hardness [1,2]. Research
ables of input and output, that could be used to substitute
and development work in the Aluminium-based composite
modelling methods for polynomial regression. The artificial
materials are currently underway to enhance the mechani-
neural network performs a significant role in understand-
cal, physical, other properties of monolithic Aluminium. Due
ing the linear and nonlinear problems in various engineering
to their low density, superior corrosion resistance, high elec-
fields. Kumar et al. applied the ANN technique to validate the
trical and thermal conductivity, Aluminium and its alloys
experimental results for surface roughness of the developed
are predominantly used as matrix material in the composite
composites [24].
development [3,4]. Studies in this field are encouraged due to
Simultaneous mathematical model and learning method
the requirements of lightweight and higher mechanical prop-
advantages would allow researchers to make comparison and
erties materials, particularly in automobile, aircraft structure,
a better understanding of the findings. Because of these major
offshore superstructure and military planes [5,6]. Compos-
benefits, both RSM and ANN is adopted in the present research
ites made of Aluminium strengthened with ceramic particles
to optimise and predict the control factors and response
such as silicon carbide, boron carbide, titanium carbide, alu-
micro-hardness, respectively. To state of the art, research on
mina etc. exhibited higher strength, rigidity, wear-resistance
interaction effects of process parameter for the microhardness
and high hardness. Several researchers investigated that the
of the Al/SiC composites by P/M route, their forecasting by RSM
addition of SiC micron or Nanoparticles in the Aluminium
and ANN, the comparison among two techniques has rarely
matrix enhances the interfacial, physical and mechanical
been reported anywhere else for the same composition/wt.%
properties through homogeneous distribution in the matrix
of reinforcements.
[4,7–9].
The current research focuses on Investigating the effect
AMCs are usually produced using Stir Casting methods
of reinforcement content (wt.%) and process parameter (sin-
and Powder Metallurgy techniques. The manufacturing pro-
tering temperature) on the microhardness of the Al-x wt.%
cess used for the synthesis of composites plays a significant
SiC (x = 5, 7.5 and 10 wt.%) composites. To develop empiri-
role in achieving a perfect distribution of particles and mini-
cal models by considering a detailed parameters correlation,
mal porosity defects. It can be achieved through the process of
evaluating interactions between independent variables, opti-
powder metallurgy owing to the major benefits of producing
mising them and providing the overall impact of variables
near net form components and preventing interfacial interac-
on the hardness of the synthesised composites utilising RSM.
tions as seen in the production of liquid melt [2,10,11].
Furthermore, ANN predictive model is established to identify
Optimisation of the process parameters is the major chal-
the impact of different control variables on the microhard-
lenge in the experimental work to save the men, materials
ness of the synthesised composites and compared with the
and money as well as to get an improved response. So, the lat-
RSM model. For comparison of the accuracy of responses pre-
est and advanced statistical tools must be adopted within the
diction another modelling Method, artificial neural networks
investigation boundary conditions. Response Surface Method
(ANN) was employed. ANN offers a complicated correlation
(RSM) is a highly advanced DOE technique, includes statistical
between variables of input and output, that could be used to
formulation for developing a model and analysing a process
substitute modelling methods for polynomial regression. To
which aims to optimise the desired response controlled by
predict the response, MATLAB ® 2019b has been used for the
multiple input parameters [12,13]. Using comparatively small
development of the ANN model based on the Feed-Forward
number of experiments, a response surface model can be used
Backpropagation Network (FFBPN).
to map a design space [14]. The Contour and surface graphs
are often used to explain both linear and nonlinear mixing
complications of mixed components. In RSM technique, cen- 2. Experimental methods
tral composite design (CCD) offers more understanding than
that of the three factorial designs, needs less experimen- 2.1. Raw materials
tal runs despite showing significant optimisation for most
of the stable processes. The central composite design is the Pure Aluminium of 10–12 ␮m average particle size, was utilised
most widely used response surface and includes a factorial as the matrix metal. The average particles size of selected
design or factorial fractional design with centre points, sup- reinforcement SiC was 37 ␮m with a purity of more than 95%.
ported by a group of axial points[15,16]. S. Sardar et al. utilises Aluminium and SiC were supplied by Otto chemicals, India. As
the RSM tool for modelling the responses wear resistance a process control agent (PCA), Toluene (C6 H5 CH3 ), supplied by
and surface roughness and also optimises the input variables Ranbaxy, India was used. For investigation purpose, a homo-
[16]. geneous mixture of matrix and reinforcements were prepared
The artificial neural network (ANN) is the computational in three combinations, (Al + 5% SiC, Al + 7.5% SiC and Al + 10%
network that attempts to artificially simulate the nerve cell SiC) respectively.
networks (neurons) of the biological nervous system (human
14038 j m a t e r r e s t e c h n o l . 2 0 2 0;9(6):14036–14050

