You are on page 1of 15

Robotics and Computer Integrated Manufacturing 19 (2003) 157–171

MRP-controlled manufacturing environment disturbed


by uncertainty
S.C.L. Koha,*, S.M. Saadb
a
Management School, University of Sheffield, 9 Mappin Street, Sheffield S1 4DT, UK
b
School of Engineering, Sheffield Hallam University, City Campus, Sheffield S1 1WB, UK

Abstract

Most identified research on uncertainty in material requirements planning (MRP)-controlled manufacturing environment, while
purporting to represent such environment, does not truly model a multi-level dependent demand system, with multi-product and
controlled by planned order release (POR) schedule based on planned lead times. This research has developed an MRP-controlled
batch manufacturing simulation model using ARENA simulation software to represent such environment. In this paper, the delivery
performance of the environment, measured with parts delivered late (PDL) and finished products delivered late (FPDL), when
disturbed by uncertainty is examined. Design of experiments and simulation studies are carried out for these uncertainties: external
late supply, internal late supply, planned set-up time exceeded, machine breakdowns, labour unavailability, tooling unavailability,
demand batch size enlargement and customer design changes. Analysis of Variance (ANOVA) results identify significant effects
from individual uncertainty as well as combination of uncertainties. The knock-on and the compound effects cause more PDL and
FPDL. The main finding is that the higher is the level of the significant uncertainties, the higher is the level of PDL and FPDL. It is
suggested that MRP-controlled manufacturing enterprise must diagnose for the significant uncertainties, so that optimum use of
buffer or slack can be implemented. Some practical implications of this work and required further work are also discussed.
r 2003 Elsevier Science Ltd. All rights reserved.

Keywords: MRP; Release logic; Uncertainty; Batch manufacturing

1. Introduction of these systems in batch manufacturing enterprises is


referred as MRP-controlled manufacturing environment.
Modern manufacturing enterprises are facing increas- Extensive research can be found in the examination
ing pressure to respond to production dynamics caused for an appropriate technique to tackle uncertainty [2,3],
by disruption of uncertainty. In this paper, uncertainty for example, the use of safety stock to tackle quantity
is defined as any unplanned events that occur during uncertainty or rescheduling to tackle timing uncertainty.
production, which disrupt orders execution. Central to However, little emphasis has been put on investigating
this are the role of the production planning and the correct representation of an MRP-controlled man-
scheduling systems within such enterprises. ufacturing environment. This environment typifies a
In batch manufacturing environment, material re- multi-product and multi-level dependent demand system
quirements planning (MRP), manufacturing resource whereby the production orders execution is governed
planning (MRPII) or enterprise resource planning by a planned order release (POR) schedule. A POR
(ERP) has been recommended as the ideal system [1]. schedule is the typical output from the MRP run [4].
Since the MRP release logic is deployed in MRPII and A range of different methods may be used for order
ERP systems, when they are used for production release with MRPII or ERP system, including Kanban
planning and scheduling, the scheduled outputs are and work-to-list. These methods are not considered in
identical. This research focuses on their roles in this study because we are interested in the performance
production planning and scheduling, therefore the use of a push system in a finite-capacitated scenario without
intervention of any buffer or slack.
*Corresponding author. Tel.: +44-1144-222-3395; fax: +44-114-
The use of Kanban may invalidate the planned release
222-3348. time in the POR schedule because it will pull order from
E-mail address: s.c.l.koh@sheffield.ac.uk (S.C.L. Koh). the preceding workstation if the current workstation is

0736-5845/03/$ - see front matter r 2003 Elsevier Science Ltd. All rights reserved.
PII: S 0 7 3 6 - 5 8 4 5 ( 0 2 ) 0 0 0 7 3 - X
158 S.C.L. Koh, S.M. Saad / Robotics and Computer Integrated Manufacturing 19 (2003) 157–171

