Documentos de Académico
Documentos de Profesional
Documentos de Cultura
i[I
X
j[J
X
b[B
X
TrC
i;j;
b
S
i;j;b
i[I
X
b[B
X
fcoq
y
i
;b
1 2y
i;b
S
i;b
j[J
X
k[K
X
Pr C
j;k
X
j;k
j[J
X
FC
j
z
j
j[J
X
k[K
X
TrC
j;k
X
j;k
1
Subject to:
j[J
X
X
j;k
D
k
;K 2
k[K
X
X
j;k
# CAP
j*
z
j
;j 3
k[K
X
N
b
X
j;k
i[I
X
S
i;j;b
;j; ;b 4
S
i;j;b
# S
0
1 2y
i;b
;i; ;j; ;b 5
j[J
X
S
0
# SC
i;b
;i; ;b 6
S
i;b
j[J
X
S
i;j;b
;i; ;b 7
y
i;b
# AD
i;b
;i; ;b 8
z
j
1 0; 1 f g; S
i;j;b
$ 0; S
0
$ 0; X
j;k
$ 0; y
i;b
$ 0:
The objective function minimizes the following costs respectively:
(1) direct cost of procuring components from suppliers;
(2) transportation cost from suppliers to plants;
Incorporating
the cost of
quality
77
(3) total quality cost at suppliers;
(4) production cost at the plants;
(5) xed cost (FC) of producing products at plant j; and
(6) transportation cost from plants to customers.
The constraints in our model start with constraint (2), which ensures that all plants are
producing enough products to meet customers demand. Constraint (3) ensures that no
plants production exceeds its capacity given it has been open for production.
Constraint (4) ensures that the suppliers will produce enough components for the
number of products demanded at the plants. Constraint (5) relates the amount shipped
to the quantity produced at each supplier compensating for defective products. We
assume that only good products are shipped to the plants (hence a 100 percent
inspection at the supplier). Constraint (6) is a supplier capacity constraint and it
ensures that suppliers capacity is not exceeded. Constraint (7) sums the total number
of good component of type b produced by supplier i. And constraint (8) places a bound
on the maximum proportion of defective allowed.
To solve the model, few assumptions are needed to be made in order to simplify the
problem so it could be solved by common industrial software, such as Lingo. If one
assumes a single component system supplied from the suppliers, then the b term in all
the variables and parameters will be dropped out. For instance, PcC
i;j;b
, which was
dened as the direct cost of component b procured from a given supplier to a given
plant, in the model will be transformed into PcC
i;j
, which is now the cost of procuring a
component from supplier i to plant j.
Going back to the perspective of viewing the supply chain as an integrated network,
meaning that suppliers are subsidiaries of the ownership of the whole supply chain,
production cost at the supplier becomes an operational parameter and our denition of
PcC
i;j
as a procurement cost will be transformed toPcC
i
, production cost at supplier i;
similarly the same analogy is preserved for all other variables and parameters:
.
Input parameters: PcC
i
, Pr C
j
, fcoq
y
i
,TrC
j;k
, and TC
i;j
.
Decision variables: S
i
, S
i;j
,X
j;k
, z
j
, and y
i
.
Constraint parameters: D
k
, SC
i
, CAP
j
, AD
i
, and N
b
The S
0
termis eliminated, as the total number of components b procured from supplier i
to plant j is transformed to the total number of components produced at supplier i, S
i
.
Based on these assumptions, the model is constructed and solutions are obtained
using Lingo 9.0. Constraint 8 was annulled for the purpose of optimizing y freely; so an
optimal y is fetched. The data sets for the problem are generated in an attempt to cover
a wide range of practical situations. For instance, the results in this paper are for a
six-supplier, three-plant, and two-customer supply chain network; our data sets
contested different practical scenarios in regard to the COQ functions. For the situation
where suppliers are operating at a high quality cost, we have simulated two suppliers
COQ functions accordingly. For situations where suppliers are running at low quality
costs two COQ functions were simulated and another two for low quality costs were
also simulated. This is done, so the quality functions are representative of different
practical scenarios.
JQME
14,1
78
Moreover, different data sets were used to solve the model. Those data sets
represented many different cost scenarios and practical situations. In this paper we
used one of the data sets, as an example, in order to be consistent, especially when
comparing the three different cases we have contested in the model. Other data sets
will yield solutions with several variations from the illustrative example given.
