Professional Documents
Culture Documents
November 5, 2003
Abstract
Singapore Food Industries (SFI) manages majority of the Singapore Ministry of Defence (MINDEF) cookhouses (which are distributed all around the Singapore Island) and provide material and
labor to prepare meals for the servicemen. This concept is known as self-contained Commercialized Cook-houses (CCH). The current mode of operation adopted by SFI is self-contained On-site
Cooking, i.e., all the meal requirement of a certain CCH is fulfilled by the CCH itself. Since there
is minimal amount of centralization in the current process, the case is to study the viability of a
new centralized operation mode and to come up with an optimal network design using optimization
techniques.
In the case we look into two levels of centralization. The first level is the centralization of the
raw material processing while the second level of centralization takes place in the preparation of the
meals. We model the problem as a Mixed Integer Programming (MIP) problem and look into two
different representations of the cost function. The first one is the step function representation while
the second one is a piecewise linear representation. The results obtained with the two representation
are slightly different, with piecewise linear representation achieving a slightly better result. At the
end of the study we show that centralization can make as much as twelve percent saving in the
total operation cost compared to the current operation.
1. Introduction
Singapore Food Industries was incorporated in 1973. It has since grown to become a leading
integrated food company in Singapore with international presence in UK, China and Australia. SFI
focuses on the 3 main strategic business areas, namely (1)Food Preparation, Manufacturing and Processing;(2) Food Distribution; (3) Hog Auction and Abattoir.
Food Preparation is a core business for SFI. As a leading institutional caterer in Singapore, SFI
operates more than 65 diverse food service outlets located in factories, schools, hospitals, and Singapore
Armed Forces camps. SFI is also very experienced in on-site catering to customers with special diets
such as hospital catering and the supply of ready-to-eat packaged meals. SFI also offers large-scale onsite and off-site catering services for formal and informal functions, by using its modern central kitchen
facilities and a large pool of experienced catering personnel and logistics support. SFI manufactures
ready meals in retort-pouch suitable for Singapore Armed Forces consumption. The company also
manufactures the Instanchef and Farmpride Heat & Serve ready meals for the supermarkets and
food service sectors.
One of the traditional strengths of SFI has been its ability to import, process and distribute meats,
with its large cold room facilities and refrigerated vehicle fleet. Its Cold Chain system guarantees the
freshest and most hygienic meats at all times. This system ensures that meats are kept at controlled
temperatures from slaughter or import to the consumer, including the transportation phase. With
modern and extensive warehousing and cold storage facilities, and a large specialized fleet of vehicles
including refrigerated vans, SFI has been in the forefront of the Food Logistics and has extensive
experience in the sourcing and import of fresh fruits, chilled and frozen meats/seafood, and other
food products. These are distributed to a wide customer base that includes supermarkets, hotels, ship
chandlers, and some government institutions. SFI runs one of the most efficient logistics operations from the unloading of food items, whether frozen, fresh, or shelf-stable to the re-packing, uploading
and distribution of these items. In addition to distribution in Singapore, SFI also exports to countries
in the Asia Pacific region including Malaysia, Indonesia, Hong Kong, Brunei, Cambodia and the
Maldives.
SFI now manages majority (about 90%) food service of the camps under MINDEF while its
competitor NTUC Foodfare manages the remainder. The current contract of SFI is expiring and
tender for a new contract is opened. With the food industry flourishing, it is expected that there will
be many interested players for this pie. Thus, it is important for SFI to be able to come up with a
good price to maintain competitiveness.
the cost of current mode of operation is quite high,it will be difficult for SFI to come out with a better
bid for the tender to remain competitive. This case looks into a new mode of operation which will
reduce cost without compromising the quality of service and standard of hygiene.