2.2. Blending of powders
The various weight percentages (5, 7.5 and 10.0 wt.%) of SiC
reinforced Al composite powders were prepared using a cen-
trifugal ball mill made by FRITSCH, Germany. The speed of
rotation and the ball to powder ratio (BPR) were 150 rpm and
10:1, respectively.
2.3. Compaction and controlled environment sintering
The blended powders were compacted into 8 mm diameter,
and 12 mm long cylindrical pellets utilising uniaxial com-
paction dies in a hydraulic pallet press (Kimaya Engineers,
India), at a compaction pressure of 450 Mpa. Zinc-stearate was
applied over the die wall before each compaction for reducing
the frictional effects at the die wall. Under a controlled envi-
ronment, the green pellets had been sintered in an electric
tubular furnace with flow regulator at temperatures of 500, 550
and 600 ◦ C for 2 h. Sintering was accomplished under a steady
supply of argon gas at a flow rate = 1 litre/min. To avoid oxida-
tion of the synthesised composites [25]. The sintered pellets
were cooled down to ambient temperature within the furnace.
2.4. Characterisation
2.4.1. X-ray diffraction
XRD of received powders and synthesised composites was
done utilising X-ray Diffractometer (D8 ADVANCE, Bruker AXS
Inc., USA) to confirm the existence of reinforcement parti-
cles (SiC). The diffractometer had been monitored employing
CuKa radiation (k = 1.5406 Å) with 40 kV operational voltage
and 40 mA current. XRD data was gathered at a sluggish scan-
ning rate of 0.02 steps/second for meticulous estimation of
existing phases. The scan range was 20◦ –80◦ in 0.01◦ steps.
The comparative analysis of XRD spectra was too executed by
observing the patterns of received Al, reinforcement SiC and
synthesised composites. Utilising EDS elemental mapping, the
degree of uniformity of composites and the dispersion of filler
material was observed.
2.4.2. Optical microscopy
The microstructure of all the sintered composite samples was
observed through an optical microscope (Model: Leica DM LM,
Wetzlar, Germany) following the preparation of the metallog-
raphy. It was used to investigate the impact of reinforcement
particles on the microstructural evolution of the composites
system. The polished and etched sections of the sintered
composites were exposed for the optical micrographic obser-
vations, homogenisation of the grain size is observed after
addition of SiC particles.
2.5. Density measurement
For all the specimens of Al and composites, the experimental
sintered densities have been measured using the principle of
Archimedes’ by standard test method (ASTM B0311-93R02E01).
For measurement of the density of the composites, HR-250
AZ analytical balance developed by A&D Company, Limited,
Tokyo, Japan was used. The figure of the analytical balance
used is given in the supplementary data file (Fig. 4). Four sam-
ples were measured for each formulation, and the mean value
of the sintered density has been reported.
2.6. Microhardness measurement
The microhardness of all the sintered composites samples was
measured according to the standard test method (ASTM E92-
82) by Vickers hardness tester at a load of 300 gf keeping dwell
time 25s. In the hardness test, the possible experimental error
was below 5%. The test was conducted at room temperature

Table 1 – Input variables and DOE levels limits.

Factor Name Units Type Minimum Maximum

A SiC Concentration wt.% Numeric 5.00 10.00

◦
B Sintering Temperature C Numeric 500.00 600.00

Table 2 – Matrix of design (central composite design).

Std Run Factor 1 A: SiC Factor 2 B: Response 1 Response 2
Concentration Sintering Sintered density Microhardness
(Wt%) Temperature (◦ C) (g/cm3) (VHN)

1 1 5 500 2.4 42.2

6 2 10 550 2.46 50.3
10 3 7.5 550 2.54 55.4
5 4 5 550 2.48 44.9
13 5 7.5 550 2.53 59.68
2 6 10 500 2.47 53.3
3 7 5 600 2.45 43.8
9 8 7.5 550 2.56 52.43
11 9 7.5 550 2.55 58.3
12 10 7.5 550 2.51 56.2
8 11 7.5 600 2.57 54.92
4 12 10 600 2.57 52.1
7 13 7.5 500 2.49 57.26
 j m a t e r r e s t e c h n o l . 2 0 2 0;9(6):14036–14050 14039

Fig. 1 – SEM micrograph of received powder (a) Aluminium particles, (b) SiC particles.

Fig. 2 – EDS spectrum of powder samples (a) Al, and (b) SiC.

(25 ◦ C), hardness measurement was noted four times for each
test sample at different locations, and the mean values were
taken into consideration for microhardness.
2.7. Experimental design
The central composite design (CCD) approach was used for the
model development in RSM to examine both the individual
and interaction effect on hardness of the synthesised com-
posites under multivariable (SiC concentrations and sintering
temperatures) conditions. The objective of applying CCD to
the variables used in the present work is that the 2nd order
model can be developed efficiently. To enable the estimate
of control parameters of a second-order model, CCD designs
are strengthened by the extra centre and axial points. The
input parameters/factors were SiC concentrations and sinter-
ing temperature, Table 1 shows the process parameters with
lower and higher levels.
The output parameters/response variables were density
and microhardness, as shown in Table 2. The descriptions
of the design matrix of the factors in the particular units
used in the surface response system, together with the hard-
ness and density responses were observed and depicted in
Table 2.
14040 j m a t e r r e s t e c h n o l . 2 0 2 0;9(6):14036–14050

Fig. 3 – XRD patterns of (a) Al powder, (b) SiC powder, depicting various phase peaks.

