See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/357825837

Box-Behnken Design Optimization of Sand Casting Process Parameters

Article in International Journal of Engineering Technologies IJET · October 2020

DOI: 10.19072/ijet.714473

CITATIONS READS
7 203

3 authors:

Blessıng Ngozı Goodluck Aliemeke Blessing Ngozi Goodluck Aliemeke

Auchi Polytechnic Auchi Polytechnic
2 PUBLICATIONS 7 CITATIONS 21 PUBLICATIONS 12 CITATIONS

SEE PROFILE SEE PROFILE

Mobolaji Humphrey Oladeinde

University of Benin
37 PUBLICATIONS 125 CITATIONS

SEE PROFILE

All content following this page was uploaded by Blessıng Ngozı Goodluck Aliemeke on 19 March 2021.

The user has requested enhancement of the downloaded file.

INTERNATIONAL JOURNAL of ENGINEERING TECHNOLOGIES-IJET
Aliemeke and Oladeinde, Vol.6, No.2, 2020

Box-Behnken Design Optimization of Sand

Casting Process Parameters
Blessing Ngozi Goodluck Aliemeke* and M. H. Oladeinde**

*Department of Mechanical Engineering, School of Engineering, Auchi Polytechnic, Nigeria

[email protected]

** Department of Production Engineering, Faculty of Engineering, University of Benin, Nigeria

([email protected])

‡Auchi Polytechnic, School of Engineering Auchi, P.M.B. 13 Auchi, Nigeria

Tel: +2348030648051, [email protected]

Received: 04.04.2020 Accepted: 31.10.2020

Abstract- Sand casting is reputed for the manufacture of engine components as a result of its ease of operation.
An assemblage of process parameters at optimal conditions leads to enhanced mechanical properties of automobile
engine components. The Response Surface Methodology Design of experiment created an experimental layout for
the sand casting process parameters and the various levels as applied in the production of engine pistons at the
foundry. The Box-Behnken Design provided a matrix of 27 experiments to be conducted. Multiple linear Regression
technique was employed to develop a mathematical model for the hardness of the aluminium alloy. The developed
model was inputted into the evolutionary Genetic Algorithm tool box as an objective model. The optimal levels
determined from the Genetic algorithm were used to carry out actual experiment in the foundry and the result was
similar to the predicted hardness value of the developed model. Statistical ANOVA test conducted showed that the
mathematical model was adequate with a R2 value of 81.02% and R2 (adjusted) value of 60.07%. The developed
model has a p-value of 0.016 which indicates that the model was significant. The optimal values obtained for
pouring temperature, vibration frequency, vibration time and runner size are 700 oC, 31.52Hz, 59.998sec and
469.69mm2 respectively.

Keywords: Sand casting, Box-Behnken, Genetic algorithm and Statistical ANOVA

1. Introduction optimize process parameters and its

corresponding response variables in an

In the Design of experiment,
output-input relationship [2]. These designs
Response Surface Methodology (RSM)
are important for fitting second order
comprises of mainly two nonlinear models
regression polynomial to dimensional
which are Box-Behnken Design (BBD) and
response surfaces. The main aim of
Central Composite Design (CCD) [1]. It is a
Response Surface Methodology (RSM) is
technique whose prime objective is to
for optimization [3].

25
INTERNATIONAL JOURNAL of ENGINEERING TECHNOLOGIES-IJET
Aliemeke and Oladeinde, Vol.6, No.2, 2020

This study entails optimization of 4 Neural network and Particle swarm

process parameters (A, B, C, and D) and Algorithm.

their response variable. In analyzing the effect of mechanical

Box-Behnken Design is an properties on internal combustion engine

experimental design and a nonlinear model pistons, an examination of the ultimate yield

of Response Surface Methodology strength of Al-GHS 1300 was done by [8].

developed by Box and Behnken in 1960 [4]. The study showed that the aluminium alloy

The Box-Behnken Design creates an posses a very high ultimate yield strength of

experimental matrix necessary for the about 1300Mpa when compared to other

combination of the process parameters aluminium alloy and composites.