No
1
2
3
4
5
6
7
8
9
10
11
No
12
13
14
15
16
17
18
19
20
21
22
No
23
24
25
26
27
28
29
30
31
32
33
Note that these locations are not the exact locations as these are classified information.
of operation, the reduction will not be significant enough to raise the competitiveness of SFI to the
desired level. The current mode of operation adopted by SFI is what is called On-site Cooking, i.e., all
the meal requirement of a certain camp is fulfilled by the camp itself.There is not much centralization
in the current mode of operation and there is much overlapping of labor. Also, there is much wastage
of utility due to under-utilization of the existing facilities. For example, at present many of the chilling
facilities are heavily under-utilized. The electricity consumption of the chilling facilities only depend
on its size and not on the actual amount of raw material placed in it. If a camps meal requirement
is not very big, the amount of raw material using up the chilling facilities resource may be far below
its capacity, thereby leading to considerable wastage of electricity usage.
strategically chosen to act as Regional Kitchens (RK). These Regional Kitchens will prepare meals to
fulfill the meal requirements of the Dinning Halls(DH) assigned to it as well as its own meal requirements. At this level, the main saving comes from the sharing of resources, mainly labor. Once the
meals are cooked, they are delivered to the Dining Halls in specialize vehicles. Note that the S.A.CCH
in Figure 3 represents Stand Alone CCH, which are special cases of the Regional Kitchens. S.A.CCH
are Regional Kitchens that only cook the meal requirement of their own Dinning Halls, i.e., there are
no other DHs are assigned to S.A.CCHs.
This mode of operation(Figure 3)will introduce a new expense though,i.e., distribution cost, which
involves the transportation of the raw materials from the Central Processing Kitchen to the Regional
Kitchens plus the transportation of the cooked meals from the Regional Kitchens to the DHs assigned
to them. But this additional transportation expense is expected to be significantly lower than the
saving achieved through the centralization of the processes.The objective of this case is then to verify
the viability of this new concept of Centralized Food Processing. Among the 33 CCHs, we will identify
the CCHs to be operated as Regional Kitchens and also decide which of the remaining CCHs are to
be served by which Regional Kitchen.
Number
of camps
Double
Kitchen
26
Muslim
Kitchen only
3
Non-Muslim
Kitchen only
19
No-Kitchen
At All
2
Total
50
USGC in Table 1 are located within the same compound and if any other camp supplies meal to all
three of them, only one vehicle is required for a certain type of meal.
Number
of camps
Non-Muslim &
Muslim meal
34
Non-muslim
meal only
13
Muslim
meal only
3
Total
50
economical to buy and put the machines at one centralized location, rather than one each at the
Regional Kitchens end.
However, from figure 4 one can easily conclude that the allocation of Regional kitchens and Dining
Halls,in this configuration, is almost solely determined by the location and capacity of the CCHs.
We also observe that the size of the clusters is relatively small. If we count the total number of
operating kitchen in the above arrangement, it has a total of 20 kitchens. Only 8 CCHs operate as
double kitchens and 12 CCHs operate as single kitchens. In term of percentage, less than 50% of
the CCHs act as double kitchens. This prompt us to question why more kitchens do not operate as
double kitchens since in total 26 CCHs have the capabilities to operate as double kitchens. Intuitively,
a double kitchen is cheaper to operate as compared to operating two single kitchens because of the
sharing of activities and resources. This question leads us to the importance of defining the right cost
function for different types of cooking mode. We will talk more about the derivation of cost function
and estimation in the model.
10
11
The processed vegetables and meat will be packaged to preserve freshness, and kept in the chiller
room(Figure 9). They will be ready to be transported to the Regional Kitchens.
The above process is expected to be completed within 16 hours. Then the transportation vehicles
at the Central Processing Kitchen will not go to all the Dining Halls around the island, but only to
the Regional Kitchens, which number will be much less the number of Dining Halls(Figure 7). Both
the refrigerated and non-refrigerated trucks will load the processed vegetables and meat and travel to
the respective Regional Kitchens using fixed routes.
The Regional Kitchens will receive the processed vegetables and meat and store them in the chiller
room. The Regional Kitchens will prepare the days meal with the supplies received. When the
cooking is done, the food will be distributed to the respective Dining Halls.
The Dining Halls will receive and unload the food from their respective Regional Kitchens. Together
with the rice and soup that the individual Dining Hall has cooked, the food is set up at the serving
counter and the food is ready to be served.
12
Number of meal
25
50
75
100
62.5+62.5
75+75
87.5+87.5
100+100
75+75+75
83+83+83
92+92+92
100+100+100
Number of pan
1
1
1
1
2
2
2
2
3
3
3
3
Number of boiler
1
1
1
1
20
2
2
2
3
3
3
3
Cooking Time
10
17.5
22.5
25
10
22.5
24.5
25
22.5
23.5
24.5
25
Gas cost
1000
1750
2250
2500
4000
4500
4800
5000
6750
7000
7350
7500
13
Electricity consumption is much affected by the type of equipment, especially refrigerator owned by
each kitchen. Some kitchen have old refrigerators while others have new refrigerators which consume
less electricity. Therefore, electricity consumption is unique to each kitchen and is assumed to be
independent of the amount of meal cooked.