Tables I-IV contain the data sets for our problem as well as optimal values of the
decision variables in our model. The value of N, which is the number of components
required to make a product, is set to a value of 3. And the optimal value of the objective
function is obtained as $641,730.179. In the tables, it is important to note that
SG i; j
S
i;j
, XT j X
j
(the sum of all products made at plant j), and
X j; k
X
j;k
.
Figure 2 shows the logistic routes the model chooses once it reaches optimality. The
model optimizes the network in accordance to operational costs and quality costs.
The solutions presented are for a network of six-suppliers, three-plants, and
two-customers and were fetched by Lingo in less than a second timing (Figure 2). The
quality functions are tabulated in Table III, among other parameters and were
simulated to resemble polynomial functions of second order, inspired by Jurans
Customers (k)
Plants ( j) 1 2
TC( j,k)
1 1.0 1.1
2 1.0 1.0
3 1.1 1.0
X(j,k)
1 1,250 250
2 0 1,050
3 0 0
Table I.
Input and output model
parameters
Plants ( j)
Suppliers (i ) 1 2 3
SG(i,j), decision variable
1 2,400 0 0
2 0 2,244 0
3 0 729 0
4 2,100 177 0
5 0 0 0
6 0 0 0
TrC(i,j), input parameter
1 1.00 1.20 1.30
2 1.20 1.10 1.25
3 1.20 1.00 1.25
4 1.10 1.15 1.05
5 1.20 1.10 1.00
6 1.20 1.10 1.00
Table II.
Input and output model
parameters
Incorporating
the cost of
quality
79
graphical presentation (Figure 1), COQ as a function ofy
i
. One can observe that the
most prevalent (i.e. constitutes high COQ) quality cost function is that of supplier 1,
produced at zero percent of defectives (Table II). This validates the model when the
cost of failure is high, consequently the model produced at zero defectives.
Special case
Lingo was able to solve the small model of six suppliers, three plants, and two
customers. However, when bigger models were contested (i.e. models with more
suppliers, plants, and customers) Lingo failed to bring about an optimal solution. With
nonlinearities present both in the objective function and in the constraints, the problem
is challenging to solve. Also the nonlinear constraints are binding and hence are more
difcult to relax. Many solution methodologies in literature propose the linearization of
Figure 2.
Optimal logistic routes
Model input parameters
Decision
variables Input parameter
Plants F PrC Capacity XT Z Customers Demand
1 11,000 100 1,500 1,500 1 1 1,250
2 10,000 102 1,500 1,050 1 2 1,300
3 9,000 104 1,500 0 0
Table IV.
Input and output model
parameters
Model input parameters Decision variables
Suppliers PcC SC fcoq (quality function) S y
1 35.5000 2400.0000 119y
2
2 39y 7 2,400.0000 0.0000
2 35.7000 2400.0000 118y
2
2 60y 12 2,400.0000 0.0651
3 35.0000 2400.0000 60y
2
2 60y 17 830.0259 0.1218
4 35.2000 2400.0000 45y
2
2 44.6y 14.5 2,400.0000 0.0512
5 34.0000 2400.0000 49y
2
2 63y 23.8 0.0000 NA
6 34.0000 2400.0000 68y
2
2 76.5y 25 0.0000 NA
Table III.
Input and output model
parameters
JQME
14,1
80
the constraint successively. In doing so, an optimal solution can be fetched. For the
relatively small model presented in this paper, the linear approximation is viable and
can produce good results. As the model gets bigger, the linearization of the constraints
will be more difcult and more calculation extensive. For this, exact solution might not
be economical and hence the simplication of the model would be needed.
The model poses several challenges to solve due to:
(1) the total cost of quality function fcoq
y
i
in the objective function is non-linear;
(2) both S
i
and y
i
are variables creating another non-linearity in the objective
function;
(3) constraint (5) involves a multiplication of two unknown quantities S
i *
y
i
and is
hence non-linear; and
(4) variables z
j
are binary in nature.
