Measurement 101 (2017) 111–117

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

Geometrical optimization of strain gauge force transducer using GRA

method
Rakesh Kolhapure ⇑, Vasudev Shinde, Vijay Kamble
Department of Mechanical Engineering, Textile and Engineering Institute, Ichalkaranji, India

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

Article history: This paper presents the geometrical optimization of strain gauge ‘Double Ended Shear Beam’ (DESB) force
Received 28 December 2015 transducer. Multi objective optimization of elastic member of transducer is carried out by maximum
Received in revised form 11 January 2017 stress sustaining capacity of strain gauge by minimizing volume. Optimal design utilize the Finite
Accepted 17 January 2017
Element Method (FEM) software. Combined Taguchi and Grey Relational Analysis (GRA) is used for
Available online 19 January 2017
optimization. Taguchi method has been applied to construct an orthogonal array based on selected
parameters. Next, multi-objective optimization is performed. As Taguchi method is unable to perform
Keywords:
multi-objective optimization combined approach of Taguchi and Grey Relational Analysis (GRA) is carried
Strain gauge
Transducer
out. This combined approach gives optimum parametric combination based on highest Grey Relation
Stress Coefficient (GRC). Analysis of Variance (ANOVA) is statistical technique is used to investigate contribution
Volume of each process parameters on the performance characteristic.
Taguchi Ó 2017 Elsevier Ltd. All rights reserved.
FEM
GRA
ANOVA

1. Introduction
Earlier in weighing industry lever scale were widely used. This
mechanism involve principle weight balancing but this process is
more time consuming and not reliable. Continuously the research
is going on for new innovative design in the engineering field.
Optimization is one of the area where design engineering are
focusing for obtaining the best result under given circumstances.
In mass and force metrology the deformation measurement of
characteristically shaped bodies, so called spring elements, is a
common principle to determine forces. The force F is acting on
the spring element and the generated deformation or strain is
measured. Load cells with glued strain gauges on it, is the most
common example for this principle [1]. A load cell is a transducer
that is used to create an electrical signal whose magnitude is
directly proportional to the force being measured [2]. Thein pre-
sented that structural and shape optimization of S type load cell
was carried out for minimizing stress, mass, displacement and
maximizing reliability index simultaneously were examined.
Structural responses as well as geometrical sensitivities are inves-
tigated by a FEM method, and reliability performance is calculated
by a reliability loading-case index (RLI) [3].
⇑ Corresponding author.
E-mail address: rockkolhapure@yahoo.com (R. Kolhapure).
The various types of load cells include hydraulic load cells,
pneumatic load cells and strain gauge load cells. Strain gauge load
cells are widely used due to its accuracy and lower unit cost as
compared with other types of load cell. For current research work
shear beam type such as ‘Double Ended Shear Beam’ load cell is
used. The characteristic of this load cell is similar to single ended
type load cell. It is widely used in weighbridge application [4].
The electrical resistance of many metals change when the
metals are mechanically elongated or contracted. The strain
gauge utilizes this principle and detects a strain by changes in
resistance [5]. A load cell is made by bonding strain gauges to a
spring material. To efficiently detect the strain, strain gauges are
bonded to the position on the spring material where the strain will
be the largest. As in all ‘Double Ended Shear Beam’ load cell strain
gauges are mounted on a thin web in the center of the cell’s
machined cavity.
2. Design of experiment
Design of experiments is a statistical technique introduced in
1920 by Sir R.A. Fisher in England. He systematically introduced
statistical thinking and principles into designing experimental
investigations, including ANOVA and factorial design concept [6].
It is used to determine the optimal factor settings of a process
and thereby achieving improved process performance, reduced
process variability and improved manufacturability of products

http://dx.doi.org/10.1016/j.measurement.2017.01.030
0263-2241/Ó 2017 Elsevier Ltd. All rights reserved.
112 R. Kolhapure et al. / Measurement 101 (2017) 111–117

and processes. The experimental design proposed by Taguchi Table 1

involves using orthogonal arrays to organize the parameters affect- Parameter and level for ‘Double Ended Shear Beam’ type load cell.

