A Scheduling Method in Multi-item Production
Adjusting to Urgent Orders Based on Job Modifiability
Masaki Samejima, Shingo Tamura, Yoshitomo Ikkai, Norihisa Komoda
Graduate School of Information Science and Techonology
Osaka University
Suita, Osaka, Japan
{samejima.masaki, tamura.shingo, ikkai, komoda} @ist.osaka-u.ac.jp
Hitoshi Goshima, Teruyuki Mori
Kometarou, Co. Ltd.
Abstract
The target ofthis research is to adress scheduling prob-
lems in order to ship multi-item orders in a rice milling
process on the same day the order is placed. Milled rice
can be stored as stock for some days and can be shipped
from stock to satisfy orders. The setup time for switch-
ing milled items are needed and work hours in a day
are given in the process. Scheduling is generated from
the producers' order forecast, orders and stock before
milling. When new orders during milling(urgent orders)
come and exceed the work-in-process stock, the schedule
must be modified in order to satisfy the orders. The pur-
pose of this problem is to minimize the number of over
time hours worked, to reduce switching time and to max-
imize throughput. In this problem, because urgent orders
accountfor 90% of the whole, stocks often run out and the
need to mill new items often occurs in multi-items.
Therefore, this research proposes a modifiable schedul-
ing method to minimize the number of switches and over
time hours worked when milling starts in a day and when
urgent orders are received. In scheduling, items to be
milled and the amount to mill are first decided using an
inventory control theory. The milling sequence is decided
using "modifiability ", which is defined by the reductibility
of milling time and the necessity of modification of items
to be milled for modifiable scheduling against urgent or-
ders. In a case where the estimated end time runs over
work hours, milling amounts which has smaller stock-out
probability are reduced in order of increasing.
The proposed method has been applied to a real prob-
lem. It has been confirmed that this method can generate
the solutions 37% better than producer's solutions.
1. Introduction
In a production field, effective scheduling plays an im-
portant role. Such schedules are created by human experts
0-7803-9402-X/05/$20.00 2005 IEEE
and take many hours, but thanks to computer progress,
rapid automated scheduling using a computer has been in-
troduced.
The target of this research is to address scheduling
problems of a multi-item rice milling process. Milled
rice can be stored as stock for some days and can be
shipped from stock to satisfy orders. Ordered items must
be shipped on the same day that the order is placed. When
switching milled items in a milling machine, some setup
time is necessary. Work hours in a day are also given
in the process. Under these constraints, scheduling is
generated from the producers' order forecasts, the orders
and the stock before milling. When new orders during
milling(urgent orders) come and these orders cause an ex-
cess of work-in-process stock, the schedule must be mod-
ified in order to satisfy the urgent orders. The purpose
of this research is to minimize the number of overtime of
hours worked, to reduce switching time and to maximize
throughput.
Currently, producers forecast the main items in ur-
gent orders based on empirical knowledge and adjust the
schedule of stock to make against each order from the
forecast and the current stock. However, many urgent or-
ders still occur that vary from the order forecast. So, when
the amount of urgent orders exceeds amount to mill, it is
necessary to mill repeatedly in order to make up the lack
of stock, and when work-in-process amount to mill exceed
the amount of orders, the excess milling needless items
takes time. Then, the number of switches increases and
overtime occurs.
"Reactive scheduling" or "real-time scheduling", is
where an initial schedule is created in an ideal environ-
ment and is modified by a rule to modify when the en-
vironment changes[3, 4]. In these researches, the meth-
ods of acquiring, comparing and evaluating the rules have
been reported. However, the initial scheduling prior to en-
vironmental changes has not been discussed enough. Be-
cause the productivity of a modified schedule depends on
1107 VOLUME 1
the schedule before modification, a rescheduling method the "order data", and the "order forecast by experts" into
that maintains productivity after modification due to envi- the system. "Order data" consists of the ordered item, the
ronmental changes is necessary. number of items in the order and the time. Since "or-
In this target problem, the amount of orders in 5/6 items der data" is generated by accepting the order, "order data"
of a whole items is too few to hold much stock. There- cannot be included in the schedule before the order is ac-
fore, the schedule of items to be milled and amount to cepted. The order forecast consists of the following two
mill has to be adjusted to accommodate urgent orders, in factors:
order to effectively mill in a limited operational capabil-
ity. In this research, the proposed modifiable scheduling
Realtime Received Milling Process
method can effectively minimize the number of switches
--* Milled Rice o-Shipped
and the amount of overtime hours worked when urgent or- Product
| ProductA:50kg L U
Rce
- = Product A
ders cause work-in-process stock. Orders