conditions [5]. The design helps to develop A study on the hardness of Silumin

quadratic response surface model which is cast piston alloy was carried out by [9]. It

used for the estimation or prediction of the investigated the effect of hardness of the

response values [6]. cast engine piston. The study inferred that

Optimization technique involves the high hardness of the piston alloy offers great

finding of maximum or minimum value of a resistance to wear and fatigue of the piston

function of a set of variables subject to some component in an engine.

constraints [7]. In casting, optimization of A mathematical model using a Box

process parameters are carried out by the Behnken Design made of 3 factors and

following techniques, Finite element, levels which was used to optimize the

Taguchi design, Response surface process parameters involved in evaluating

Methodology, Genetic algorithm, Artificial Pharmaceutical complexes [10]. It was

concluded that the resveratrol ratio had

26
INTERNATIONAL JOURNAL of ENGINEERING TECHNOLOGIES-IJET
Aliemeke and Oladeinde, Vol.6, No.2, 2020

greater influence on the two responses than employed to determine which of the factors

temperature and time. The checkpoint and are considered significant. The optimized set

R2 values were very high which shows that of conditions of the process parameters was

the optimization was properly validated. An used for validation of the experiment and the

investigation of the optimization of process mole fraction of the ester was found to be

parameters of organic ester distillation using 0.8435 which showed a great correlation

Response Surface Methodology was done between the theoretical value and the

by [10]. The process parameters employed measured organic ester.

in the distillation are reboiler duty, reflex Box Behnken Design and Central

and feed ratios. The response variable in this Composite Design of the Response Surface

study is the mole fraction of the organic Methodology were used by [11] to model

ester. The Minitab software was used to relationship between responses and the

develop the Box Behnken Design process parameters for squeeze casting. The

experimental matrix needed to display the process parameters of squeeze casting

various conditions for the distillation examined in the study are squeeze pressure,

process. A total of 15 randomized pressure duration, die temperature and

experiments were carried out and the various pouring temperature. A high and low

responses recorded. In analyzing the Box operating level was also assigned to each

Behnken Design of the Response Surface process parameter. The experiments were

Methodology a regression model containing conducted to determine 3 selected responses

main, squared and interaction terms was which are Yield strength, ultimate tensile

developed. A confidence level of 95% was strength and Surface roughness.

27
INTERNATIONAL JOURNAL of ENGINEERING TECHNOLOGIES-IJET
Aliemeke and Oladeinde, Vol.6, No.2, 2020

The developed models were statistically Design second order quadratic regression

subjected to model adequacy test. Surface model was developed. The obtained

plot showed that parameters like die quadratic regression model was used as the

temperature, squeeze pressure and pouring objective function for the Genetic

temperature contributed greatly to the Algorithm. It was concluded in the study

surface roughness model. The study that the 4 process parameters had great

exhibited 15 random test cases that were influence on the Materials Removal Rate

used to make prediction on the responses by (MRR).

applying the developed model. The response This study is aimed at optimizing

surface roughness had better prediction of sand casting process parameters using the

test cases for Central Composite Design Box-Behnken Design experimental

(CCD) than the Box Behnken Design (BBD) methodology.

when compared. 2.0 Materials and Methods

Also, a 4-process parameter The production of the cast

optimization of EN 19 carried out by [12] component was carried out by the sand

used a Box-Behnken Design of Response casting process. The mould cavity was

Surface Methodology to create an prepared to readily accept the molten metal

experimental matrix in investigating the when poured and to help reduce the number

effect of flow rate of coolant, speed, depth of wasteful casting. The mould cavity has

and feed rate of the molten steel on a the core made of green sand well placed in it

Material Removal Rate (MRR). A number to create an internal cavity for the piston.

of 28 experiments of varying conditions The molten metal was scooped out from the

were observed. The nonlinear Box-Behnken

28
INTERNATIONAL JOURNAL of ENGINEERING TECHNOLOGIES-IJET
Aliemeke and Oladeinde, Vol.6, No.2, 2020

Crucible on the furnace with the aid of a into the already prepared mould cavity on

ladle and introduced into the mould through the mould vibrating machine. The poured

the pouring basin to the runner before molten metal was vibrated by the machine

entering the casting cavity. The scrap during solidification at selected frequency of

pistons were heated in the crucible furnace 10 Hz, 30 Hz and 50 Hz. The Rockwell

and it attained its molten state for duration hardness machine was used to determine the

of fifty minutes. The pouring temperatures hardness value of the aluminium alloy

were measured with the aid of digital component. The process parameters used in

thermocouple. The molten metal was poured this experiment are shown in Table 1.