Same as in the case of electricity, the water consumption is not much affected by the amount of
meal cooked. Based on the information provided by the SFI management, the inconsistency in water
consumption is due to the cleaning service and water leaking. In this project, electricity and water
consumption is treated as a fixed cost, unique to each camp.
The above analysis can be directly used to forecast the utility cost for single kitchen.For double
kitchen and dining hall however, some modifications need to be made.The fixed cost for double kitchen
is assumed to be the same as that of a single kitchen because there is a sharing of usage between muslim
and non-muslim kitchens. For dining hall, the fixed cost is approximated to be 50% of the single kitchen
one, because only light cooking need to be done in the Dinning Halls. For electricity and water in the
double kitchen, the same cost data as a single kitchen is used, because it is assumed that electricity
and water usage are independent of the amount of meal cooked. On the other hand, the electricity
and water fixed cost is again approximated to be 50% of the single kitchen one.
to describe the labor cost function based on the data given by SFI.
The total operating cost for Regional kitchen is obtained by adding the labor and utility cost
function as shown in Figure 13. This figure relates the number of meal produced and served during
lunch per day, and the monthly cost.
The transportation cost involves the delivery cost of raw materials from CPK to the selected RK
and the delivery cost of cooked meal from RK to the respective DH assigned to the RK. Both types
of delivery consist of a fixed and variable cost. The main contribution of the fixed cost includes driver
allowances, depreciations, monthly maintenance cost and miscellaneous cost whereas fuel cost is the
only variable cost which depend on the distance travelled.
0 otherwise
1 if the muslim kitchen in camp i is operating at level
M
production t
yit =
0
otherwise
0
otherwise
15
In the formulation, 3 levels of production means that for any operating single kitchen (muslim or
non-muslim), there are 3 possible levels at which meals can be produced. For example, if the maximum
meal production of a certain camp is Pmax , then level 1 would be Pmax /3 , level 2 would be 2Pmax /3
and level 3 would be Pmax . Below is the description on how the sample points are taken.In figure 14,
Cf ixed represents the cost incurred if the kitchen is not operating, in other words, the camp operates
as a DH.Lets say the total number of meal required to be cooked by a particular single kitchen is
denoted by Pcook , then the corresponding cost attributed to this demand requirement would be one
of the following possibilities:
Non-Muslim
kitchen level
1
1
1
2
2
2
3
3
3
Muslim
kitchen level
1
2
3
1
2
3
1
2
3
Double
kitchen level
1
2
3
4
5
6
7
8
9
n P
n
P
i=1 j=1
n
P
i=1
N
M M
(cN
ij xij + cij xij ) +
hi (1 yi ) +
n
P
i=1
n P
3
P
i=1 t=1
N + f M yM )
(fitN yit
it it
n P
9
P
i=1 k=1
D
sik yik
gi yi CV
The objective function minimizes the weighted sum of the following cost components:
17
1. Delivery cost of cooked meal from Regional Kitchen to Dining Hall, given by
n P
n
P
i=1 j=1
n P
3
P
i=1 t=1
N + f M yM )
(fitN yit
it it
N
M M
(cN
ij xij + cij xij ).
n P
9
P
i=1 k=1
D +
sik yik
N
and 1 respectively, then, due to the constraints which will be explained later, variables y1 , y13
M
D
, y11 , and y17 will be active (notice from table 5 that non-muslim kitchen level 3 and muslim
kitchen level 1 corresponds to double kitchen level 7). It means that the operating cost is equal
N + fM s .
to the sum of muslim and non-muslim kitchen minus the saving obtained, or f13
17
11
M
On the other hand, if camp 1 only operates its muslim kitchen at level 1, then only y1 and y11
are active. Finally, if none of the kitchens in camp 1 is operating, then no variable is active. It
means that only the term h1 is left.
n
P
i=1
gi yi .