If the y values for all suppliers were xed, then issues 2 and 3 above disappeared and
the problem becomes a special case one. y
i
is now an input to the model and hence
constraint 8 would be eliminated. There are two motives for this simplication; a
solution methodology motive and a practical application-oriented motive. The rst
motive is based on the fact that most approaches, in solving a nonlinear model with
nonlinearity both in the objective function and in the constraints, rely on some form of
successive linearization. As the model size of the model increases, optimality becomes
difcult to achieve in such cases. When y is xed, the model becomes linear; hence easy
to solve. The other motive is an application-oriented motive. Management of the supply
chain network can wish to have all entities in the supply chain working at the same
percent of defectives and might focus on the overall cost of achieving such a scenario.
In Figure 3, one can observe that at high defect ratios (i.e. high ys) the overall
objective value increases. An increase in defectives will have a consequence of a
corresponding increase in production cost, as we need to compensate for defective
products by producing more. Alternatively, a decrease in y will cause an increase in
quality costs and hence an increase the overall objective value. The graph
demonstrates an initial decrease of the objective value for low values of y and
Figure 3.
Graph of the values of the
objective function vs
percent of defectives
Incorporating
the cost of
quality
81
sequential increase later for the objective values for higher y values, which concurs
with the previous two statements.
Referring back to the original problem, for smaller problems Lingo was capable of
solving the problem. But for larger problems, there are possible directions for solving
the model. The method of Lagrangian relaxation can possibly be benecial in solving
the model. Also, as mentioned in this section, when y is xed the model becomes linear
and easy to solve. Hence, other methods can be constructed to exploit this special
characteristic in the model.
Impact of COQ
This section illustrates the impact of incorporating COQ into the supply chain
network. In order to address this, the model presented previously (i.e. the COQ
integrated into the supply chain network model) was modied by removing the
COQ terms from the model (i.e. the basic supply chain network model). Looking
back at the integrated model, if we were to remove the cost of quality term, the
P
i[I
P
b[B
fcoq
y
i
;b
1 2y
i;b
S
i;b
term, from the objective function then the objective
function will be free of quality related terms. In addition, the constraint 5, S
i;j;b
#
S
0
1 2y
i;b
would also need to be eliminated. In doing so, the integrated model will be
transformed into the basic supply chain network model. The basic model was solved
with the same data sets used above. Tables V-VIII present the results of the basic model.
As both models were solved using the same data sets, the results can be compared
as follows: In the integrated model, the value of the objective function was obtained to
be $641,730. In the latter case (i.e. COQ not incorporated), the value of the function was
$552,295; indicating a difference of approximately 16 percent. This difference is due to
the contribution of COQ into the overall operation. This cost would have been hidden if
one were not to include COQ in the model.
Moreover, when COQ was not incorporated into the model (Figure 4), it can be seen
that suppliers 1 and 2 were no longer furnishing any material to any of the plants
according to the optimal solution from the basic model. Suppliers 1 and 2 with a
better COQ structure no longer dispensed any material to any plant. Alternatively, the
optimal network solution from the basic model chooses the suppliers 5 and 6 despite
the fact that the failure costs at these two suppliers were high. Hence, solutions from
the basic model, if implemented, will be prone to potential future problems at the plants
as well as throughout the network. Nonconformance costs will eventually outweigh the
Customers (k)
Plants ( j) 1 2
TC( j,k)
1 1.0 1.1
2 1.0 1.0
3 1.1 1.0
X(j,k)
1 1,250 250
2 0 1,050
3 0 0
Table V.
Input and output model
parameters
JQME
14,1
82
cost savings in the latter network. When one also considers issues of delays, inventory,
and reputation, the benets of incorporating COQ will outweigh the latter case.
With COQ incorporated into supply chain network design, suppliers behavior in
regard to quality issues will be scrutinized. Suppliers whom are incurring high cost of
quality will be less competitive or attractive even if they were to produce at lower costs.
As shown, if we were to compare the two different scenarios, one with COQ not
included in the design of the supply chain network and one with COQ included. In the
rst, the nal optimal network will choose key suppliers who have low costs. No
information is inferred in regard to the quality nonconformance cost. In this scenario a
supplier that is running at a high quality nonconformance cost is treated similarly to
Plants ( j)
Suppliers (i) 1 2 3
TrC(i,j)
1 1.00 1.20 1.30
2 1.20 1.10 1.25
3 1.20 1.00 1.25
4 1.10 1.15 1.05
5 1.20 1.10 1.00
6 1.20 1.10 1.00
SG(i,j)
1 0 0 0
2 0 0 0
3 0 2,400 0
4 450 0 0
5 2,400 0 0
6 1,650 750 0
Table VI.