idle, which may result in early release of order. Early manufacturing environment under demand and lead-
release contradicts with the MRP release logic within time uncertainties. The effects from the use of different
which order will never be released until the release date lot-sizing rules were also considered. It was concluded
is reached [5]. This is to avoid part shortages at the that these uncertainties could be tackled by using
associated lower level bill of material (BOM) to process appropriate lot-sizing rules. A multi-product and
the related part at the higher level. To ensure the multi- multi-level dependent demand system was developed,
level dependent demand relationship is permanently which the production orders execution was controlled
linked, particularly, under a finite-capacitated manufac- by POR schedule. Additional algorithm was coded to
turing environment, this release logic plays a vital role control order release by checking availability of all
in evaluating the effectiveness of release timing and required components. Nevertheless, their order release
quantity for parts at the subsequent process. logic ignored the feasibility of parts’ early completion,
Work-to-list lists a series of jobs that need to be which it allowed order at the higher levels BOM chain to
completed in a batch. Some enterprises use this as their be released early. The consequences of this logic are that
production execution mode [6]. The use of work-to-list is some part shortages would occur in the process and in
unnecessary in this study because batch release can reality the part that was affected by the shortages would
easily be designed in the POR schedule; therefore one have never been completed if buffer or slack does not
design architecture has been eliminated. This will be exist.
discussed in the later section. To represent MRP-controlled manufacturing
A correct representation of MRP-controlled manu- environment, similar simulation models have been
facturing environment is crucial to demonstrate and developed by Enns [1], Ho and Carter [9] and
analyse how uncertainty affects the production and Homem-de-Mello et al. [10]. The above omissions are
enterprises’ performance, and to effectively provide a the most common, made by the researchers in this field.
knowledge base for prevention and/or reaction. Little research can be found, which has modelled
the MRP release logic in its entirety. They were the
closest representation of MRP-controlled manufactur-
2. Literature review ing environment.
Kanet and Sridharan [11] examined late delivery of
Review of the literature on representation of MRP- raw materials, variations in process lead-times, inter-
controlled manufacturing environment to examine its operation move times and queue waiting times in MRP-
performance when disturbed by uncertainty has been controlled manufacturing environment. To model such
carried out. This section discusses the past research and environment, they represented demand by inter-arrival
comments on the effectiveness of previous representa- time rather than defined from the master production
tion of such environment. schedule (MPS). Matsuura et al. [12] adopted the same
Minifie and Davies [7] developed a dynamic MRP- approach to model demand. This has resulted in the
controlled manufacturing system simulation model absence of an order release schedule to control produc-
to study the interaction effects of demand and supply tion orders execution. Since the MRP lead-time was not
uncertainties. These uncertainties were modelled in represented in anywhere in the system, it did not model
terms of changes in lot-size, timing, planned orders the MRP release logic. The production orders execution
and policy fence on several system performance was controlled by a flow line rather than processed in
measures, namely late deliveries, number of set-ups, batch. This logic ignored the MRP multi-level depen-
ending inventory levels, component shortages and dent demand. Additionally, only a single product was
number of exception reports. It was concluded that modelled and there was no control over the release
the system performance is significantly affected when timing of the orders. The use of MRP needs to be
disturbed by demand and supply uncertainties. justified with a complex multi-product set-up because
They used POR schedule to release order into the simple product variety does not realistically require the
simulation model. However, their simulation model assistance of such systems. Their models are a more
allowed partial order release when some required parts accurate representation of an independent demand
at the lower level BOM chain are available on hand. system or a flow line.
This logic violates MRP release logic in two-fold: (1) Yang and Pei [13] examined the effects of engineering
order at the upper level BOM cannot be released until changes using integration between a design model
all parts at the lower level BOM are available, and (2) and an MRP-controlled manufacturing system simula-
this order cannot be released early or it can only be tion model. In their study, only engineering changes
released when its release date is reached or at a later during planning were examined. These changes
release date (late release). were transformed into the manufacturing BOM before
Using a similar research methodology, Brennan and the production begins. It was concluded that early
Gupta [8] examined the performance of MRP-controlled engineering changes could be accommodated easily in
S.C.L. Koh, S.M. Saad / Robotics and Computer Integrated Manufacturing 19 (2003) 157–171 159