3. Results and discussions

3.1. Characterisations of received powders

The microstructures of received Al and SiC powders are

depicted in SEM micrographs Fig. 1. It is observed from SEM
micrograph that, Al powders have spherical and droplet like
morphology, whereas SiC powders have an ellipsoidal shape
with a sharp edge, tip and points.
The XRD pattern of received Al powder is shown in Fig. 3(a).
Al (111), Al (200), Al (220) and Al (311) crystal phases are the
major constituent components of Al powder and their peaks
were observed at a diffraction angle (2) = ∼38◦ , 45◦ , 65◦ and
78◦ respectively. In comparison, the SiC powder XRD pat-
tern as depicted in Fig. 3(b). The phases ␤-SiC (111) and ␤-
SiC (220) peaks were at a diffraction angle (2) = ∼36◦ and 60◦
respectively. These pure powder XRD patterns have a vital role
in examining the phase structure of the Al/SiC composites.
These XRD peaks pattern can be utilised to validate the Al-SiC
composite production with the existence of related Al and SiC
crystal phases.
The results of EDS for Al powder and SiC powder are repre-
Fig. 4 – Analytical balance used for recording the sintered
sented in Fig. 2. Al spectrum Fig. 2(a), contains a few amounts
density of the composites.
of oxygen owing to oxidation at the surface. This oxide layer
guards the surface against wear and corrosion. The SiC spec-
trum Fig. 2(b) shows the individual peaks of Si and C elements results, no new peak existed, suggesting that no new phase
indicating each element are present in the sample. was developed throughout the sintering process. Thus, it can
be noted that the composites samples produced are strength-
3.2. Characterisation of synthesised Al/SiC composites ened with SiC. The calculations have been made by using
DIFFRAC® plus software (Bruker AXS Inc.).
3.2.1. XRD of Al/SiC AMCs
The X-ray diffractometer analysis was done for phase recog- 3.2.2. Microstructures of Al/SiC AMCs
nition and existence of reinforcement particles (SiC) in the The microstructure of the Aluminium and composites is
synthesised sintered composites. The XRD plot of developed expressed in optical microscopy images Fig. 6. Optical
composites with different SiC concentrations is presented in microscopy of the polished and etched surface of the com-
Fig. 5. The existence of Aluminium and SiC was verified from posites sintered pellet show that the reinforcing the SiC
their corresponding peaks of the XRD graph. The highest peak has resulted in grain size homogeneity. Introduction of SiC
was observed for Aluminium followed by SiC. assisted in the homogenous distribution of grain size because
It was noticed that the peak intensity referring to the SiC of the pinning action of SiC at the grain boundaries, limit-
phase had begun to increase with the weight per cent of SiC ing grain growth while sintering [28]. The refined grains of Al
particles in the AMCs. The results observed are in line with matrix is observed in synthesised composites (Fig. 6b). The
the findings of the previous researchers [9,25–27]. In all XRD consistent grain boundaries, as shown in Fig. 6(b) indicates
 j m a t e r r e s t e c h n o l . 2 0 2 0;9(6):14036–14050
Fig. 5 – XRD peaks pattern of Al-x wt.% SiC (x = 0, 5, 7.5 and
10) composites.
that SiC did not provide an obstacle in the sintering operation
owing to the successful dispersion inside the Al matrix. The
agglomerations of the SiC particles are observed in Fig. 6(b) and
also in microhardness indentation Image 8(b). Fig. 6(c) depicts
the SiC particles distribution within the Aluminium matrix.
The distribution of SiC particles in the Al matrix was observed,
no agglomerations of reinforcements are seen hence leads to
the highest increment in microhardness of the composites.
The microhardness indentation image is shown in Fig. 8(c)
also reveal the uniform distribution of the SiC particles in the
Al matrix. The uniform distribution of carbide reinforcement
plays a pivotal role in the enhancement of the microhardness.
The uniformity in the grain size is quite important as it pro-
vides the homogeneity in the mechanical properties across
the entire material’s structure [29].
Fig. 6 – Optical microscopy observations of (a) Aluminium, (b) Al + 5% SiC, (c) Al+ 7.5% SiC and (d) Al + 10%SiC.
14041
The grain boundaries and refined grains are visible in
Fig. 6(d). The clustering of SiC particles is observed in the vicin-
ity of the grains as a black regime alongside the grain boundary
for 10 wt.% SiC compositions, as shown in Fig. 6(d) and in hard-
ness indentation photo Fig. 8(d). The non-uniform grains size
was observed, and it leads to decrement in the hardness of the
composites.
Fig. 7 depicts the elemental mapping of sintered Al –
7.5 wt.% composites synthesised through cold compaction.
SiC particles distribution in Al matrix is observed as homo-
geneous, discrete and random. The existence of elements
Aluminium(Al), Silicon (Si), Carbon (C) and Oxygen (O) was
confirmed by elemental mapping of the composites, as
shown in Fig. 7. The oxygen element content is owing
to Al-oxide layers formed in mixing and sintering pro-
cess.
3.2.3. Microhardness investigations
The nonhomogeneous grain size dispersion is observed, and
the average size of the grain seems identical to the grains of
Al. The prediction of the overall performance of the developed
composites is necessary to analyse its bulk properties, includ-
ing surface properties. Thus, Vickers microhardness test is
performed to analyse the bulk properties of the synthesised
composites. For each specimen, the microhardness of the
composites is measured four times, and the mean results are
presented in Fig. 9.
The increment in microhardness for all the sintered com-
posites was observed due to the presence of SiC reinforcement
as compared to the Al sample. Such results are consistent with
previous studies finding [30,31]. It is observed from Fig. 9 that
the maximum increment in microhardness of synthesised
composites is 60% at 7.5 wt.% SiC as compared to received
Aluminium. This increment in microhardness is attributed to
14042 j m a t e r r e s t e c h n o l . 2 0 2 0;9(6):14036–14050

Fig. 7 – Elemental maps of Al + 7.5% SiC Aluminium matrix composites.