For scheduling, it is necessary to decide what items before start Order Forecast

to mill, the amount to be milled and the milling se-

quence. The amount of milled items required to be in PProrcer
stock is decided using an inventory control theory[l, 2, 5]
E5~
--&-Generate Urent Orderi UrentOre

that prevents stock-out. The items included in the safety

stock(enough stock to prevent stock-out) are decided by
the experts' order forecast and have to be milled to boost
the stock up to the limited stock decided by forecast.
When the sequence of items to be milled is decided, at-
tention must be focused on reducing the amount of milling
time and making any necessary modifications to the items,
to address urgent orders late in the working day. From
these two characteristics, "modifiability" is defined and
the milling sequence is decided. The items that has
smaller modifiability are milled earlier.In a case where the
end time of the schedule is after the given work hours,
amount to mill that has a larger stock-out margin is re-
duced.
2. A Scheduling Problem in Multi-item Pro-
duction
2.1. Outline of the problem
The target of this research is to solve a scheduling prob-
lem whereby multi-items in a rice milling process can be
shipped on the same day that orders are placed(Figure 1).
Milled rice can be held in stock for some days and can
be used to fill orders to be shipped from the stock, but ex-
pired stock must be discarded. The setup time is needed to
switch milled items and the work hours in a day is given in
the process. The initial schedule is generated from the ex-
perts' order forecast, the outstanding orders and the milled
items already in stock. When new orders arrived during
milling(urgent orders) which exceed the work-in-process
stocks(stock+amount of milling item), the schedule has to
be modified in order to complete the orders. The evalua-
tion factors of this problem are to minimize the number of
overtime of hours worked and the number of machine set-
ting switches, and to maximize throughput. The purpose
of this problem is to generate highly-evaluated schedules
that satisfy these constraints as well as ensuring that all
orders received in a day are shipped on the same day they
were placed.
The experts can forecast the main items that make up
the majority of orders. They then input the "stock data",
Figure 1. Outline of the problem
* The average number of main items in the order fore-
cast
Experts can forecast the number of main items con-
tained in the weeks order to 1/6 of the whole items
using their expert knowledge and the past order his-
tory. Therefore, the average amount in the weeks is
input.
* The average number of total orders
Experts can forecast the amount of orders theat will
occur in a day from the average total amount of past
orders. Therefore, an average total amount of orders
is input.
2.2. Constraints
Scheduling for the mill has the following constraints:
* Order constraints
All orders received in a day must be shipped the same
day.
* Constraints regarding the unit amounts to be milled
The amount to mill has to be multiples of a given unit
amount.
* Constraints regarding the probability of stock expir-
ing
The expired stock rate has to be below a given pro-
portion.
2.3. Evaluation factors
The target problem for the evaluation is the last sched-
ule for each day of the week. There are three evaluation
factors:

1108 VOLUME 1
 1. Minimizing overtime(T) forecast. For this reason, problems are solved by modify-
To minimize the number of overtime hours worked ing the schedule. Therefore, it is necessary to modify both
in a week. the items to be milled and the amount to mill in the sched-
ule to accommodate urgent orders. The milling sequence
2. Minimizing the number of switches(N) of items must be contrived, because the schedule can only
To minimize the number of switches made. Each be modified after urgent orders are received.
switch takes ten minutes and requires labor.
3. A Scheduling Method Based on Job Modi-
3. Maximizing throughput(S)
fiability
To maximize the throughput amount of the mill per
hour.
3.1. Approach
The following evaluation function is defined as the In order to cope with the difficulties mentioned in sec-
weighted(wj) sum of the above three evaluation factors tion 2.4, the scheduling method must respond to urgent
in this problem. orders in order to avoid stock-out and overtime. There-
fore, both the items to be milled and the amount to be
evaluation function =-w1 x T - w2 x N + W3 X S milled are decided to avoid stock-out, and modifications
are then made to avoid overtime. The issue in repeatedly
The purpose of scheduling is to maximize the value of modifications is that the current schedule must be effec-
the evaluation against the executed schedules. tively modifiable. In this paper, a method for deciding
the milling sequence in a modifiable schedule is proposed.
2.4. Characteristics of this problem The method for deciding items to be milled, amount to be
In this problem, the following problems are caused by milled, and the milling sequence is as followings:
many urgent orders, which account for 90% of the total
number of orders(Figure 2): 1. Deciding what items to be milled and amount of mill