4. Reduction due to the unnecessary cost, when there are deliveries to CCHs located in the same
compound. This is given by the the term CV . Suppose that CCHs A, B, and C are located in
one compound, then only one vehicle is needed to deliver to those CCHs instead of three. Since
in component 1 we have included 3 vehicles, we should eliminate the cost of 2 vehicles. This
condition will be explained more in the constraint 7 subsection.
2.3.5. Constraints
1) Saving Constraint:
Since every CCHs is treated as double kitchen, we would like to have a relationship between yD
ik ,
N
M
yit , and yit . If non-muslim kitchen operates at level 1 and the muslim kitchen operates at level 1,
the corresponding double kitchen operates at level 1 as well.Similarly, if both non-muslim and muslim
kitchen operate at level 3, the corresponding double kitchen operates at level 9.The relationship among
variables is represented by the following Boolean expression:
D
N
M
yi1
= yi1
AN D yi1
..
.
D
N
M
yi9
= yi3
AN D yi3
..
.
D
N
M
D
2yi9
yi3
+ yi3
yi9
+ 1 i I3
18
Since yi is active if at least a kitchen is operating, then in all CCHs except CCHs which do not
have any kitchen facility at all the following will hold
yi =
3
X
N
(yit
M
yit
)
t=1
9
X
D
yik
i I1 I2 I3
k=1
Since only one operating mode can exist at any one time for CCHs in sets I1 , I2 , and I3 , then the
following constraint should be applied
9
X
D
yik
1 i I1 I2 I3
k=1
3
X
N
yit
1 i I1 I2 I3
t=1
3
X
M
yit
1 i I1 I2 I3
t=1
, t {1, 2, 3}
Similarly, for those CCHs that only have muslim kitchen, the following constraints will apply
D
yik
= 0 i I2
, k {1, ..., 9}
N
yit
= 0 i I2
, t {1, 2, 3}
Finally, for those CCHs that do not have any kitchen at all,
D
yik
= 0 i I4
, k {1, ..., 9}
N
M
yit
= 0 , yit
= 0 i I4
M
xN
ij = 0 , xij = 0 i I4
, t {1, 2, 3}
, j I1 I2 I3 I4
j J3
xM
ij = 0 i I
j J4
N
aN
j xij
3
X
N
bN
it yit
i I1 I3
t=1
jJ1
The above constraint will not be correct if we generalize to the case of zero non-muslim demand (case
N
when J1 is changed to J1 U J3 ). Because if xN
ij is equal to zero, there will be no conclusion about yit .
Similarly, for non-zero muslim meal demand requirement
X
M
aM
j xij
3
X
M
bM
it yit
i I2 I3
t=1
jJ2
xM
ij = 1 j J2
iI2 I3
5) Self-Serving Constraint:
This constraint will force each Regional Kitchens to satisfy their own demand for a particular meal
that they produce.
3
X
N
N
xii =
yit
, i I1 I2 I3
t=1
xM
ii
3
X
M
yit
, i I1 I2 I3
t=1
xM
tabu1(i),tabu2(i) = 0 i tabu index
In addition, it is important to avoid distance close to 20 km between RK to DH. Therefore, a penalty
as big as 1000 times the usual cost would be impose for distance between 15 km and 20 km.
7) Proximity Constraint:
As mentioned earlier, some CCHs are located within one compound and the distance among them
is only about 500 meters. To model this, suppose camp A and B is in the same compound. If the
demand for a particular meal for both CCHs are fulfilled by RK outside the compound, then, this RK
should only send one vehicle for both CCHs. In addition, CCHs A and B should be fulfilled by the
same RK. To model this, we need to put constraint
N
xN
iA = xiB
M
xM
iA = xiB
Then, since only one vehicle is needed to deliver the food, we should add the following component in
the objective function to reduce the number of vehicle needed
N
M M
N N
M M
(cN
iA xiA ciA xiA )or(ciB xiB ciB xiB )
Now, if lets say there are 3 CCHs A,B, and C in one compound. Then, the constraint should be
changed to
N
N
xN
iA = xiB = xiC
M
M
xM
iA = xiB = xiC
k
P
i=1
21
i ai
k
P
i=1
k
P
i=1
i f (ai )
To model the additional constraint that at most two consecutive coefficients i can be nonzero,
we define binary variable zi , i = 1, ..., k 1, which is equal to 1 if ai x < ai+1 , and zero otherwise.