Fig. 8 – Microhardness indentation images of base Al and composites formulations.

 j m a t e r r e s t e c h n o l . 2 0 2 0;9(6):14036–14050
Fig. 9 – Influence of weight percentage of SiC on
microhardness of composites.
a uniform distribution of SiC reinforcement in the Al matrix
as shown in microstructure image Fig. 6(c) and indentation
image Fig. 8(c). The increment in the hardness of Al-SiC
composite compared with base Aluminium is attributed to
multiple phenomena such as: (i) Uniformly distributed SiC
particles in the matrix, (ii) Obstructions by SiC particles in
dislocation movements (Orowanian strengthening), (iii) Grain
refinement of the matrix (Hall–Petch strengthening), Hard-
ness because of the harder SiC particles [13]. However, the
hardness of composites declines with the additional inclu-
sion of 10 wt.% SiC microparticles but having higher hardness
value than the neat Al and Al/5 wt.% SiC composites. The
reason for decrement is attributed to the clustering of par-
ticles [25]. The clustering of SiC particles is observed in
Fig. 8(d).
Fig. 10 – 3D contour plot of SiC concentration and sintering temperature on hardness response.
14043
3.3. RSM modelling
The RSM models for sintered density and hardness were devel-
oped by utilising experimental results using design expert
software version 12. A complete quadratic model comprising
of the terms linear, square and interaction were developed.
The equation of the developed model for comparing the
responses (Hardness and sintered density) to input variables
A and B using design expert software can be expressed as
follows:
Let A = Reinforcement (SiC) concentrations (wt.%),
B = Sintering temperature (◦ C) as mentioned in Table 1.
Micro hardness = +56.32 + 56.32A − 0.3233B − 0.7000AB
− 8.53A2 − 0.0407B2 (1)
Sintered density = 2.55 + 0.0367A + 0.0233B − 0.0025AB
− 0.0841A2 + 0.0059B2 (2)
Table 3 depicts the ANOVA table for hardness which is used
to derive the Eq. (1). As demonstrated, the F-value of the model
is 12.03 indicates the developed model is significant. There is
only a 0.25% chance that an F-value this large could occur due
to noise. The Lack of Fit F-value of 0.36 implies the Lack of Fit
is not significant relative to the pure error. There is a 78.31%
chance that a Lack of Fit F-value; this large could occur due to
noise. Non-significant lack of fit is good as we want the model
to fit. These results are similar to previous research [32].
The experimental data gathered from central composites
design, shown in Table 4, has been analysed and interpreted by
the Response Methodology (RSM). Design expert software (ver-
sion12) was used to conduct the analysis part of the present
work.
14044 j m a t e r r e s t e c h n o l . 2 0 2 0;9(6):14036–14050

Table 3 – ANOVA table for hardness model.

Source Sum of squares df Mean squares F-value p-Value

Model 341.07 5 68.21 12.03 0.0025 Significant

A-SiC concentration 102.51 1 102.51 18.07 0.0038
B-Sintering temperature 0.6273 1 0.6273 0.1106 0.7492
AB 1.96 1 1.96 0.3456 0.5751
A2 200.99 1 200.99 35.44 0.0006
B2 0.0046 1 0.0046 0.0008 0.9781
Residual 39.70 7 5.67
Lack of fit 8.53 3 2.84 0.3650 0.7831 Not significant
Pure error 31.17 4 7.79
Cor total 380.77 12

Table 4 – Response (Hardness) model summary.

Source Std. Dev. R2 Adjusted R2 Predicted R2 Press

Linear 5.27 0.2709 0.1250 −0.3605 518.05

2FI 5.53 0.2760 0.0347 −1.6293 1001.16
Quadratic 2.38 0.8957 0.8213 0.6543 131.64 Suggested
Cubic 2.51 0.9176 0.8022 0.8140 70.82 Aliased

Bold values indicates the most feasible model suggested by RSM tool i.e Qudratic out of four models mentioned in the table 4.

Fig. 11 – Microhardness plot of actual VS predicted.

The Hardness Response model summary is shown in
Table 4. Adequacy measures R2 , adjusted R2 and predicted R2
are shown in Table 4. All indicators of adequacy are in reason-
able agreement and demonstrate a considerable relationship.
The Predicted R2 of 0.6543 is in reasonable agreement with the
Adjusted R2 of 0.8213 since the difference is less than 0.2. The
analysis of the variance outcome for the hardness model indi-
cates that significant model terms are the main effect of the
two process parameters (concentration of reinforcement and
sauntering temperature) along with the interaction effect of
the two parameters [24].
3.3.1. Hardness effect analysis on process variables
For the conciseness, it is decided to explain the effect of pro-
cess variables on the hardness only. Figs. 10–14 demonstrates
Fig. 12 – Perturbation plot depicting the effects of all
variables simultaneously with a central reference point of
0.00 on the hardness.
the relations between the hardness response and the process
variable. In the experimental study ranges, each plot displays
the influence of two process parameters against the other
parameter set at the central point values. The surfaces of the
response better represent the ability of each factor to affect
the hardness. It is observed from Fig. 10 (3D plot of response
surface), reinforcement (SiC) concentration is found to have
the most powerful effect on the hardness of the composites,
preceded by the sintering temperature.
The graph of experimental outputs and predicted output
is depicted in Fig. 11, it can be observed that the correlation
between the actual value and predicted by the model of the
 j m a t e r r e s t e c h n o l . 2 0 2 0;9(6):14036–14050 14045