production. They did not consider late engineering 3.1. Multi-level dependent demand
changes that might occur during production, which
could render significant deviations to the POR schedule A typical POR schedule contains customer order
on order release timing and quantity. Since the MRP number, part number, net requirement, release date
release logic used was not discussed in detail in their and due date. It is a collection of scheduled output of
paper, definite conclusion cannot be made on the basis all orders in the MPS. It acts as the controller to
of the effectiveness of their model. release order for production execution in the simulation
Although much research can be found claiming model.
have had modelled the MRP release logic in its entirety, Since the simulation model executes forward in time,
little evidence are available to support and explain all orders in the POR schedule have to be ranked in
the implementation of the logic. Although the sugges- chronological order by release date and part number.
tions proposed by the researchers may or may not be This ranking enables parts at the lower level BOM for
affected by the effectiveness of the representation each product with smaller release date to be processed
of such environment, the simulation results will certainly prior to their sub-assembly and finished product.
vary and this may subsequently affect the validity of Nevertheless, the orders in the POR schedule do not
the suggestions in relation to the performance measures have a unique identity likewise in the BOM to link the
used. For example, if early release is persisted, the parts demand dependency. Although their due date can
implication will be unnecessary higher inventory level as be closely tracked with the release date of the related
the parts are not required until the scheduled date. part at the upper level BOM, there could be an identical
Consequently, this gives a faulty inventory level and it release date for a different part, which may create
could lead to a deceptive proposal on reducing the level ambiguity in establishing parts demand dependency.
of inventory. Additionally, if delay has occurred during production,
Two of the most important findings from this such tracking may not work due to lead-time variations.
literature review are: (1) MRP release logic has not To overcome these limitations, a tagging configura-
been represented in its entirety, and (2) simulation tion is developed. The tagging configuration is derived
modelling appears to be the most common research from the concept known as the parent and child in MRP
methodology used to model MRP-controlled manufac- system. Child is referred to required parts or compo-
turing environment. nents to assemble or manufacture parts at the upper
This research aims to develop a batch manufacturing level BOM chain. Parent is defined as parts, sub-
system simulation model to truly represent an MRP- assemblies or finished products, at the upper level
controlled manufacturing environment in its entirety. BOM chain, which is dependent on completion and
This paper presents the developmental and experimental availability of all required child before they can be
work on modelling its operations and performance when processed.
disturbed by several uncertainties simultaneously. It also Using this concept, several tagging attributes are
provides some useful insights into the relative signifi- designed but the most important attributes are: (1) part
cance of the uncertainties. tag, (2) parent tag, and (3) child number tag. Part tag is
configured from the concatenation of customer order
number and part number. Parent tag is configured from
3. Simulation model development the parent’s part tag. Child number tag is configured
from the sum of the required number of child for a part.
This research has developed a batch manufacturing Consider an illustrative example. Let:
system simulation model to represent an MRP-con- Customer order number for product j¼x;
trolled manufacturing environment. It is programmed Child number tag for product j, which is a parent¼7
using SIMAN V simulation language within the (assume it requires a total of 7 child);
ARENA simulation software. The baseline in the The part numbers for the child of product
simulation model is a series of resources (machine, j¼a,b,c,d,e,f,g;
labour or tool) and its associated queues. Child number tag for part a¼5;
The most challenging work in order to successfully The part numbers for the child of part a ¼ h; i; j; k; l;
model such environment is to effectively design The tagging configuration will generate the following:
and implement the MRP release logic in a simulated The part tag of product j¼xj;
finite-capacitated manufacturing system. This includes The parent tag of product j¼x0; (a value of zero is
modelling of a multi-level dependent demand system assigned because it does not have a parent)
and an order release timing control. We look at The part tag of part a¼xa, the part tag of part b¼xb,
the information in a typical POR schedule to investigate y the part tag of part g¼xg;
how this information can be used to model such The parent tag of part a¼xj, the parent tag of part
environment. b¼xj, y the parent tag of part g¼xj;
160 S.C.L. Koh, S.M. Saad / Robotics and Computer Integrated Manufacturing 19 (2003) 157–171