Then we can model the minimization problem involving piecewise linear cost function as the following
mixed integer programming:
minimize
s.t
k
P
i=1
k
P
i f (ai )
i = 1
l=1
1 z1 ,
i zi1 + zi,
k zk1 ,
k
P
zi = 1,
i = 2,..,k - 1
i=1
zi {0, 1}
22
23
set
set
set
set
set
set
set
set
set
of
of
of
of
of
of
of
of
of
all
all
all
all
all
all
all
all
all
CCHs
Single Muslim Kitchens
Single Non-Muslim Kitchens
Double Kitchens
CCHs with demand for Muslim food
CCHs with demand for Non-Muslim food
CCHs with zero demand for Muslim food
CCHs with zero demand for Non-Muslim food
CCHs with no kitchen
2.4.4. Parameters:
aN
j :
aM
j :
bN
i :
bM
i :
D
bi :
hi :
gi :
Cij :
index tabu:
tabuij :
level :
B:
fD
il :
aD
il :
the cost of supplying the raw materials from the Central Kitchen
(C.K.)to the ith CCH
Cost of delivering meals from the ith camp to the jth camp
Number of kitchens that are further than 20km apart from each
other
contains index of CCHs that are tabu
number of pieces in the piecewise linear cost funtion
arbitrary big number
Cost of running the ith CCH as a double kitchen at level l .
Production level of the ith Double Kitchen at level l.
1
if the ith CCH is choosen to be a Regional Kitchen
yi =
0
otherwise
yiN =
yiM =
yiD =
(
xN
ij
otherwise
if the ith CCH is choosen to be a Regional Kitchen to serve
Muslim meal
otherwise
if the ith CCH is choosen to be a Regional Kitchen
to serve both Muslim and Non-Muslim meal
otherwise
1
0
if the ith Regional Kitchen serves Non-Muslim meal to the jth CCH
otherwise
1
0
if the ith Regional Kitchen serves Muslim meal to the jth CCH
otherwise
(
xM
ij =
D
il :
iN :
iM :
25
zilD :
pD
i :
pN
i :
pM
i :
cij xN
ij +
PP
iI jI
cij xM
ij +
P
iI
gi yi
The transportation cost involves the cost of transporting the processed raw materials from the
Central Processing Kitchen to the Regional Kitchens and the cost of transporting the cooked meals
from the Regional Kitchens to the respective Dinning Halls assigned to them.
(ii) Operating cost
P
M
(7.5928 pN
i + 5.0324 ai + 24855)
iI
P
N
+
(15.069 pM
i + 17.377 ai + 12411)
PiI
D
(7.5928 iN + (5.0324 aM
i + 24855)yi )
iI
P
D
(15.069 iM + (17.377 aN
i + 12411)yi )
iI
P
iI
D (1
fi1
P level
P
iI l=1
P
yiD )
iI
D
D
il fil +
hj (1 yj )
P
24855(1 yiN )
12411(1 yiM )
iI
iI
The overall operating cost comprise of the operating cost of the Single Kitchens, the operating of the
Double Kitchens and the operating cost of the Dinning Halls. As mentioned before, the operating cost
M
of the Single Kitchens are modelled as two linear functions z = 7.5928 pN
i + 5.0324 ai + 24855(NM)
M
N
and z = 15.069 pi +17.377 ai +12411(M), while the operating cost of the Double Kitchen is modelled
as a two piece linear function. The operating cost of the Dinning Halls only depend on the amount of
meals to be served, which is fixed for a certain Dinning Hall. Hence the operating cost of a Dinning
Hall is just a constant term.
Since the operating cost of a Single Kitchen and a Double Kitchen is completely different, we
have to make sure that when a CCH is operating as a Double Kitchen, the operating cost of its
constituent Single Kitchens does not show up in the objective function. That means that if the ith
M
M
N
CCH is a Double Kitchen, then the terms 7.5928 pN
i +5.0324 ai +24855 and 15.069 pi +17.377 ai +
12411 corresponding to the two constituent Single Kitchens of that Double Kitchen has to go to zero.
26
Although we have defined binary variables yiD to indicate whether a CCH is a Double Kitchen or
not, we cannot use it to drive the above two terms to zero, as this will make the objective function
non-linear. Therefore, we define new variables iN and iM which are zero if the ith CCH is not a
M
th CCH is a Double Kitchen.