the design space. Fig. 12 shows the plot of perturbation of

the two variables on the hardness response. The response
is illustrated by altering only a single factor across its range
while maintaining another factor constant. The influence of
variable A (SiC content) was observed for the range between
5–10 wt.%.
It can be observed the hardness increases initially and then
decreases at 7.5 wt.% onwards. It is observed from Fig. 12 that
the parameter B (sintering temperature) has less influence on
the response as compared to parameter A.
Fig. 13 shows the plot of interaction, that reveals that
there is an interaction between the two variables on the
response hardness, which is also manifested in the mathe-
matical model of hardness.

Fig. 13 – SiC Concentration and sintering temperature

interaction effect on hardness response.
Fig. 14 – Desirability plot of micro-hardness at optimum
process variables.
response is up to the satisfactory level. The model does not
reveal any abrupt variation of the continual variance [18].
The real data points in the graph are in near agreement
with the predicted ones suggesting that the quadratic model
is an effective model for estimating the independent variable’s
response [15].
Perturbation graph is an essential diagram depiction to
analyse the influence of all variables at a given location in
3.3.2. Optimisations analysis of process variables
The optimum conditions for the process parameter are pre-
dicted by using the numerical optimisation component of the
design expert software. The various input constraints goals
of process variables are chosen to optimise the response, as
presented in Table 5.
The optimum conditions of input variables were obtained
through solving the prediction equation of quadratic model
obtained from experimental findings and surface plot analysis
of the response. The set of input parameters and the response
with the achieved optimum are presented in Table 6, where
the respective desirability was 0.68368, and 0.63921 for the two
suggested optimum solutions. The desirability ranges from 0
to 1, depending on the proximity of the response to the target
[33].
The RSM provides the optimised input parameter sets sug-
gested to get maximum sintered density and microhardness
in the course of development of the composites as depicted in
the above Fig. 14. The maximum microhardness 56.824 VHN
and density 2.539 g/cm3 can be achieved at a sintering tem-
perature of 550 ◦ C and SiC content of 8.09%. The experimental
findings were further analysed through ANN for critical review
and being compared with RSM.
3.3.3. Experimental verification on optimum conditions of
process variables
For the confirmation of the accuracy of the optimised
condition produced by response surface methodology, four

Table 5 – Constraints to optimise the response.

Name Goal Lower limit Upper limit Lower Weight Upper Weight Importance

A: SiC Conc. Minimise 5 10 1 1 3

B: Sintering, Temp. target = 550 545 555 1 1 3
Sintered density maximise 2.4 2.71 1 1 3
Microhardness maximise 42.2 60 1 1 3

Table 6 – Numerical optimisation and composite desirability values.

No of solutions SiC(%) Sintering temp (◦ C) Sintered density Microhardness (VHN) Desirability

1 8.09 550.000 2.539 56.824 0.68368

2 7.296 550.000 2.533 55.931 0.63921

The bold values indicates the best possible solution with highest value of hardness and desirability among the two solutions.
14046 j m a t e r r e s t e c h n o l . 2 0 2 0;9(6):14036–14050

Table 7 – Experimental verificaton under optimum conditions.

Run Sintered density (g/cm3 ) Microhardness (VHN)

Experimental Predicted % Error Experimental Predicted % Error

1 2.476 2.539 2.48 55.492 56.824 2.34

2 2.513 2.539 1.02 56.137 56.824 1.20
3 2.489 2.539 1.96 55.246 56.824 2.77
4 2.508 2.539 1.22 56.102 56.824 1.27
Mean 1.67 1.89
Standard deviation 0.674 0.782

Table 8 – Artificial neural network operation parameters.

Name Type

Network configuration 2-10-2

Number of hidden layers 1
Number of hidden neurons 10
Transfer function used Logsig (sigmoid)
Number of patterns used for training 9
Number of patterns used for testing and 4
validation
Number of epochs 1000

3.4.1. Training network and testing

Fig. 15 – Neural Network Configurations (2-10-2 topology)
selected as the prediction model for the density and
hardness.
additional experiments were carried out under the same
conditions as those given by solution1. Table 7 displays
the outcomes of repeated experiments, and the predicted
response results in solution 1 are analogous to the experimen-
tal results.