ing the process and the levels at which they should be varies. Sr. No. Parameter Unit Level 1 Level 2 Level 3
Instead of having to test all possible combinations like the factorial 1 Length mm 137.90 197 256.10
design, the Taguchi method tests pairs of combinations [7]. 2 Height mm 35 50 65
Optimization is a technique for obtaining the best result under 3 Thickness mm 30.10 43 55.90
given circumstances. In design, construction, and maintenance of
any engineering system, engineers have to take decisions at several
stages. The ultimate goal of all such decisions is either to minimize Table 2
the required effort or to maximize the desired benefit. Since the Experimental values of different variables.
required effort or the desired benefit in any practical situation Experiment No. Parameter 1 Parameter 2 Parameter 3
can be expressed as a function of certain decision variables,
1 137.9 35 30.1
optimization can be defined as the process of finding the condi- 2 137.9 50 43
tions that give the maximum or minimum value of a function 3 137.9 65 55.9
[8]. ‘Double Ended Shear Beam’ type load cell structure and 4 197 35 43
5 197 50 55.9
working of this load cell is similar to simply supported beam which
6 197 65 30.1
aids bending. So, to avoid excessive bending various dimensional 7 256.1 35 55.9
parameters are studied, those are ‘length’, ‘height’, ‘thickness’ and 8 256.1 50 30.1
‘cavity dimensions (i.e. cavity length, cavity height and cavity 9 256.1 65 43
radius)’. As strain gauges are fit inside the cavity at an angle 450,
so it is not critical parameter. Remaining three parameters, namely
Table 3
‘length’, ‘height’ and ‘thickness’ are considered critical parameters
Analyzed data and S/N ratio of observed results.
shown in Fig. 1. To get better strength, avoid maximum bending
and increase the life of load cell above stated parameters are Orthogonal Analyzed data S/N ratio (DB)
array
selected as critical dimensional parameters. The parameters and Volume Sensitivity Volume Sensitivity
corresponding levels for ‘Double Ended Shear Beam’ type load cell, (mm3) (lstrain/N
103)
is shown in Table 1. 1 105626.342 12.40 100.475 38.132
EN 24 Steel is a material used, having high wear resistance, high 2 224775.931 6.20 107.035 44.152
3 388948.825 3.70 111.798 48.636
toughness and strength property with its material density
4 213486.677 8.40 106.587 41.514
7840
105 Kg/mm3, Young’s Modulus 2.1
105 N/mm2, Poisson 5 412802.109 4.80 112.315 46.375
Ratio 0.3 and Yield strength 600 N/mm2. To carry out analysis 6 256762.306 15.00 108.191 36.478
following assumptions were made, 7 358130.032 6.00 111.081 44.437
8 239508.947 19.80 107.586 34.067
9 515508.588 4.50 114.245 46.936
(1) Load cell is fixed to rigid support by means of bolts.
(2) Uniformly distributed load is acting on top surface of load
cell.
(3) At the location of strain gauges values of strains are Sensitivity and volume are the two response variable are
measured. selected. ‘‘Larger the better” quality characteristic for sensitivity,
L9 OA is selected represented in Table 2 to make the further whereas ‘‘smaller the better” quality characteristic for volume. Sig-
experiments. This array specifies 9 experiments. The L9 OA com- nal to noise (S/N) ratio is determined for response variable, using
prising of 3 parameters with each having 3 levels are represented Eq. (1) for ‘‘smaller the better” and Eq. (2) ‘‘larger the better” char-
in the Table 2. acteristic. Values of S/N ratio are tabulated in Table 3.

Fig. 1. ‘Double Ended Shear Beam’ type load cell with its critical dimensional parameters [9].
R. Kolhapure et al. / Measurement 101 (2017) 111–117 113

Fig. 2. Steps in FEM.