Input and output model
parameters
Model input parameters Decision variables
Suppliers PcC SC fcoq (quality function) S y
1 35.50 2400 119y
2
2 39y 7 0 NA
2 35.70 2400 118y
2
2 60y 12 0 NA
3 35.00 2400 60y
2
2 60y 17 2,400 NA
4 35.20 2400 45y
2
2 44.6y 14.5 450 NA
5 34.00 2400 49y
2
2 63y 23.8 2,400 NA
6 34.00 2400 68y
2
2 76.5y 25 2,400 NA
Table VII.
Input and output model
parameters
Model input parameters
Decision
variables Input parameter
Plants F PrC Capacity XT Z Customers Demand
1 11,000 100 1,500 1,500 1 1 1,250
2 10,000 102 1,500 1,050 1 2 1,300
3 9,000 104 1,500 0 0
Table VIII.
Input and output model
parameters
Incorporating
the cost of
quality
83
the one that is operating at a lower quality nonconformance cost, given they both have
the same production cost. Even if that supplier has a lower production cost, if chosen,
additional costs would result due to product failures; this in turns would disburse the
savings realized initially. Hence, choices made solely on production cost could sacrice
quality and incite additional quality nonconformance costs. The scenario of
incorporating COQ in supply chain network design will ensure the lowest overall
cost, because it reduces the probability of defective and hence the probability of
additional cost which might be due to corrective action.
Conclusions and future work
Supply chain network design (SCND) is an important problem and attracts the
attention of many researchers. However, no research has been done so far to integrate
the vital concept of Cost of Quality into the network designs. This study attempts to
incorporate COQ in SCND. We have modeled COQ into SCND using a nonlinear
mathematical programming model. The non-linearity in the objective function,
represented by the convex quality functions, and the nonlinearity present presented in
the constraints have not deterred us from achieving insightful results. The quality
functions that were used are valuable as they can represent mathematically the quality
system of a given supplier. While, COQ costing differ among different companies, one
can most certainly analyze the behavior of COQ with respect to y, and infer a
mathematical function to represent it. Hence, using functions, we can bypass the
complications of accounting systems and infer costs, which correspond to percent of
defectives, which can be related to production costs. Our model had not only sought the
optimal quality level but had realized a solution which takes into account day to day
tradeoffs in supply chain operations. By minimizing the overall cost of the supply
chain we have fetched a reasonable y that is low enough to deter extra cost of
reproduction and high enough as not to drive COQ up.
Just like COQ was modeled at the suppliers, it can be also modeled at the plant.
Although in this study COQ was ignored at the plant, further research could model
COQ at both supplier and plants simultaneously. Also, further research could address a
multi-product sourcing and distribution network. This will increase the number of
decision variables and the number of constraints in the model and will add more
complexity to the bill of material parameters, producing more interrelations between
Figure 4.
Optimal network for the
case when COQ is not
incorporated into the
model
JQME
14,1
84
parameters and decision variables. The complexity of such a problem would perhaps
be implausible to solve by Lingo and would eventually require the formulation of a
solution methodology that might take advantage of any special characteristics of our
objective function and constraints. As alternative means of solving more complex and
larger models, we will investigate using some stochastic search procedures such as
tabu search or genetic algorithms.
References
Arntzen, B.C., Brown, G.G., Harrison, T.P. and Trafton, L.L. (1995), Global supply chain
management at digital equipment corporation, Interfaces, Vol. 25 No. 1, pp. 69-93.
Beamon, B.M. (1998), Supply chain design and analysis: models and methods, International
Journal of Production Economics, Vol. 55, pp. 281-94.
Breitman, R.L. and Lucas, J.M. (1987), PLANETS: a modeling system for business planning,
Interfaces, Vol. 17, January-February, pp. 94-106.
British Standards Institution (1991), Quality Vocabulary Quality Concepts and Related
Denitions, BS 4778: Part 2, British Standards Institution, London.
British Standards Institution (1995), Quality Management and Quality Assurance-Vocabulary,
BS EN ISO 8402, British Standards Institution, London.