Table 1.0: Process parameters and their various levels

Process parameters LEVELS
Level 1 Level 2 Level 3
Pouring temperature, A (0C) 700 725 750
Vibration Frequency, B(Hz) 10 30 50
Vibration time, C (seconds) 30 45 60
Runner size (mm2) 180 335 490

2.1 Box-Behnken Design (BBD) III. Developing a mathematical second

The Response Surface Methodology order regression response model

design procedure used in this study as stated IV. Predicting various response values

by [13] is as follows from each set of process parameters

I. Design of experimental matrix for values so as to ascertain the optimal

conducting of the experiment using the combination

conditions ascribed to the process V. Determining main and interaction

parameters effects of experimental parameters

II. Developing a column for the response through dimensional plots.

variables after the specified number of

runs

29
INTERNATIONAL JOURNAL of ENGINEERING TECHNOLOGIES-IJET
Aliemeke and Oladeinde, Vol.6, No.2, 2020

The general form of equation for Response Surface Methodology (RSM) is

representing a 4-parameter response (Y) in a

𝑌 = 𝐹(𝐴, 𝐵, 𝐶, 𝐷) (1)

𝑌 = 𝛽0 + 𝛽1 𝐴 + 𝛽2 𝐵 + 𝛽3 𝐶 + 𝛽4 𝐷 + 𝛽5 𝐴2 + 𝛽6 𝐵2 + 𝛽7 𝐶 2 + 𝛽8 𝐷2 + 𝛽9 𝐴𝐵 + 𝛽10 𝐴𝐶 + 𝛽11 𝐴𝐷

+ 𝛽12 𝐵𝐶 + 𝛽13 𝐵𝐷 + 𝛽14 𝐶𝐷 + 𝜀 (2)

Where A, B, C, and D are the process parameters. While

Β0, β1,β2,β3,β4-------------------------β14 are the regression coefficients

The quadratic model developed contains 4-process parameter and 3-level experiment.

main, squared and interaction terms [14]. The experimental matrix contains the

The nonlinear model is notable in the columns of various factor levels {high (+1),

Response Surface Methodology (RSM) for medium (0) and low(-1)}, experimental run

addressing experimental design of a orders and the standard orders. The run

minimum of 3 factors and 3 levels [15]. order was used to conduct the experiments

Box-Behnken Design has an advantage of and the standard orders were used for the

fewer numbers of runs or experiments than randomization of experiment and the actual

Central Composite Design involving 3 or 4 order of the experiment. The randomization

process parameters. In this study, Box- ensures independency among the conditions

Behnken Design of Response Surface in the various runs. The Box-Behnken

Methodology was employed in carrying out experimental design matrix used for

the experiment. Minitab 17 software was obtaining the hardness response values is

used in getting the experimental matrix. The shown in Table 2.

design matrix has a stipulated 27 runs for a

30
INTERNATIONAL JOURNAL of ENGINEERING TECHNOLOGIES-IJET
Aliemeke and Oladeinde, Vol.6, No.2, 2020

3. Results and discussion Hardness Testing Machine is presented in

The results from the hardness Table 2.