114 R. Kolhapure et al. / Measurement 101 (2017) 111–117
n ¼ 10log10 ½mean of sum of square of measured data ð1Þ
n ¼ 10log10 ½mean of sum squares of reciprocal of measured data
ð2Þ
2.1. Finite Element Method (FEM)
Finite Element Method (FEM) is a numerical technique, which
give near accurate solutions to complex field problems. FEM
involves dividing the complex structures into known number of
smaller structures. This is called discretization or meshing, which
makes the technique more effective in analyzing irregular shaped
structures of engineering problems. Structural analysis for con-
structed orthogonal array represented in Table 2 is carried out by
finite element analysis (FEA). Fig. 2 shows steps of FEM.
To maximize sensitivity of ‘DESB’ load cell, value of strain can-
not exceed above 1000 lstrain. As load carrying capacity of ‘DESB’
load cell is 49050 N (300% of rated capacity) [9] we apply pressure
of 131 N/mm2 as shown in Fig. 2d. Following assumptions are
made, to achieve above objective,
(1) Load cell is fixed at its two end.
(2) Uniformly distributed load (UDL) is applied on top face of
load cell.
3. Results and discussion
3.1. Effect of process parameters on volume
In order to study the effect of process parameters on volume,
experiments were conducted using L9 OA represented in Table 2.
The analyzed data and signal to noise ratio (S/N) are given in
Table 3. The Minitab 16Ó software tool is used to analyze the
parameters.
From Fig. 3 it is clear that, Volume increase with increase in
dimension of length, height and thickness of load cell. As volume
is ‘‘smaller the better” type quality characteristic, from Fig. 3, it
can be seen that the third level of length (A3), third level of height
(B3), third level of thickness (C3) result in minimum value of
volume.
Fig. 3. Mean of S/N ratio plot for volume.
3.2. Effect of process parameters on sensitivity
From Fig. 4 it is clear that, sensitivity increase with increase in
dimension of length, and decrease with increase in thickness.
Sensitivity increase with increase in height first and then decrease.
As sensitivity is ‘larger the better’ type quality characteristic, from
Fig. 4, it can be seen that the third level of length (A3), third second
of height (B2), first level of thickness (C1) result in maximum value
of sensitivity.
3.3. Multi response parametric optimization
Taguchi method is used for optimizations of parameters using
signal to noise ratio. Higher signal to noise ratio means closer to
optimal of parameters. It can optimize the single response only
and unable to optimize if the number of responses are more than
one [10]. In many cases, parameters cannot be set only for one
response, as the objective would be to minimize and maximize
some response. So for current research work there is a need of
multi response optimization. Grey Relational Analysis (GRA)
method is used for multi response optimization. Which is used to
solve the complicated interrelationships among the various multi-
ple performance characteristics effectively. The multi response
optimization first convert multiple objective into single objective
[11]. From the GRA, a grey relational grade (GRG) is obtained to
evaluate the multiple performance Characteristics. As a result, con-
version of complicated multiple performance characteristics can be
single performance characteristic. The various steps in GRA include
normalization of process parameters for calculating grey relational
coefficient (GRC).
In order to optimize volume and sensitivity simultaneously,
grey relational analysis (GRA) has been utilized. For multi response
optimization using GRA, following steps have been used,
(1) Identification of performance characteristics and process
parameters to be evaluated.
(2) Selection of process parameters levels.
(3) Conduct the experiments at different settings of parameters
based on the orthogonal array.
(4) Convert the experimental data into S/N values.
(5) Normalization of S/N ratio.
 R. Kolhapure et al. / Measurement 101 (2017) 111–117 115

Fig. 4. Mean of S/N ratio plot for sensitivity.