2 Table 8.
(X − mean)
Standard deviation(s) =
(n − 1)
The mean % error of sintered density and microhardness
is 1.67 and 1.89, respectively. These findings demonstrate the
comparable values for the responses produced from exper-
imentation and predicted from RSM tool of Design-Expert
software. Thus, the precision of the optimal conditions pro-
duced by the software is verified experimentally.
3.4. ANN (artificial neural networks) Model
development
ANN consists primarily of three major strata: the input, the
hidden and the output layers. The input layer comprises of
entry nodes that collect the user’s data and transmit it through
a hidden layer to the output layer. The number of hidden
layers may be varied according to the scope of the available
data [34,35]. Three-layer architecture is chosen in the currently
established model, as illustrated in Fig. 15. One input layer
having SiC concentration, sintering temperature as the entry
nodes, one hidden layer with ten hidden nodes and one out-
put layer with two output nodes, namely sintered density and
hardness.
ANN model development consists of two-step viz., training or
learning step, and recalling or testing. The network changes
its weights in the learning process to reduce the discrep-
ancy between projected outputs and targeted goals [36]. To
train the network, an ANN model of feedback propagation
was used. The network learning and testing were performed
using the software package MATLAB 2019b. Microhardness
and density data were used to develop the ANN model.
In this research work modelling, the Levenberg–Marquardt
(LM) algorithm was used because of the reliability and
processing time [37]. In references to this research, the param-
eters for artificial neural network operation are presented in
Fig. 16 illustrates the comparison made between the
training, validation, testing and combined collection of exper-
imental and projected data. The overall performance curve
showed a strong correlation between experimental and
expected results due to model training accuracy.
Therefore, the trained model offered minimal error to the
prediction, and it may also be used to project the unknown
value for the upcoming set of data. The network has been suc-
cessfully trained with a near-unit determination coefficient
(R = 0.99918), as depicted in Fig. 16. The coefficient of regression
(R), which represents the relationship between the output and
the goal, having the overall value of 0.99934, which is closer to
1 and it implies that the performance is better. The ANN pre-
dicted values are closer to the experimental values implying a
small error difference. The developed model can, therefore, be
used effectively to predict the density and hardness of Al/SiC
composites.
3.4.2. Model validation
It was considered that the 0.0001 intended error values would
be appropriate for artificial neural network training. The num-
ber of iteration-dependent error variance for an artificial
neural network chosen for the concentration of reinforce-
 j m a t e r r e s t e c h n o l . 2 0 2 0;9(6):14036–14050 14047

Fig. 16 – Predictive regression graphs for training, validation, and testing of the developed network.

Table 9 – Compare of RSM and ANN.

Error prediction Response surface microhardness

RSM ANN
2
R 0.8957 0.9868
RMSE 1.729 1.035
MAPE 0.490 0.060

Average ANN performance error is 0.3621% and 2.40% for

the density and hardness, respectively, while testing all the
trained and testing pattern. ANN is an effective tool to predict
the hardness of the materials [38].