experiment conducted using Rockwell

Table 2: Box-Behnken Experimental Values for Hardness

Run Standard Pouring Vibration Vibration Runner Hardness
order Order temp,A(oC) frequency,B(Hz) time,C(sec) size,D(mm2) H
1 19 700 30 60 335 67.9
2 15 725 10 60 335 62.7
3 12 750 30 45 490 53.5
4 1 700 10 45 335 55.5
5 6 725 30 60 180 62.6
6 8 725 30 60 490 64.0
7 23 725 10 45 490 57.8
8 5 725 30 30 180 61.0
9 9 700 30 45 180 65.0
10 14 725 50 30 335 63.8
11 13 725 10 30 335 56.6
12 2 750 10 45 335 56.8
13 3 700 50 45 335 60.0
14 16 725 50 60 335 64.6
15 4 750 50 45 335 57.0
16 25 725 30 45 335 69.5
17 10 750 30 45 180 70.0
18 18 750 30 30 335 68.5
19 20 750 30 60 335 59.8
20 26 725 30 45 335 65.5
21 17 700 30 30 335 58.5
22 27 725 30 45 335 68.5
23 11 700 30 45 490 72.1
24 7 725 30 30 490 61.0
25 21 725 10 45 180 57.8
26 22 725 50 45 180 62.0
27 24 725 50 45 490 59.9

In this study a multiple linear response variable and the various process

regression is needed to establish a parameters.

relationship between the Hardness as

31
INTERNATIONAL JOURNAL of ENGINEERING TECHNOLOGIES-IJET
Aliemeke and Oladeinde, Vol.6, No.2, 2020

A multiple linear regression model as given response variable Y and predictor variables

in equation 3 shows a relationship of A1, A2, A3 and A4

𝑌 = 𝛽0 + 𝛽1 𝐴1 + 𝛽2 𝐴2 + 𝛽3 𝐴3 + 𝛽4 𝐴4 … … … … … … . . +𝛽𝑘 𝐴𝑘 + 𝜀 (3)

where 𝛽0 = intercept and 𝛽1 , 𝛽2 , 𝛽3 , 𝛽4 are are independent variables and 𝜀 stands for

regression coefficients. Also 𝐴1 , 𝐴2 , 𝐴3 , 𝐴4 the error term

3.1 Box-Behnken Design Analysis for Hardness

The mathematical model for hardness developed regression model of Box-Behnken Design is as given

using the Response Surface Methodology nonlinear in equation 4

𝐻 = −2816 + 6.75𝐴 + 2.83𝐵 + 9.51𝐶 + 1.157𝐷 − 0.00392𝐴2 − 0.01550𝐵 2 − 0.00733𝐶 2

− 0.000091𝐷2 − 0.00215𝐴𝐵 − 0.01200𝐴𝐶 − 0.001523𝐴𝐷 − 0.00442𝐵𝐶

− 0.000169𝐵𝐷 + 0.000151𝐶𝐷 (4)

3.2 Significance Test for the Box- matrix. The essence of the test is to ascertain

Behnken Design Regression Model for the significance of the main, multiple and

Hardness H interaction parameters present in the

Significance test was carried out for regression model. The Table 3 depicts the

the hardness model obtained from the analysis of variance (ANOVA) for the Box-

nonlinear Box-Behnken experimental Behnken regression model for hardness.

Table 3: Analysis of Variance (ANOVA) for Box-Behnken Regression Model for Hardness
Source Degree of Adj SS Adj MS F-value P-value
Freedom
Model 14 511.11 36.51 3.61 0.016
Linear 4 69.96 17.49 1.73 0.208
A 1 15.19 15.19 1.50 0.244
B 1 33.67 33.67 3.33 0.093
C 1 12.61 12.61 1.25 0.286
D 1 8.50 8.50 0.84 0.378
Square 4 207.66 51.92 5.13 0.012
A2 1 32.01 32.01 3.16 0.100
B2 1 205.01 205.01 20.25 0.001
C2 1 14.52 14.52 1.43 0.254
D2 1 25.23 25.23 2.49 0.001
2-way Interaction 6 233.48 38.91 3.84 0.023

32
INTERNATIONAL JOURNAL of ENGINEERING TECHNOLOGIES-IJET
Aliemeke and Oladeinde, Vol.6, No.2, 2020
AB 1 4.62 4.62 0.46 0.512
AC 1 81.00 81.00 8.00 0.015
AD 1 139.24 139.24 13.75 0.003
BD 1 1.10 1.10 0.11 0.747
CD 1 0.49 0.49 0.05 0.830
Error 12 121.502 10.13
Lack of fit 10 112.84 11.28 2.60 0.309
Pure error 2 8.67 4.33
Total 26 632.61
R2=81.02% R2 (Adj) =60.07%