(6) Calculate the grey relational coefficient.
(7) Calculate the grey relational grade by using the weighing
factor for the performance characteristics.
(8) Analyze the experimental results using the grey relational
grade and statistical analysis of variance.
(9) Select the optimal levels of process parameters.
Conduct the confirmation experiment to verify the optimal pro-
cess parameter settings [12].
The obtained S/N ratio data are converted to normalized values
using Eqs. (3) and (4). The normalized data is more evenly
distributed without units and it is used for further analysis. The
Eq. (1) used for ‘smaller the better’ and Eq. (2) for ‘larger the better.
And their values are represented in Table 4. Lager the normalized
results correspond to better performance and result should be
equal to 1.
max x0i ðkÞ  x0i ðkÞ
xi ðkÞ ¼
max x0i ðkÞ  min x0i ðkÞ
x0i ðkÞ  min x0i ðkÞ
xi ðkÞ ¼
max x0i ðkÞ  min x0i ðkÞ
where xi ðkÞ is the value after grey relational generation,
max x0i ðkÞ and min x0i ðkÞ are the largest and smallest value x0i ðkÞ.
3.4. Determination of deviation sequence
The deviation sequence Doi ðkÞ is the absolute difference
between the reference sequence x0i ðkÞ and the comparability
sequence xi ðkÞ and it is calculated from normalized value repre-
sented in Table 4. It is determined using Eq. (5). Values are repre-
sented in Table 4.
Doi k ¼ jx0 ðkÞ  xi ðkÞj ð5Þ
3.5. Calculation of grey relational coefficient (GRC)
GRC are calculated to express the relationship between the
ideal and actual normalized S/N ratio. The grey relational
coefficient can be expressed by Eq. (6). Values are represented in
Table 5.
ð3Þ
Dmin þ n  Dmax
c0;i ðkÞ ¼ ð6Þ
D0i ðkÞ þ n  Dmax
ð4Þ where Dmin is the smallest value of 0i(k) = minimin k |x⁄0(k)  x⁄i (k)|
and Dmax is the largest value of 0i(k) = maximax k |x⁄0(k)  x⁄i (k)|,
x⁄0(k) is the ideal normalized S/N ratio, xi⁄(k) is the normalized
comparability sequence and n distinguishing coefficient n 2 ð0; 1Þ
generally it is considered as 0.5 [12–13].

Table 4
Normalized S/N ratio and deviation sequence.
Table 5
Normalized S/N ratio Deviation sequence Grey relational coefficients and grade values.

Volume Sensitivity Volume Sensitivity Volume Sensitivity GRG Rank

0.000 0.721 1.000 0.279 0.333 0.642 0.488 7
0.476 0.308 0.524 0.692 0.488 0.419 0.454 9
0.822 0.000 0.178 1.000 0.738 0.333 0.536 6
0.444 0.489 0.556 0.511 0.473 0.494 0.484 8
0.860 0.155 0.140 0.845 0.781 0.372 0.576 4
0.560 0.834 0.440 0.166 0.532 0.751 0.642 3
0.770 0.288 0.230 0.712 0.685 0.413 0.549 5
0.516 1.000 0.484 0.000 0.508 1.000 0.754 1
1.000 0.117 0.000 0.883 1.000 0.361 0.681 2
116 R. Kolhapure et al. / Measurement 101 (2017) 111–117

Table 6
ANOVA for grey relational grade.

Parameter DOF Seq SS Adj SS Adj MS F P % Contribution

Length 2 0.042981 0.042981 0.0214905 14.02 0.067 53.3515
Height 2 0.021018 0.021018 0.010509 6.86 0.127 26.0892
Thickness 2 0.013498 0.013498 0.006749 4.4 0.185 16.7548
Error 2 0.003065 0.003065 0.0015325 – – 3.8045
Total 8 0.080562 – – – – 100.0000

3.6. Determination of weighted grey relational grade (GRG)
The grey relational grade is calculated by averaging the values
of grey relational coefficient corresponding to each performance
characteristic. The overall performance characteristic of the multi-
ple response process depends on the calculated grey relational
grade. It can be calculated by using Eq. (7) and values are repre-
sented in Table 5.
ANOVA table shows the percentage contribution of each
parameter. It is clear from ANOVA table that parameter only length
is significant for multi objective response whereas remaining
parameters are less significant.
Using Taguchi method, response table has been generated to
separate out the effect of each level of process parameters on grey
relational grade as represented in Table 7.