3.5. ANN and RSM results comparison for Al/SiC

composites

Fig. 17 – Performance curve of the developed model (Mean
squared error (MSE) Vs. number of epochs.
ment is shown in Fig. 17. The best validation performance
was achieved at 379 epochs, and developed network attains
accuracy even at a lower value of epochs.
RSM and ANN are prediction tools capable of solving multi-
variate linear and nonlinear correlation problems. From the
prior studies, RSM and ANN have been a powerful tool for
the estimation of process variables for the manufacturing of
composite materials[24,39]. These models have been used to
identify the impact of input parameters on output variables
and also to describe the relations between variables. Differ-
14048 j m a t e r r e s t e c h n o l . 2 0 2 0;9(6):14036–14050
Fig. 18 – The comparison between RSM and ANN surface
hardness values for synthesised composites.
ent statistical errors were estimated to compare the predicted
data gathered from established RSM and ANN, as depicted in
Table 9. It can be noted from the table R2 value of 0.8957 and
0.9868, for RSM and ANN respectively. This outcome indicates
that there is a higher deviation to the prediction of Response
surface methodology than that of ANN prediction.
The comparison between the predicted RSM and ANN val-
ues are further demonstrated in Fig. 18. The results revealed
that the ANN model is more suitable over RSM to analyse
the interacting variables and predictions. Comparison of the
results had been proved to predict the hardness close to exper-
imental gathered readings with a confidence interval of 95%.
It is observed from Fig. 18, ANN model estimation is more
reliable than the model based on RSM, and thus it is recognised
that ANN has a preferable modelling capability. The finding is
supported by a few recent studies [21,32,40].
4. Conclusions
The present research work investigates the synthesis,
microstructural characterisations, hardness behaviour of the
Al-x wt.% SiC (x = 5, 7.5 and 10 wt.%) sintered composites syn-
thesised by powder metallurgy technique. RSM technique was
used for the design of experiments, modelling and optimi-
sation of process variables. Further, ANN prediction of the
hardness and density for Al/SiC composites were done and
resembled the results achieved by the RSM techniques. From
this investigation, the following major conclusions are drawn.
(1) The SiC micron particles were encapsulated and evenly
distributed in the Al matrix, which was verified by ele-
mental mapping and optical microscopy. The effect of
SiC addition on the Vickers microhardness of the synthe-
sised composites results in the increment with increasing
SiC particles. The hardness of the composites was higher
than the base Al in all compositions. XRD and optical
microscopy analyses verified the presence of SiC particles
in Al composites microstructures.
(2) The composites with 7.5 wt.% reinforcement (SiC) content
and 550 ◦ C sintering temperature reveal the highest Vick-
ers hardness 59.68 VHN. Further incorporation of SiC leads
to a reduction in composites hardness due to the agglom-
eration of the reinforcing element at the interface of the
matrix particles.
(3) Utilising the RSM design of experiments (Design-Expert
software), a quadratic model was developed to predict the
Al/SiC composite density and hardness in aspects of the
content of SiC and sintering temperature. The response
assessed from the model equation shows a strong agree-
ment with the results observed.
(4) An ANN model was developed to forecast the hardness of
the synthesised Al/SiC composites. The surface hardness
values predicted in the ANN model are in proximity to the
experimental results.
(5) RSM and ANN models are considerably well correlated and
have an R-square value of 0.8957 and 0.9868, respectively
(closer to unity). This finding clearly shows that there is a
larger deviation to the forecast of RSM than that of ANN.
The RMSE and MAPE for RSM were 1.725 and 0.490 while
for ANN it was 1.035 and 0.060 respectively. The results
indicate that the ANN model is more accurate than RSM
to analyse interactions the parameters and prediction.
Authors contribution
Mohammad Azad Alam (Conceptualization, Investigations
& conduction-major activities), Hamdan H. Ya (Conceptu-
alization), Mohammad Azeem (experimental conduction-
assistance), Hussain P.B (supervision), S.M Sapuan (super-
vision), Rehan Khan (writing-reviewing), Sajjad Arif (experi-
mental conduction- assistance). Ansari A.H (Conceptualiza-
tion),
Declarations of competing interest
The authors declare no conflict of interest.
Declaration of Competing Interest
The authors report no declarations of interest.
Acknowledgements
The authors admiringly acknowledge the support of Mechan-
ical Engineering Department, Universiti Teknologi Petronas,
Malaysia for granting PhD Scholarship under GA scheme,
and authors also would like to thank Mechanical Engineer-
ing Department (ZHCET, Aligarh Muslim University, Aligarh,
India) for providing necessary facilities for experiments and
Advanced Engineering Materials and Composites Research
Centre (AEMC), Department of Mechanical and Manufacturing
Engineering, Universiti Putra, Malaysia for informative sup-
port throughout the work.
references
[1] Verma N, Vettivel SC. Characterisation and experimental
analysis of boron carbide and rice husk ash reinforced
 j m a t e r r e s t e c h n o l . 2 0 2 0;9(6):14036–14050
AA7075 aluminium alloy hybrid composite. J Alloys Compd
2018;741:981–98.
[2] Fernández H, Ordoñez S, Pesenti H, González RE, Leoni M.
Microstructure homogeneity of milled aluminum
A356-Si3N4 metal matrix composite powders. J Mater Res
Technol 2019;8:2969–77.
[3] Surya MS, Prasanthi G. Manufacturing, microstructural and
mechanical characterisation of powder metallurgy processed
Al7075/SiC metal matrix composite. Mater Today Proc 2020.
[4] Halil K, Ismail OI, Sibel D, Ramazan Çi. Wear and mechanical
properties of Al6061/SiC/B4C hybrid composites produced
with powder metallurgy. J Mater Res Technol 2019;8:5348–61.
[5] Şenel MC, Gürbüz M, Koç E. Fabrication and characterisation
of synergistic Al-SiC-GNPs hybrid composites. Compos Part B
Eng 2018;154:1–9.
[6] Şenel MC, Gürbüz M, Koç E. New generation composites with
graphene reinforced aluminium matrix. Eng Mach
2015;56:36–47.
[7] Hafizpour HR, Sanjari M, Simchi A. Analysis of the effect of
reinforcement particles on the compressibility of Al-SiC
composite powders using a neural network model. Mater
Des 2009;30:1518–23.
[8] Mohit H, Arul Mozhi Selvan V. Optimization of the tensile
strength of sintered Al6061/SiC nanocomposites using