Double Kitchen, or equal to pN
i and pi respectively if the i
Also, the operating costP
functions do notPgo to zero when P
the amount of meal cooked is zero.
D (1 y D )
N)
Hence, add the last term,
fi1
24855(1
y
12411(1 yiM ), to make sure that
i
i
iI
iI
iI
cij xN
ij +
PP
cij xM
ij +
gi yi
iI jI
iI jI
iI
P
P
M
N
+
(7.5928 pN
(15.069 pM
i + 5.0324 ai + 24855) +
i + 17.377 ai + 12411)
iI
iI P
P
D)
D
(7.5928 iN + (5.0324 aM
+
24855)y
(15.069 iM + (17.377 aN
i
i
i + 12411)yi )
iI
P level
P
iI l=1
iI
D
D
il fil +
P
iI
hj (1 yj )
P
iI
D (1 y D )
fi1
i
P
iI
24855(1 yiN )
P
iI
12411(1 yiM )
2.4.7. Constraints
Constraints 1 ,2 and 3 enforce the relationship between the binary variables yiN , yiM , yiD and yi . Binary
variable yiD indicates whether a CCH is Double Kitchen or not, and it can be equal to one only if both
of its constituent Kitchens are open at the same time, that is, yiD can be equal to one only if both yiN
and yiM are equal to one. Similarly yi indicates whether a CCH is a Regional Kitchen or a Dinning
Hall and it is equal to one if any one of yiN , yiM or yiD is equal to one.
yiN + yiM
yiN
yiM
yiD + 1 ,
i I
(1)
2yiD
i I
(2)
i I
(3)
Constraints 4 and 5 are to ensure that demand of each of the CCHs is met and that only one
Kitchen can supply to a certain camp.
X
xN
j Nonzero NM
(4)
ij = 1 ,
iI
xM
ij
= 1,
j Nonzero M
(5)
i I
(6)
i I
(7)
iI
= yiN ,
=
yiM
27
If a CCH has no demand for Non-Muslim food, it cannot be assigned to a Non-Muslim R.K. and
vice versa
xN
ij
= 0,
j Zero NM, i I
(8)
xM
ij
= 0,
j Zero M, i I
(9)
Muslim kitchen cannot be opened in a CCH with only Non-Muslim facility and vice versa.
yiN
= 0,
i Single M
(10)
yiM
= 0,
i Single NM
(11)
To ensure that a camp can supply another camp only if it is a Regional Kitchen.
xN
ij
xM
ij
yiN ,
yiM
i I, j Nonzero NM
i I, j Nonzero M
(12)
(13)
i Single M
(14)
yiD
i Single NM
(15)
yiD = 0 ,
i No Kitchen
(16)
yiD
j No Kithcen
(17)
= 0,
Single Muslim kitchens cannot serve Non-Muslim food and vice versa
yiN
= 0,
i Single M
(18)
yiM
= 0,
i Single NM
(19)
= 0
i Single NM
(20)
xM
tabui1 ,tabui2
= 0
i Single M
(21)
If several CCHs, say j1 and j2 are within the same compound, the assignment to those CCHs must
N
N
be the same, that is xN
ij1 = xij2 . In the objective function we deduct the cij1 xij1 term which reflects
the fact that if the CCHs in that complex are active, we only need one vehicle to make the delivery
for all the CCHs within the compound.
Integrality constraints:
xN
ij
{ 0,1}
i I,j I
(22)
xM
ij
yiN
yiM
{ 0,1}
i I,j I
(23)
{ 0,1}
i I
(24)
{ 0,1}
i I
(25)
28
yiM
xM
ij ,
i I, j I
(27)
jI
N
bN
i yi ,
pM
i
M
bM
i yi
i I
(28)
i I
(29)
When the kitchen is operating as a Double Kitchen, we have to make sure that the terms corresponding to the Single Kitchen Costs in the objective function go to zero. We also have to ascertain
that it is operating at the same production level as the sum of the production levels of the single
kitchens associated with the Double Kitchen. In summary, we want the following relation to hold
true:
yiD = 0 ilN = ilM = 0
M
M
yiD = 1 ilN = pN
i , il = pi
B yiD ,
B
ilN pN
i
ilM pM
i
N
pi ilN
M
pM
i il
yiD
B(1 yiD ) ,
B(1
B(1
B(1
yiD )
yiD )
yiD )
i I, l = 1,..,level
(30)
i I, l = 1,..,level
(31)
i I, l = 1,..,level
(32)
i I, l = 1,..,level
(33)
i I, l = 1,..,level
(34)
i I, l = 1,..,level
(35)
D
= 1,
il
i I
(36)
l=1
D
zilD ,
il
D
il
D
il
level
X
D
zil1
D
zil1
+
,
zilD = 1 ,
zilD
i I, l = 1
(37)
i I, l = 2,..,level - 1
(38)
i I, l = level
(39)
i I
l=1
29
(40)
level
X
D
D
il ail ,
i I
(41)
l=1
When a CCH is operating as a Double Kitchen, we have to ensure that its production level is equal
N
M
to the sum of the production levels of its two constituent Single Kitchens, i.e., pD
i = (pi + pi ) only
D
when yi = 1 .