The part tag of part h¼xh, the part tag of part i¼xi, On the other hand, if all required child has completed
y the part tag of part l¼xl; their production earlier than their due date, the
The parent tag of part h¼xa, the parent tag of part parent will start its production at TNOW, which
i¼xa, y the parent tag of part l¼xa; is earlier than the release date. In contrast, the
Recognition of parent and child is shown by the parent should never be released earlier than the release
identical values between the part tag of product j and date.
the parent tag of its child. The example has also Given that we need to control over the order release
considered the case of multi-level dependent demand timing, it is important to be able to evaluate the validity
(parent=part a, child¼part h,i,j,k,l). The values of the of any potential release. Evaluation of the difference
child number tag give the expectation for the number of between TNOW and release date gives three potential
matching values sets. In this example, there are 7 sets of releases scenario: (1) prompt, (2) late, and (3) early.
matching values between product j and its child; and Fig. 1 shows the order release timing logic, which is
there are only 5 sets between part a and its child. An designed to eradicate the above limitations.
identical parent tag value for each of the part tag of part If a part is about to be released, and the TNOW
a–g and h–l is expected at their respective lower level equals its release date, the part will be released without
BOM chain. To summarise, by comparing the values of delay. This scenario is known as the prompt release,
the part tag and the parent tag, such tagging configura- which it can be achieved when the lead-time is accurate
tion could clearly recognise and verify the parts demand and the planned resources are available. This scenario
dependency. may or may not be achieved at all time due to finite-
The tagging configuration is coded using Visual Basic capacitated manufacturing and disruption to production
Application. The results are known as an advanced POR from uncertainty.
schedule. The advanced POR schedule contains all In such case, the second scenario is feasible. Late
typical information in a POR schedule with the addition release is the outcome when the TNOW is greater than
of the tagging attributes. Using this information, the the release date. The part will be released at TNOW. If
parts multi-level dependent demand is permanently this is a child for another parent, it is likely that the
linked in the simulation model. parent will also be released late due to the tardiness of its
child under the condition where buffer or slack does not
exist. This scenario may or may not be true if buffer or
3.2. Order release timing slack is available.
For example, if the planned lead-times in the MRP
Owing to the assumption of infinite capacity in MRP system for the required child of a parent have been
to generate the release date, there is no guarantee that over-estimated, this will result in early completion of
the parts can be released according to plan in the the child if the planned resources are available to process
simulation model when resource constraints or uncer- the parts. As a result, the third scenario may be feasible
tainties are applied. We examine the default time that because the parent is in a position to be released.
the simulation model uses. We found that every part is However, MRP will only release order when the
being released at its first feasible current simulation time release date is reached or thereafter, hence the earliness
(from time zero). In SIMAN V, this time is stored in a (release date—TNOW) is offset to counteract with
reserved variable known as time now (TNOW) [14]. this scenario. The implication is similar to the prompt
However, the release date is the start time for produc- release scenario, with the exception that the delay is
tion in an MRP-controlled manufacturing environment, relatively smaller for the third scenario.
and hence this time should be used instead of the
simulated start time.
Delaying the order release date before the first
Parts to be released
operation begins has successfully solved this problem.
The TNOW is being forwarded in time. This has
resulted in the orders to be released according to their Yes
TNOW = Release date? Prompt release
release date if tardiness does not occur.
Nonetheless, release of parent in such time implies No
that they are independent of the completion time of Yes
Route parts to
their child. If some required child has not yet completed TNOW > Release date? Late release
routing
their production, the parent will start its produc-
No
tion with child shortages and it is still able to complete
the batch. In contrast, the parent should never be Delay: Release date - TNOW;
Prevent early
release
released until all required child arrived. When they
are arrived, the parent will be released at that time. Fig. 1. Order release timing logic.
S.C.L. Koh, S.M. Saad / Robotics and Computer Integrated Manufacturing 19 (2003) 157–171 161