Customer:P Many This method decreases the incidence of stock-out

Product A: O00kg using an inventory control[l, 2, 5] based on stock-
Product B:50kg to-make. In this problem, it is difficult to estimate
Start End
the order amounts of each items from the past order
amounts, because urgent orders often occur. There-
C AD
Modify
F E AB7 .... l 2 _
Over-I time fore, the main items are milled to maintain stock lev-
nXwg-.Xn0W=FE.zr^lg2WI2gRg=~ ~ ~~~~~ ~~~~~ ~~~~~ ~~~~~ ~~. .~. .. .. . els using order forecast, while sub items are milled
" TL TI to make up for amount of a stock-out.
By urgent orders, amount to mill It is necessary to mill the same
is over amount of orders. in order to make up lack of the stock. 2. Deciding the process sequence of items to be milled
Overtime occurs. The number of switches increases. As mentioned in section 2.4, the schedule can't
Figure 2. Characteristics of this problem be modified before the incidence of urgent orders.
Therefore, "modifiability" is defined by the re-
ductibility of milling time and necessity of making
modifications. In deciding the process sequence of
* The increase of repeatedly switching items to be items to be milled, an item with the larger modifia-
milled. bility is milled later in order to generate a modifiable
When the amount of urgent orders to be filled ex- schedule.
ceeds the amount of milling that was based on the
order forecast, it is necessary to repeatedly mill the 3. Reducing item to be milled and the amount to be
same item to make up for the lack of stock. milled
* The excess of hours worked overtime milling items Sometimes, the amount of milling required for stock-
for urgent orders. to-make is great and more milled items are added be-
cause of urgent orders. Therefore, the items to be
When the amount of urgent orders is lower than milled and amount to be milled are reduced in order
amount to mill, unnecessary milling over urgent or- to decrease overtime and the number of switches and
ders on the day that the orders occur requires time. to minimize incidence of stock-outs.
Although it is necessary to generate a schedule which
prevents a lack of stock or an excess of the amount to mill, In the following section, we describe the details of this
it is difficult to improve the accuracy of the experts' order method.

1109 VOLUME 1
3.2. A method for deciding items to be milled and
amount of mill using inventory control theory
In the above method, items to be milled and the amount
to be milled are decided using inventory control theory. To
decide, a forecasted average order quantity is used. There-
fore, in deciding what items to be milled and the amount
to be milled, different methods are used for the main items
and the sub items.
1. Main items
In inventory stock theory, "safety stock" is intro-
duced, which statistically avoids stock-out against
orders while following normal distribution. Safety
stock is defined in the following formula, where ks
indicates the constant "service level" (the acceptable
fill rate) and or is standard deviation of past orders.
safety stock = Qd + ks (J
Next, the "acceptable maximum stock" is defined as
stock limits to avoid expired stock. It is defined in
the following formula, where kL is the "acceptable
expire coefficient" decided by acceptable rate of ex-
pired stock, and d is the maximum stock period.
acceptable maximum stock = d Qd - kL
3.3. The method for deciding the milling sequence
based on job modifiability
The quality of a modified schedule depends on the
milling sequence, so it is necessary to decide on a milling
sequence that allows for flexible modifications to be made
to a schedule to accomodate urgent orders at any time.
Therefore, items which have a small level of milling time
reductibility or a small rate of modification necessity, are
milled earlier in order to allow for adjustments to be made
to accommodate urgent orders. Such items are explained
in the following way:
* A small level of milling time reductibility
As mentioned in section 3.2, because larger stock
amounts of main items are kept as safety stock,
the amount of milling of large safety stock items is
large. Large safety stock items require more milling
time(Figure 4). Therefore, such items goes to be
milled earlier.
Safety stock of A is large. end
Start
,... ................
Overtime
A D FE B]
................. ... ...