response surface methodology. Mater Today Proc 2020.
[9] Almotairy SM, Boostani AF, Hassani M, Wei D, Jiang ZY. Effect
of hot isostatic pressing on the mechanical properties of
aluminium metal matrix nanocomposites produced by dual
speed ball milling. J Mater Res Technol 2020;9:1151–61.
[10] Nassar AE, Nassar EE. Properties of aluminium matrix Nano
composites prepared by powder metallurgy processing. J
King Saud Univ – Eng Sci 2017;29:295–9.
[11] Xie MX, Li YX, Shang XT, Wang XW, Pei JY. Microstructure and
mechanical properties of a fiber welded socket-joint made of
powder metallurgy molybdenum alloy. Metals 2019;9:640.
[12] Khuri AI, Mukhopadhyay S. Response surface methodology.
Wiley Interdiscip Rev Comput Stat 2010;2:128–49.
[13] Sharma VK, Kumar V, Joshi RS. Parametric study of
aluminium-rare earth based composites with improved
hydrophobicity using response surface method. J Mater Res
Technol 2020;9:4919–32.
[14] Kumar SS, Erdemir F, Varol T, Kumaran ST, Uthayakumar M,
Canakci A. Investigation of WEDM process parameters of
Al–SiC–B4C composites using response surface methodology.
Int J Light Mater Manuf 2019.
[15] Khajelakzay M, Bakhshi SR. Optimization of spark plasma
sintering parameters of Si3N4-SiC composite using response
surface methodology (RSM). Ceram Int 2017;43:6815–21.
[16] Sardar S, Pradhan SK, Karmakar SK, Das D. Modeling of
abraded surface roughness and wear resistance of
aluminum matrix composites. J Tribol 2019;141:25–9.
[17] Daniel G. Principles of artificial neural networks, vol. 7.
World Scientific; 2013.
[18] Hafeez A, Ammar Taqvi SA, Fazal T, Javed F, Khan Z, Amjad
US, et al. Optimisation on cleaner intensification of ozone
production using Artificial Neural Network and Response
Surface Methodology: parametric and comparative study. J
Clean Prod 2020;252:119833.
[19] Roohi R, Jafari M, Jahantab E, Aman MS, Moameri M, Zare S.
Application of artificial neural network model for the
identification the effect of municipal waste compost and
biochar on phytoremediation of contaminated soils. J
Geochem Explor 2020;208:106399.
[20] Negassi M, Suarez-Ibarrola R, Hein S, Miernik A, Reiterer A.
Application of artificial neural networks for automated
analysis of cystoscopic images: a review of the current status
and future prospects. World J Urol 2020:1–10.
14049
[21] Kannaiyan M, Karthikeyan G, Thankachi Raghuvaran JG.
Prediction of specific wear rate for LM25/ZrO2 composites
using Levenberg-Marquardt back-propagation algorithm. J
Mater Res Technol 2020;9:530–8.
[22] Kurt HI, Oduncuoglu M, Yilmaz NF, Ergul E. A Comparative
Study on the Effect of Welding Parameters of Austenitic
Stainless Steels Using Artificial Neural Network and Taguchi
Approaches with ANOVA Analysis 1845.
[23] Dinaharan I, Palanivel R, Murugan N, Laubscher RF.
Application of artificial neural network in predicting the
wear rate of copper surface composites produced using
friction stir processing. Aust J Mech Eng 2020:1–12.
[24] Kumar R, Chauhan S. Study on surface roughness
measurement for turning of Al 7075/10/SiCp and Al 7075
hybrid composites by using response surface methodology
(RSM) and artificial neural networking (ANN). Meas J Int
Meas Confed 2015;65:166–80.
[25] Arif S, Alam T, Ansari AH, Bilal M, Shaikh N. Morphological
characterisation, statistical modelling and tribological
behaviour of aluminum hybrid nanocomposites reinforced
with micro- nano-silicon carbide. J Asian Ceram Soc
2019;00:1–15.
[26] Ravindran P, Manisekar K, Rathika P, Narayanasamy P.
Tribological properties of powder metallurgy–processed
aluminium self lubricating hybrid composites with SiC
additions. Mater Des 2013;45:561–70.
[27] Reddy MP, Shakoor RA, Parande G, Manakari V, Ubaid F,
Mohamed AMA, et al. Progress in natural science: materials
international enhanced performance of nano-sized SiC
reinforced Al metal matrix nanocomposites synthesised
through microwave sintering and hot extrusion techniques.
Prog Nat Sci Mater Int 2017;27:606–14.
[28] Aziz EIA. Characterisation of the microstructure and
mechanical properties of Al/SiC composite produced by FSP
technique, vol. 9; 2019. p. 28–33.
[29] Bisht A, Srivastava M, Kumar RM, Lahiri I, Lahiri D.
Strengthening mechanism in graphene nanoplatelets
reinforced aluminum composite fabricated through spark
plasma sintering. Mater Sci Eng A 2017;695:20–8.
[30] El-kady O, Fathy A. Effect of SiC particle size on the physical
and mechanical properties of extruded Al matrix
nanocomposites. Mater Des 2014;54:348–53.
[31] Khdair AI, Fathy A. Enhanced strength and ductility of Al-SiC
nanocomposites synthesised by accumulative roll. Integr
Med Res 2019;9:478–89.
[32] Rahimi MH, Shayganmanesh M, Noorossana R, Pazhuheian
F. Modelling and optimisation of laser engraving qualitative
characteristics of Al-SiC composite using response surface
methodology and artificial neural networks. Opt Laser
Technol 2019;112:65–76.
[33] Gopalakannan S, Senthilvelan T. Application of response
surface method on machining of Al–SiC nano-composites.
Measurement 2013;46:2705–15.
[34] Varol T, Canakci A, Ozsahin S. Artificial neural network
modeling to effect of reinforcement properties on the
physical and mechanical properties of Al2024-B4C
composites produced by powder metallurgy. Compos Part B
Eng 2013;54:224–33.
[35] Arif S, Alam MT, Ansari AH, Shaikh MBN, Siddiqui MA.
Analysis of tribological behaviour of zirconia reinforced
Al-SiC hybrid composites using statistical and artificial
neural network technique. Mater Res Express 2018;5:56506.
[36] Mahanta S, Chandrasekaran M, Samanta S, Arunachalam R.
Multi-response ANN modelling and analysis on sliding wear
behavior of Al7075/B4C/fly ash hybrid nanocomposites.
Mater Res Express 2019;6:0850h4.
14050 j m a t e r r e s t e c h n o l . 2 0 2 0;9(6):14036–14050
[37] Muthukrishnan N, Davim JP. Optimisation of machining
parameters of Al/SiC-MMC with ANOVA and ANN analysis. J
Mater Process Technol 2009;209:225–32.
[38] Shaikh MBN, Raja S, Ahmed M, Zubair M, Khan A, Ali M. Rice
husk ash reinforced aluminium matrix composites:
fabrication, characterisation, statistical analysis and artificial
neural network modelling. Mater Res Express 2019;6:056518.
[39] Antil SK, Antil P, Singh S, Kumar A, Pruncu CI. Artificial
neural network and response surface methodology based
analysis on solid particle erosion behavior of polymer matrix
composites. Materials (Basel) 2020;13.
[40] Nwobi-Okoye CC, Ochieze BQ. Age hardening process
modeling and optimisation of aluminum alloy A356/Cow
horn particulate composite for brake drum application using
RSM, ANN and simulated annealing. Def Technol
2018;14:336–45.