3.3. Simulation model logic lower level BOM will be processed. When they have
completed their production, they have to search for their
The designs of the multi-level dependent demand parent. Thus, they cannot be disposed. However, their
system and the order release timing control are parent does not yet exist in the simulation model
embedded in the simulation model logic. Fig. 2 shows because they have not been read. This prohibits the
the simulation model logic. searching and hence the parts demand dependency is
The priority for production orders execution is lost.
determined by the release date. The earlier is the release These problems can be solved in two methods: (1) by
date, the higher is the execution priority. To model such modelling a pre-defined queue for every completed child
priority rule, the release date and the part number for all to wait for their parent to arrive, or (2) by pre-routing
orders in the advanced POR schedule are pre-ranked every parent to a pre-defined queue at the beginning of
in chronological order. Since ARENA permits reading the simulation to wait for the completion of their child.
of sequential data [14], such ranking automates the Both methods have been tested and we found that they
execution priority in the simulation model. give identical results, but the latter method reduced the
To route the orders for production process to the simulation time by 97%. Therefore, this method is
planned resources, a routing number is initially assigned implemented in the simulation model.
to each part in the advanced POR schedule. The routing Pre-routing the parent requires upgrading their record
consists of the set-up and operation times at the planned sequence to the forefront of the advanced POR
machine, labour or tool for producing a part. There is a schedule. To verify whether the part is a parent, only
destination in the simulation model where the routing is those order that has a child number tag greater than zero
stored with respect to each routing number. By assign- is upgraded. This arrangement enables pre-routing of
ing the part routing number to the sequence Number every parent to a pre-defined queue in the simulation
(NS), a reserved variable in SIMAN V for defining model. The simulation model needs to recognise the
operations sequence [14], the parts can be easily directed threshold of such pre-routing. Therefore, the same
to its own routing. Batch manufacturing is modelled by parent verification is coded. If this condition is true, it
delaying the operation time for a batch. Given that the verifies for the existence of a parent, and the parent is
operation time in the routing is defined in a unit basis, routed to a holding bay (pre-defined queue). Until a
this time is multiplied with the net requirement of the point where this condition is false, the next part will be
order. the first child, which has the earliest release date.
The values of the attributes in the advanced POR The child will be routed to its routings for production.
schedule are read into the simulation model record-by- Before production, the release date will be evaluated
record. In such case, it is expected that those child at the according to the order release timing logic. When it is

Start

Advanced
POR Read
Schedule

No Part = Child at the Once a child/parent is


Child number tag>0?
lowest level BOM completed:

Yes

Part = Parent Evaluate release


Route to its queue date SEARCH, No
ParentQSet(Holding): Record
(Parent tag == Part performance
tag);
ParentQSet(Holding) Route for supply
Dispose
Yes

No
Child number tag = REMOVE: Parent entity,
Child number tag = 0? End
Child number tag - 1 ParentQSet(Holding);

Yes

Route parent for


Evaluate release date
manufacture

Fig. 2. Simulation model logic.