The Box-Behnken Design (BBD) values shows that over fitting occurred in

hardness model showed that two squared the model. A further test on the model

effect terms (B2 and D2) and two interaction adequacy was carried out by applying the

terms (AC and AD) are significant. Also the Normal probability and residual plots as

sharp difference between the R2 and R2 (adj) shown in Fig. 1

(a) (b)

Fig. 1: (a) Normal Probability plot for the Nonlinear Hardness (b) Residual plot for the Hardness Data

The normal probability plot shown in the variation of residuals of the treatment

Fig. 1 (a) indicates that the proximity levels and the distribution of data satisfy the

between the residual points and ideal normal normality condition.

distribution diagonal line is very high and Also Response surface plot was

this connotes that the data is normally carried out for the various interaction terms

distributed. Also the Fig. 1 (b) showed that involved in the Box-Behnken Design

33
INTERNATIONAL JOURNAL of ENGINEERING TECHNOLOGIES-IJET
Aliemeke and Oladeinde, Vol.6, No.2, 2020

hardness model. The plot shown in Fig. 2(b) value. The plots on Figs. 2 a, c and d

shows that the interaction between pouring showed interaction of process parameters

temperature and the vibration time is very that have moderate effect on the hardness

significant and it also result in high hardness value.

(a) (b)

Hardness plot 4

(C) (d)

Fig. 2: (a-d) Response Surface Plots for Hardness Interaction Terms

3.3 Genetic Algorithm Analysis for the probability adopted are 85% and 0.01

Developed Equation respectively. A number of 100 generations

The regression models obtained and 100 seconds time limit were used for the

through multiple linear regressions were optimization.

used as the objective function in the Lower bound of parameters = {700, 10,

MATLAB genetic algorithm tool. The 30, 180}

population size used was 50, number of Upper bound of parameters = {750, 50,

variable used is 4, crossover and mutation 60, 490}

34
INTERNATIONAL JOURNAL of ENGINEERING TECHNOLOGIES-IJET
Aliemeke and Oladeinde, Vol.6, No.2, 2020

The Table 4 shows the tested levels used in the hardness model from the genetic

and best optimal levels for the parameters algorithm tool.

Table 4: Result of Optimal Levels from Genetic Algorithm on BBD Hardness Model

Factor Parameter Level range Optimal level

A Pouring temp(0C) 700-750 700.000
B Vibration frequency(Hz) 10-50 31.523
C Vibration time(secs) 30-60 59.998
D Runner size 180-490 469.696
The fitness value is 70.789

3.4 Validation of Models experimental layout for the parameters and

In addition to the model validation the various levels applied in this study. It

carried out by ANOVA and R2 statistic provided an experimental matrix of 27 runs

further model validation was carried out for which were conducted and the test results

the developed models by performing recorded. Multiple linear Regression

experiment using the determined optimal technique was employed to develop a

values. The predicted values were mathematical model for the response.

determined by inputting the optimal Statistical ANOVA test conducted showed

conditions obtained from the optimization that the model was adequate with a R2

model into the developed nonlinear (adjusted) value of 60.07%. The developed

regression models. Also, the conditions were model has a p-value of 0.016 which

used to carry out actual experiment and the indicates that the model was significant. The