1X m
cðx0 ; xi Þ ¼ cðx0 ðkÞ; xi ðkÞÞ ð7Þ 3.7. Prediction of grey relational grade under optimum parameters
m i¼1
After evaluating the optimal parameter settings, the next step is
where cðxO ; xi Þ the grey relational grade for the jth experiment and
to predict and verify the improvement of quality characteristics
m is the number of performance characteristics.
using the optimal parametric combination.
It is clearly observed from Table 5 grey relational grade that the
The optimal Grey relational grade (gopt) is predicted using
process parameter ‘‘setting of experiment No. 8” has the highest
Eq. (8).
grey relational grade (0.754) thus the eight number experiment
gives the best multiple performance characteristics among the
nine experiments. Basically, larger the grey relational grade, better gopt ¼ T þ ðA ð1=2=3Þ  TÞ
 þ ðB
 ð1=2=3Þ  TÞ  ð1=2=3Þ  TÞ
 þ ðC  ð8Þ
the corresponding multi-objective characteristics. The obtained
GRG is considered a single response for designed experiments  = overall mean of the response
T
and analysis of variance (ANOVA) is carried out for finding most A ð1=2=3Þ ; B  ð1=2=3Þ = average values of response at the first
 ð1=2=3Þ ; C
significant parameters which affect the multi objective response or second or third levels of parameters A, B and C respectively.
which is a statistical technique. In the present research work
ANOVA is carried out on Minitab 2016. ANOVA is given in Table 6.
gopt ¼ 0:574 þ ð0:661  0:574Þ þ ð0:619  0:574Þ
ANOVA calculations are based on F-ratio, which is the ratio
between the regression mean square and the mean square error þ ð0:628  0:574Þ ¼ 0:76
which is also called as variance ratio. This ratio is used to measure
the most affecting parameters. If calculated value of F-ratio is By using optimum combination A3B3C1, from Table7 validation
higher than its tabulated value, then the factor is significant [14]. experiments were performed and corresponding values were dis-
played in Table 8. This value indicates that there is improvement
in Grey Relational grade from 0.754 to 0.808 i.e., a total of 0.054
Table 7
Response table for grey relational grade (GRG). process is improved.

Levels Length (mm) Height (mm) Thickness (mm)

1 0.492 0.507 0.628
4. Results and discussion
2 0.567 0.595 0.540
3 0.661 0.619 0.554
Max 0.661 0.619 0.628 Based on ranking order obtained from Table 5 and Fig. 5, exper-
Min 0.492 0.507 0.540 iment no.8 is considered the best response value. Basically, larger
Delta 0.169 0.113 0.088 the grey relational grade, better the corresponding characteristics.
Rank 1 2 3
From the response table for grey relational grade, the best combi-
Total mean value of GRG is 0.574. nation of the process parameters is set with A3B3C1.

Table 8
Predicted and experimental values.