162 S.C.L. Koh, S.M. Saad / Robotics and Computer Integrated Manufacturing 19 (2003) 157–171

completed, it exits the manufacturing system. The exit is These products consist of three to five BOM levels
coded to signify for the completion of a routing. Then, (multi-level) with a mixture of 60% purchased and 40%
it searches its parent in the holding bay by finding a manufactured parts. There are a total of 434 different
matching value between the part tag and the parent tag. parts in this product range. This range represents a
When the matching value is found, it removes the parent complex routing assignment for each part. Their MPS
entity in the holding bay. The removal is coded to are run through an MRP system, which result in some
signify for a successful search. After removal, the 50,000 batches of orders in the POR schedule. Their
parent’s child number tag is reduced in one. This values for each tagging attributes are generated and then
reduction is coded to lead to the eventual stage (child the advanced POR schedule is integrated into the
number tag=0), where all required child are completed. simulation model for the experiments.
At this stage, the entity can be disposed when all Ten types of machines, six types of assemblers
required performance measures are collected. and inspectors, and four types of tooling are modelled.
A parent may have more than one child. Hence, the Table 1 shows the resources capacity levels. All
next step is to check whether all required child is machines are set to work on a 24 h basis, whilst all
completed before it can be released under the control of labour work on 3 eighth shifts.
the order release timing logic. If these conditions are not It has been assumed that no alternative machines are
being satisfied, the parent will continue to stay in the available if the required machines breakdown and no
holding bay to wait for the other required child. alternative routing can be executed if an order needs to
Otherwise, the parent will be released for production. be expedited. Additionally, no multi-skilling labour has
To accommodate for those parts that have double also been assumed. The tooling assignment for each part
roles (parent and child) for different parts, the same is fixed; therefore no flexible set-up can be performed if
search, remove and child number tag reduction processes the required tools are not available. These assumptions
will recur until a negative search result is obtained. This are designed towards setting-up the experiments that are
result implies that the parts multi-level dependent without the influence of any buffer or slack.
demand has been entirely processed. These processes
cease when all orders in the advanced POR schedule are 4.2. Performance measures
processed.
Owing to the objective and the design of this
experiment, the effects of uncertainty can be clearly
4. Experimental design measured by the amount of negative effects on delivery.
Timely delivery is becoming increasingly important in
The objective of the experiment is to investigate the today’s competitive business environment; hence the
net effects of uncertainty on delivery performance in an performance on tardiness is measured.
MRP-controlled manufacturing environment. The net
effects are referred to the outcome after we have relaxed
the influence of any buffer or slack on the effects of Table 1
uncertainty. This is designed to provide clear insight and Resources capacity levels
direction to MRP-controlled manufacturing enterprises, Resource Type Capacity
for deciding their priority in tackling the uncertainty.
Guillotine Machine 1
Several important issues in the experimental design
Break Press Machine 1
are addressed: (1) what parameters settings to be used, Drill Machine 1
(2) what performance measures to be used, and (3) what Coiler Machine 4
uncertainties to be modelled, at how many levels and at Finished Coiler 1 Machine 1
what levels. The assumptions used for the experiments Finished Coiler 2 Machine 1
Turret Punch Machine 2
are discussed within the relevant issues.
Weld Machine 3
Mixer Machine 5
4.1. Parameters settings Setting Machine 1
Assembly station 1 Labour 12
The simulation model is parameterised using the data Assembly station 2 Labour 1
Assembly station 3 Labour 1
collected from a transformer manufacturer, which use
Assembly station 4 Labour 2
an MRP system for production planning and schedul- Deburr Labour 1
ing. They operate in batch manufacturing. We study Inspector Labour 1
their product range to identify the most active products Tool Set 1 Tool 1
sales over the last 2 y. This has led to ten products Tool Set 2 Tool 1
Tool Set 3 Tool 2
(multi-product) and hence the MPS is set-up based on
Tool Set 4 Tool 2
their demand.
S.C.L. Koh, S.M. Saad / Robotics and Computer Integrated Manufacturing 19 (2003) 157–171 163

Dual tardiness performance measures are examined: enterprises (including the transformer manufacturer),
(1) the proportion of finished products delivered late which found eight uncertainties that are most likely to
(FPDL), and (2) the proportion of parts delivered late affect customer delivery performance. These are external
(PDL). In this experiment, tardiness is defined as late supply, internal late supply, planned set-up time
TNOW>due date. This condition is evaluated before exceeded, machine breakdowns, labour unavailability,
a part is disposed in the simulation. If this condition is tooling unavailability, demand batch size enlargement
true, the part will be recorded and added to the tardiness and customer design changes. Their experiments are
counter. designed. Table 2 shows a brief description of each of
If the part involved is a finished product, the FPDL is the uncertainties.
increased by 1/total number of finished products in the The number of levels per factor determines the total
advanced POR schedule. Recognition for a finished number of experiments required. To reduce the number
product is decided by a parent tag with a value of zero of experiments required, a half factorial of eight
being assigned to its secondary concatenated fragment uncertainties each at two levels is designed. This requires
(parent’s part number). If it is not a finished product, a total of 12  28 ; 128 experiments. Two levels are
the PDL is increased by 1/total number of parts in the considered sufficient in this study because it enables
advanced POR schedule (50,000—total number of the uncertainties to be modelled realistically at high and
finished products). low levels. Ten replications are run for the pilot
Both measures are designed to give different views on experiments to achieve a normal distributed response.
the effects of uncertainty. For example, if a part is Therefore a minimum of 1280 simulation run is
delivered late, it will certainly be recorded in the PDL. required.
However, this part may be used as a child for other A normal distributed response is measured by an
parts, which means its multi-level demand dependency acceptable half-width (h) value, considered to be less
may result in late delivery on its related parent. than 5% of the sample mean [15], whilst the confidence
Therefore, a part that is delivered late as a result of level is set in all experiments to be 95%. If the error is
uncertainty may lead to larger PDL. On the other hand, p5%, a normal distributed response is obtained. Thus
such effects may not be affecting FPDL if buffer or slack no further replication is required. Otherwise, further
is employed to expedite the affected orders. Since we replication is required until the limit is satisfied. See [14]
aimed to ensure that there is no buffer or slack in the for the procedures to calculate the h value and the
simulation model, it will be interesting to see whether number of further replication. Recursion of this
there is any difference in terms of the effects on procedure has resulted in a total of 1534 simulation
deliveries. run for this design of experiment.
Interviewing the transformer manufacturer gives de-
4.3. Uncertainties tails for the best and the worst scenario for setting the
levels of uncertainties. These levels are tested in the
The literature review appears to cover a diversified experiments in conjunction with a sensitivity study
range of uncertainties. However, there is no consensus to investigate the responsiveness of the simulation
on why those uncertainties were examined. Without model to such levels. This is performed to ensure the
establishing for the correct reasons, MRP-controlled levels are within the tolerance of the simulation model to
manufacturing enterprises might not rationalise their give a sensible outcome. Table 3 shows the modelling
effort. methods for each uncertainty and the selected levels
A business model for diagnosing underlying causes from the results of the sensitivity study. This is the
of uncertainty was developed by Koh [6]. This model replica of a realistic level that is experienced by the
was applied to 126 MRP-controlled manufacturing transformer manufacturer.