obtained results were compared with the optimal values obtained for pouring

predicted values. temperature, vibration frequency, vibration

4. Conclusion time and runner size are 700oC, 31.52Hz,

The Box-Behnken Design of 59.998sec and 469.69mm2 respectively. The

experiment was used to develop an developed optimal levels were used to carry

35
 INTERNATIONAL JOURNAL of ENGINEERING TECHNOLOGIES-IJET
Aliemeke and Oladeinde, Vol.6, No.2, 2020

out actual experiment in the foundry and the A6061 Pure Aluminium)”, International
Journal of Mechanical engineering, Vol. 3,
No. 1, pp. 4 – 13, 2016.
result was similar to the predicted hardness
[9] M. T. Azhagan, B. Mohan and A. Rajadurai,
value of the developed model. “Optimization of Process Parameters to
Enhance the Hardness on Squeeze Cast
References Aluminium Alloy AA6061”, International
Journal of Engineering and Technology,
Vol.6, No.1, pp. 183-190, 2014.
[1] G. C. M. Patel, P. Krishna, P. R. Vundavilli,
and M. B. Parappagouder, “Multi-objective [10] A. Giwa, A. A. Adeyi and S. O. Giwa,
Optimization of Squeeze Casting Process “Empirical Modelling and Optimization of
Using Genetic Algorithm and Particle Swarm PAME Reactive Distillation Process Using
Optimization”, Foundry Engineering, Vol. 35, Minitab”, International Journal of Scientific
No.3, pp. 172-186, 2016 and Engineering Research, Vol.6, No.6,
pp.538-549, 2015.
[2] A. I. Khuri and S. Mulchopadhyay, “Response
Surface Methodology”, WIREs [11] G. C. M. Patel, P. Krishna and M.
Computational Statistics, Vol. 2, pp. 128-150, Parappagouder, “Modelling of Squeeze
2010. Casting Process Using Design of Experiment
and Response Surface Methodology”,
[3] K. P. Praveen and M. Arum, “Multi-objective International Journal of Cast Metals Research,
Genetic Algorithm Based Optimization of Vol. 28, No.3, pp.167-180, 2015.
Process Parameters for Hard Part Turning”,
International Journal of Research in [12] R. Kumar, R. and C. Grewal, “Improvement
Engineering and Technology, Vol. 3, No.8, in Hardness of LM-6 Aluminium Green sand
pp. 25-30, 2015. Castings by Taguchi Method”, Asian Journal
of Engineering and Applied Technology,
[4] M. A. Tekindal, H. Bayrak, B. Ozkaya and Y. Vol.2, No.2, pp.11-18, 2013.
Genc, “Box-Behnken Experimental Design in
Factorial Experiment”. Turkish Journal of [13] M. Manohar, J. Joseph, T. Selvaraj and D.
Field Crops, Vol. 2, No.17, pp.115-123, 2012. Sivakumar, “Application of Box-Behnken
Design to Optimize the parameters for
[5] A. Agarwal, V. Kharb and V. A. Saharan, “ Turning Inconel 718 Using Coated Carbide
Process Optimization, characterization and Tools”, International Journal of Scientific and
Evaluation of Resveratrol-Phospholid Engineering Research, Vol.4, No.4, pp.620-
Complexes Using Box-Behnken Statistical 643, 2013.
Design”, International Current Pharmaceutical
Journal, Vol.3, No.7, pp. 301-308, 2014. [14] V. S. Mudakappanavar and H. M.
Nanjundaswamy, “Multi-objective Optimiza-
[6] L. Yue, “Design of Experiment to support tion of Process Parameters During
Automated Assembly Planning”, Master Solidification of Hypoeutectic Al-Si Alloy
Degree Thesis, Oregon State University, Casting Using Genetic Algorithm”,
Oregon, 2016. International Journal of Latest Trends in
Engineering and Technology, Vol. 2, No. 4,
[7] M. Jayaram, “Parameter Optimization of pp.424-434, 2013.
WEDM for AA6061 Using Genetic
Algorithm”, International Research Journal of [15] N. Anuar, A. F. Adnan, N. Saat, N. Aziz and
Latest Trends in Engineering and Technology, R. Taha,“Optimization of Extraction
Vol.2, No.3, pp. 96-104, 2015. Parameters by Using Response Surface
Methodology, Purification and Identification
[8] T. Akhil, K. Naresh, A. Khurshid and P. A. of Anthocyanin Pigments in Fruit”, The
Kumar, “Analysis on Four Stroke Single Scientific World Journal, Vol.2, pp.10-21,
Cylinder Engine Piston by using Aluminium 2013.
Alloys (AI–GHS 1300, Al-SiC-Graphite

36

View publication stats