Cakravastia, A., Toha, I.S. and Nakamura, N. (2002), A two-stage model for the design of supply
chain networks, International Journal of Production Economics, Vol. 80, pp. 231-48.
Camm, J.D., Chorman, T.E., Dill, F.A., Evans, J.R., Sweeney, D.J. and Wegryn, G.W. (1997),
Blending OR/MS, judgment, and GIS: restructuring P&Gs supply chain, Interfaces,
Vol. 27 No. 1, January-February, pp. 128-42.
Campanella, J. (1991), Principles of Quality Costs; Principles, Implementation and Use, 3rd ed.,
ASQ Quality Press, Milwaukee, WI.
Chopra, S. and Meindl, P. (2001), Supply Chain Management: Strategy, Planning, and Operation,
Prentice-Hall, Upper Saddle River, NJ.
Christy, D.P. and Grout, J.R. (1994), Safeguarding supply chain relationships, International
Journal of Production Economics, Vol. 36, pp. 233-342.
Cohen, M.A. and Lee, H.L. (1988), Strategic analysis of integrated production-distribution
systems: models and methods, Operations Research, Vol. 36 No. 2, pp. 216-28.
Cohen, M.A. and Lee, H.L. (1989), Resource deployment analysis of global manufacturing and
distribution networks, Journal of Manufacturing and Operations Management, Vol. 2,
pp. 81-104.
Jayaraman, V. and Ross, A. (2005), A simulated annealing methodology to distribution network
design and management, European Journal of Operational Research, Vol. 144, pp. 629-45.
Juran, J.M. (1979), Quality Control Handbook, 3rd ed., McGraw-Hill, New York, NY.
Lee, H.L. and Billington, C. (1993), Material management in decentralized supply chains,
Operations Research, Vol. 41 No. 5, pp. 835-47.
Melo, M.T., Nickel, S. and Saldanha de Gama, F. (2005), Dynamic multi-commodity capacitated
facility location: a mathematical modeling framework for strategic supply chain
planning, Computers and Operations Research, Vol. 33, pp. 181-208.
Min, H. and Zhou, G. (2002), Supply chain modeling: past, present, and future, Computers and
Industrial Engineering, Vol. 43, pp. 231-49.
Incorporating
the cost of
quality
85
Pyke, D.F. and Cohen, M.A. (1993), Performance characteristics of stochastic integrated
production-distribution systems, European Journal of Operational Research, Vol. 68 No. 1,
pp. 23-48.
Roden, S. and Dale, B.G. (2001), Quality costing in a small engineering company: issues and
difculties, The TQM Magazine, Vol. 13 No. 6, pp. 389-99.
Santoso, T., Ahmed, S., Goetschalckx, M. and Shapiro, A. (2005), A stochastic programming
approach for supply chain network design under uncertainty, European Journal of
Operational Research, Vol. 167, pp. 96-115.
Svoronos, A. and Zipkin, P. (1991), Evaluation of one-for-one replenishment policies for
multi-echelon inventory systems, Management Science, Vol. 37 No. 1, pp. 68-83.
Towill, D.R. (1991), Supply chain dynamics, International Journal of Computer Integrated
Manufacturing, Vol. 4 No. 4, pp. 197-208.
Vidal, C.J. and Goetschalck, M. (2000), Modeling the effect of uncertainties on global logistics
systems, Journal of Business Logistics, Vol. 21 No. 1, pp. 95-121.
Voudouris, V.T. (1996), Mathematical programming techniques to debottleneck the supply
chain of ne chemical industries, Computers and Chemical Engineering, Vol. 20,
pp. s1269-74.
Further reading
Geoffrion, A. and Graves, G. (1974), Multi-commodity distribution system design by Benders
decomposition, Management Science, Vol. 29 No. 5, pp. 822-44.
Newhart, D.D., Stott, K.L. and Vasko, F.J. (1993), Consolidating product sizes to minimize
inventory levels for a multi-stage production and distribution systems, Journal of the
Operational Research Society, Vol. 44 No. 7, pp. 637-44.
Corresponding author
Amar Ramudhin can be contacted at: amar.ramudhin@etsmtl.ca
JQME
14,1
86
To purchase reprints of this article please e-mail: reprints@emeraldinsight.com
Or visit our web site for further details: www.emeraldinsight.com/reprints