In order to avoid a stock-out, the main items are kept n

7J~.Modify
in larger amounts than safety stock. Therefore, using
these formulas, the items to be milled and amount to
C
.A .
D F

be milled are decided as shown in Figure 3. Amount to mill is large. Reductibility is small.
* Items to be milled Figure 4. Large safety stock items
Items whose current stock level is below the
safety stock level are milled. * A small rate of modification necessary.
* Amount to be milled As shown in Figure 5, it is not efficient to repeat-
The milling amount of each item to be milled is edly mill items that have already been milled be-
decided using the following formulas: cause the number of switches increases. Therefore,
to avoid milling items repeatedly, those that have a
amount to be milled = acceptable maxcimum stock large stock-out margin(= prospective stock - amount
-current stock of orders) are milled earlier.
-total amount of current orders
Acceptable Start Urgent Order end
Maximum Stock
Safety Stock
IC A_ D F E BI
Current Stock |J~..M odify C is stock-out.
IC A D F EW 1BCI
1 otal Amount of
nI frr wAntOrcl-r-
Current of Stock is below The item which has already milled.
Safety Stock Product A
The number of switches increases.
Figure 3. A method to decide amount of mill Figure 5. Small stock-out margin items
2. Sub items In order to decide the milling sequence based on the above
Safety stock and acceptable stock are "O" in sub characteristics, items which have in smaller rate of modi-
items. The amount to be milled is calculated us- fiability, defined in the following formula, are milled ear-
ing the following formula to replenish the reduced lier:
amount of an out-of-stock item:
Amount to be milled -current stock modifiability -safety stock
+order amount -stock-out margin

1 11 0 VOLUME 1
3.4. A method for reducing items to be milled and 4.1. Target data
amout of milling required For experiment, two real orders of a continuous 5 day
In the final phase of working hours in a day, items that and 6 day week are prepared. The total amount of orders
are out-of stock due tourgent orders must be added to the in a day is between 15000kg and 20000kg, and the num-
schedule, and an excess of working hours then occurs. ber of milled item is twenty. "Item ID" is assigned to each
Therefore, a method that allocates time margins for urgent items from 1 to 20. The items from 1 to 10 are the main
orders is proposed, in order to reduce the excess of work- items, while the items from 11 to 20 are sub items. The
ing hours. When an excess of working hours occur, a time working hours consists of 450 minutes the and order time
margin is allocated by reducing the amount of milling of is 390 minutes. The maximum stock period (d) is three
the larger stock-out items(Figure 6). days. Other parameters and a formula expressing constar-
The time margin is decided using the "Average amount ints are decided as follows:
of total orders". Milling performance indicates the
amount the milling process can mill in one minute. * Safety stock coefficient ks: The service level is set to
of total orders
ave. amount 90%, so that the safety stock coefficient ks = 1.64.
time margin =work hours-
milling perf ormance
Start Main items * Acceptable coefficient kL: The acceptable expired
end stock rate is set to 1%, so that acceptable coefficient
...........
Before kL = 2.33
Reduction I 17 | 1
After
Reduce
v
X/
A-
Overtime
* Milling amount constarints:The unit milling amount
is 600kg. The amount to mill is rounded out to mul-
Reduction 499--h tiples of 600kg.
Time margin amount
amoun to m'll
to mll = F Famount millx
600
to
] 0
Figure 6. Allocation of time margin
In the evaluation function defined in section 2.3, the
When an excess of working hours occurs, the amount weights are defined in Table 1. The weights are decided
of milling is reduced using the following process: by experts based on their experience and are not changed
1. In order to keep the stock of main items above safety in this problem.
stock level, amount of the largest stock-out margin Table 1. Evaluation function of weights
for item is reduced using the following formula (Fig- The kind of weights weight
ure 7); 8
wl l
Acceptable
Maximum Stock W2 p 40
Amount to modiS W3 1l 5
Safety Stock
r 1 _ -~~~~Current
C Stock
Stc 4.2. Experiment of the proposed method
Amount to mill This proposed method is developed on a computer and
Milli g up to Safety Stoc
0
Total Amount of
Current Orders
then applied to the target. Comparisons of the evaluation
between the result using experts' schedule and the result
Figure 7. Reduction in amount of mill using the proposed method, is shown in Table 2. A com-
parison of the evaluation is explained in the following for-
mula:
2. When the time margin can be allocate (time margin eval. by a proposed method
comparison of eval. - x 100
< work hours - end time) as the result of recalculat- eval. by producer
ing the end time, this process is finished.
3. If the items that can be reduced(safety stock < stock-
out margin+amount of mill) remains, the process
goes back to 1. Otherwise, the process goes on to
4.
4. The larger stock-out margin items are deleted. This
process is not performed on items whose stock-item Time(min.)
margin is over 0 or whose time margin can be allo- Figure 8. Comparing schedules on the 1st
cated. day in 1st data
Next, comparisons between schedule the experts' results
4. An Application to a Practical Problem and the schedule using the proposed method is shown in
Figure 8. Figure 9 shows the transition of a schedule that
The proposed method has been applied to a practical is modified when urgent orders are received. In Figure 10,
problem in order to verify its effectiveness of the method. the results of the schedule on the days in the week using