N
M
D
pD
i (pi + pi ) B(1 yi ) ,
i I
(42)
(pN
i
i I
(43)
pM
i )
pD
i
B(1
yiD )
i I
(44)
pN
j
j I
(45)
pM
j
j I
(46)
jI
M
aM
j xij
jI
30
Proposal
SFI
SMA
Number of
Double kitchen
8
10
Proposal
SFI
SMA
Proposal
SFI
SMA
Number of
Single NM kitchen
7
5
Number of
Non-Muslim truck
27
24
Utility
Cost(S$)
356,653
330,114
Labor
Cost(S$)
1,773,287
1,622,563
Number of
Single M kitchen
5
1
Number of
Muslim truck
21
20
Transportation
Cost(S$)
116,588
108,207
Number of
Dining Hall
31
34
Total
truck
48
44
Total
Cost(S$)
2,246,531
2,060,884
95% demand
Single NM Single M
UMIA
UARZ
UHDZ
USTZ
UPLZ
UDBZ
USGB
USWA
USBZ
ULCA
UBPZ
UJRB
Double
UCGA
UAQA
UKJB
UPSB
UMJZ
UKTA
UNSA
USWB
UTAZ
UJRB
10
Double
UCGA
UAQA
UKJB
UPSB
UMJZ
UKTA
UNSA
USWB
UTAZ
UTGA
10
31
The demand requirement calculated from 95 percentile data is higher than from the 85 percentile
one. It is observed that at 95 percentile demand, more Regional Kitchens need to be operated. For
single Non-Muslim kitchen, there is no difference in the choice of Regional Kitchens. The two demand
requirements require exactly the same set of single Non-Muslim kitchens. On the other hand, at 95
percentile demand six more single Muslim kitchens are required. In addition to UARZ, it need to
open USTZ, UDBZ, USWA, ULCA, UBPZ and UJRB. As for Double Kitchens, only UJRB need to
be converted into a single Muslim Kitchen and open a new Double Kitchen, UTGA.
32
33
Number of
Double kitchen
8
8
Proposal
SFI
SMA
Proposal
SFI
SMA
Number of
Single NM kitchen
7
5
Number of
Non-Muslim truck
27
25
Utility
Cost(S$)
356,653
341,818
Labor
Cost(S$)
1,773,287
1,535,738
Number of
Single M kitchen
5
14
Number of
Muslim truck
21
11
Transportation
Cost(S$)
116,588
88,502
Number of
Dining Hall
31
23
Total
truck
48
36
Total
Cost(S$)
2,246,531
1,966,058
Please note that the tabulated Cost estimates have been altered due to the confidential nature of the data
34
The Regional Kitchen configuration at 85 percentile and 95 percentile demand data is shown in
Table 9.
85% demand
Single Nm Single M
USGB
ULCA
UMIA
USTZ
UHDZ
UBPZ
UPLZ
UJRB
USBZ
UMJZ
UCPZ
UCMZ
USLB
UKTA
UDBZ
USWA
UGCZ
UAQA
UTGA
UTNZ
6
14
95% demand
Single Nm Single M
USGB
ULCA
UMIA
USTZ
UHDZ
UBPZ
UPLZ
UJRB
USBZ
UMJZ
UCPZ
UCMZ
USLB
UKTA
UDBZ
USWB
UGCZ
UAQA
UTGA
UTNZ
Double
UKJB
UPSB
UNSA
USWB
UCGA
UARZ
UTAZ
14
Double
UKJB
UPSB
UNSA
USWA
UCGA
UARZ
UTAZ
It is observed that most of the Kitchen types remain unchanged. In both configurations, there are
6 Single Non-Muslim kitchens, 14 Single Muslim Kitchens and 7 Double Kitchens. Only USWA which
was originally a Single Muslim kitchen is operating as a Double Kitchen in the new configuration while
USWB which was originally a Double Kitchen is a Single Muslim Kitchen in the new configuration.