Sr. No. Parameters Initial setting Prediction value Validation value

1 Optimal Parameters A3B2C1 A3B3C1 A3B3C1
2 Length (mm) 256.1 256.1
3 Height (mm) 50 65
4 Thickness (mm) 30.1 30.1
5 Grey Relational Grade 0.754 0.761 0.808
6 Improvement in grey relational grade 0.054
 R. Kolhapure et al. / Measurement 101 (2017) 111–117
Fig. 5. Graph shows ranking of experiment no. with respective grade value.
5. Conclusions
The aim of research work is to find the optimum shape of strain
gauge based load cell to get the better response or in other words
to improve the load cell performance by minimizing volume and
maximizing sensitivity. Design of experiment techniques have
been utilized for investigation and optimization of selected param-
eters, in order to get better response The conclusion based on multi
objective optimization using Taguchi with GRA are summarized as
follows:
(1) ‘Double Ended Shear Beam’ type load cell, used for research
work having volume 328196.92 mm3 and sensitivity
4.70
103 lstrain/N. The Optimization method gives
results as volume 317948.92 mm3 and sensitivity
15.30
103 lstrain/N i.e. volume is reduced by 3.12% and
sensitivity is increased by 69.28%.
(2) Using GRA, Initial setting A3B2C1 grade i.e., 0.754 increases
by using new optimum combination A3B3C1 up to 0.808
means there is increment in grade of 0.054. Therefore, using
present approach process parameters have been successfully
optimized for better performance of load cell.
(3) The Analysis of Variance resulted that length has major
influence on the sensitivity and volume.
(4) The optimal results of this paper expected to provide design
guidance for force sensor with significantly.
Acknowledgment
The authors would like to thanks Mr. R.D. Kulkarani (MCS
Instruments & Engineers) for providing load cell for creating 3d
model. We are also thankful to Mr. R.V. Tambad (FRI, Ichalkaranji)
117
for giving information about load cell that include its working
principal and calibration procedure.
References
[1] M. Kuhnel, F. Hilbrunner, H. Buchner, G. Jager, E. Manske, T. Frohlich, Traceable
measurement of mechanical parameters of double bending beam force
transducers according to EN ISO 376, Measurement 51 (2014) 336–342.
[2] J.M. Drouet, Y. Champoux, Designing a strain gauge transducer for dynamic
load measurement in using numerical simulation, Procedia Eng. 72 (2014)
304–309.
[3] C.K. Thein, Structural sizing and shape optimization of a load cell, Int. J. Eng.
Technol. 2 (7) (2013) 196–201.
[4] L. Rice, Load cell and weigh module handbook, Rice Lake Weighing System
(2010) 1–3.
[5] K.H. Thakkar, V.M. Prajapati, B.D. Patel, Performance evaluation of strain gauge
based load cell to improve weighing accuracy, Int. J. Latest Trends Eng.
Technol. 2 (1) (2013) 103–107.
[6] D.C. Montgomery, Design and Analysis of Experiments, seventh ed., John Wiley
and Sons Inc., Hoboken, 2009.
[7] J. Antony, F. Antony, Teaching the Taguchi method to industrial engineers,
Work Study 50 (4) (2001) 141–149.
[8] S.S. Rao, Engineering Optimization Theory and Practice, fourth ed., John Wiley
and Sons Inc., Hoboken, 2009.
[9] ADI ARTECH Transducers Pvt. Ltd., Cat. No. 702010-06.
[10] D. Das, A. Sahoo, R. Das, B. Routara, Investigations on hard turning using coated
carbide insert: Grey based Taguchi and regression methodology, Procedia
Mater. Sci. 6 (2014) 1351–1358.
[11] M. Durairaj, D. Sudharsun, N. Swamynathan, Analysis of process parameters in
wire EDM with stainless steel using single objective Taguchi method and multi
objective grey relational grade, Procedia Eng. 64 (2013) 868–877.
[12] U. Dabade, Multi-objective process optimization to improve surface integrity
on turned surface of Al/SiCp metal matrix composites using grey relational
analysis, Procedia CIRP 7 (2013) 299–304.
[13] A.S. Shivade, V.D. Shinde, Multi-objective optimization in WEDM of D3 tool
steel using integrated approach of Taguchi method and grey relational
analysis, J. Ind. Eng. Int. 10 (2014) 149–162.
[14] M.K. Das, K. Kumar, T.Kr. Barman, P. Sahoo, Optimization of process
parameters in plasma arc cutting of EN31 steel based on MRR and multiple
roughness characteristics using grey relation analysis, Procedia Mater. Sci. 5
(2014) 1550–1559.