111 VOLUME 1
the proposed method are illustrated. The number on the * A method for deciding the amount to mill
schedule in Figure shows the item ID.
From the results of Table 2, in each evaluation factor, As shown in Figure 8, amount of items to mill using the
overtime, the number of switches and throughput, the re- proposed method is larger than that of the experts, so the
sults of the proposed method are superior to the experts' number of switches decreases. It is confirmed that a pro-
method. From Figure 8-10, the methods for deciding posed method can decide the appropriate amount to mill.
what items to mill, amount to mill, and the milling se- * A method for deciding the milling sequence of items
quence is verified as follows:
Table 2. Comparing evaluation As a result of the schedule made by the producers in
experts the proposed Figure 8, item "7" is processed first and earlier than item
1 st data method "2" in experts' schedule, but it is not milled in the sched-
over time(minutes) 227 100 ule made by the proposed method. From Figure 9, in the
a number of switching 58 32 initial schedule made by the proposed method, item "7"
throughput(kg/h) 2405 2588 is allocated but the following item "2", which has more
evaluation value 7889 10860 safety stock. During the processing of item "2", because
comparison of eval. 100% 137% there are few orders for item "7" and there are other urgent
2nd data || | orders, the low-priority item "7" is deleted and is milled
over time(minutes) 140 120 the next day(Figure 10). It is confirmed that the proposed
a number of switching 58 47 method can decide an effective milling sequence to reduce
throughput(kg/h) || 2144 2219 the number of switches. Therefore, the effectiveness of
evaluation value 7280 8255 the proposed method in this research can be confirmed.
comparison of eval. 100% 113%
Time to Schedule 5. Conclusion
Omin X...................... ...5

In this problem, because urgent orders account for 90%

in
................. ........... of the whole, stocks often run out and the need to mill
new item occurs in multi-items. This research proposes
...........................
.............................. a modifiable scheduling method to minimize the num-
................................................
ber of switches and the amount of overtime worked when
1 1 1 1 1 1 1 1 1 1 1.,1 1. . . . . .
................................. ,111111
milling starts in a day and when urgent orders are received
18

14 1.

..... that exceed the work-in-process stock. In this research, a

14 scheduling method is proposed for a milling process that
60 120 180 240 300 360 420450 produces multi-items and that adjusts to accommodate ur-
Time(min) gent orders based on job modifiability. The items to be
milled and amount to mill are decided based on an inven-
Figure 9. Transition of a schedule by the
proposed method on the 1st day in lI1st data
st data
tory control theory. The milling sequence is decided using
modifiability. The proposed method was applied to a real
Daty
g problem. It was confirmed that the proposed method can
1st
...jj

14 generate solutions 37% more effective than the experts'

14
ATll OU1n
2nd day
\
i

. m 1KJ-
15 13
3rd dayl 6 iIgi i X M S.1m g 11919
f1
References
4th day ..
- 19 [1] D. M. Lambert, J. R. Stock, and L. M. Ellram. Fundamen-
5th day I I _I l _ [ mi -. -218 17 16 tals of Logistics Managemaent. McGraw-Hill, Inc, 1980.
[2] J. D. C. Little. A proof of the queuing formula:l=Aw. Op-
eration Research, 9:383-387, 1961.
......................................................

........................................................................

60 120 180 240 300 360 420450 [3] R. Shafaeri and P. Brunn. Workshop scheduling using prac-
Time(min.) tical (inaccurate) data part 1: The performance of heuristic
scheduling rules in a dynamic job shop environment using
Figure 10. Week's schedule by the proposed a rolling time horizon approach. Int. J. of Production Re-
method in 1st data search, 37(17):3913-3925, 1999.
[4] M. J. Shaw, S. Park, and N. Raman. Intelligent schedul-
* A method for deciding what items to mill ing with machine learning capabilities -the induction of
scheduling knowledge. IEE Trans., 24(2):156-168, 1992.
As shown in Figure 8, by milling only out-of-stock [5] T. E. Vollman, W. L. Berry, and D. C. Whybark. MANU-
items, a schedule that has less items is created. It is con- FUCTURING PLANNNING AND CONTROL SYSTEMS.
firmed that the proposed method can decide what items McGraw-Hill, Inc, 1997.
were milled prior:

1112 VOLUME 1