There are some changes in the Dinning Hall assignment though. The cluster sizes are smaller in the
new configuration. This is to be expected, because with bigger demands, the number of camps a
Kitchen can serve cannot be as high as with smaller demands. It is found that most of the R.K.s
whose Dinning Hall assignments have changed are those originally operating at near peak capacities.
4. Conclusion
In this case, a location and assignment problem of the Commercialized Cook House (CCH) was
looked into. In order to save the total cookhouse operating cost, the Singapore Food Industries (SFI)
decided to restructure its operation from On-site Processing to Centralized Food Processing. The
focus of the project was to come up with a centralized mode of operation without compromising the
existing quality of service. Two different models were developed in this project, the step cost function
model and the piecewise linear model.
The most challenging part of this project was to determine the cost parameters of the model. Two
cost components were considered, the operating cost and the transportation cost. For operating cost,
36
it was found out from regression analysis that only gas usage and labor cost have relationships with
production and serving volume, and thus only the two of them are considered in the operating cost
model. For electricity and water, the cost was treated as a monthly fixed cost since they showed a
poor relationship with the production volume. In the model, the transportation cost was modelled as
a sum of a fixed cost and distance-dependent cost. The fixed cost was mainly contributed by vehicle
depreciation, maintenance, and driver allowance. As for distance-dependent cost, only fuel usage was
considered.
The effort for data gathering, data validation, cost estimation and error detection was substantial
and involved close corporation with the SFI managers and staffs. Furthermore, the multitude of
parameters as well as uncertainties regarding the demand requirements made it necessary to perform
some statistical analysis. Finally the two models were implemented using OPL studio. The result
showed that a network configurations which can reduce the total cost by about 10%, compared to the
network configuration proposed by SFI, is theoretically possible.
Below it briefly compares the three different network configurations, one proposed by SFI and the
other two obtained through the two models developed in the project.
The main attributes of SFI network configuration are the small cluster sizes and the relatively
small number of Double Kitchens. As a result, the total number of operating kitchens is relatively big.
The Dinning Hall assignments are also relatively simple; the Regional Kitchens only supply to nearby
Dinning Halls. As a result, the daily operations are less complicated and hence easier to manage.
In both of the configuration obtained with the two new models, the cluster sizes are relatively big,
with some of the clusters containing 10 or more CCHs. The Dinning Hall assignments are also quite
complicated. Consequently, the daily operations may be more complicated and harder to manage.
With step cost function model, the double kitchen assignment seems to be the most dominating
type, followed by Single Non-Muslim and Muslim kitchen. On the other hand, the most dominating
type in the result of piece wise linear model is single Muslim kitchen, followed by double kitchen
and single Non-Muslim kitchen. In the original cost function derived from the regression analysis,
the Muslim kitchen has a bigger slope but a lower y-intercept value compared to the Non-Muslim
kitchen. That means, it is more cost effective to operate the Muslim kitchens in smaller production
level. With the step cost function model, the information about the slope and intersection value at a
low level of production could not be captured as accurately as with the piecewise linear model. The
minimum production level of a kitchen is one-fourth the maximum capacity of that kitchen, and since
the Muslim demands are not that big, one Muslim kitchen can provide meals to quite a large number
of Dinning Halls, which explains the lesser number of Muslim kitchens with this model. On the other
hand, with the piecewise linear model, it is possible to capture the exact level of production. This
results in many of the Muslim kitchens being operated in small production level. Most of the Single
Muslim kitchens in this configuration are actually producing meals for themselves only.
Both of the problem formulations are based on the 85 percentile demand data obtained through
statistical analysis, although the actual demand may be stochastic in nature. The models, as such,
cannot capture this time variant nature of the input demand data. Therefore a possible future work
can be to further refine our model to incorporate this randomness in data.
37