Table 2
Uncertainty descriptions

Uncertainty Description

External late supply External suppliers deliver later than scheduled


Internal late supply Internal suppliers, e.g. workstations of previous stages of production, complete their production
later than scheduled, thus they deliver later than scheduled
Planned set-up time exceeded Machines take longer than the planned time to set-up
Machine breakdowns Machines failure or stoppage
Labour unavailability Labour scheduled for production are not present
Tooling unavailability Tooling scheduled for production are not present
Demand batch size enlargement Customer increases demand or order size of scheduled product
Customer design changes Customer changes ordered product specification, hence resulting in additional operations required
164

Table 3
Uncertainty modelling methods and the selected levels

Modelling method Levels

Uncertainty Parts/orders/resources Probability Low High


subjected distribution

External late supply Purchased parts with lead-time variation Discrete F ¼ 2%


M ¼ 1 day M ¼ 3 days
Internal late supply Manufactured parts with lead-time variation Discrete F ¼ 2%
M ¼ 1 day M ¼ 3 days
Planned set-up time Manufactured parts with machine set-up time Discrete F ¼ 5%
exceeded variation
M ¼ 15 min M ¼ 240 min
Labour unavailability Assemblies or parts for inspection with lead-time Discrete F ¼ 5%
variation
M ¼ 15 min M ¼ 240 min
Tooling unavailability Manufactured parts that require tool sets for Discrete F ¼ 5%
machining with lead-time variation
M ¼ 30 min M ¼ 480 min
Demand batch size Products with varied order size after being Discrete F ¼ 100%
enlargement scheduled
M ¼ 1 order per product M ¼ 10 orders per product
Customer design changes Manufactured parts with operations Discrete F ¼ 4% F ¼ 10%
routings variation
Machine breakdowns Break Press, Coiler and Weld failure Exponential F (Break Press)=125 days
with MTBF and MTTR F (Coiler)=50 days
F (Weld)=63 days
Gamma M (Break Press)=(300 min, 2) M (Break Press)=(1200 min, 2)
M (Coiler)=(120 min, 2) M (Coiler)=(1200 min, 2)
M (Weld)=(150 min, 2) M (Weld)=(1200 min, 2

Keys: F =Frequency: M=Magnitude: min=minute


S.C.L. Koh, S.M. Saad / Robotics and Computer Integrated Manufacturing 19 (2003) 157–171
S.C.L. Koh, S.M. Saad / Robotics and Computer Integrated Manufacturing 19 (2003) 157–171 165

In general, each uncertainty is randomised with a estimated by the designers and the engineers. The design
specific probability distribution. The distribution used is changes are assumed to not involve any additional
dependent on the uncertainty. Parts that are subjected to replenishment of purchased parts. Magnitude of un-
the uncertainties are expressed in terms of frequency and certainty is not assigned because the lead-time variation
magnitude. Frequency models the proportion of parts has already been considered in the revised routing. For
that are subjected to uncertainty. Magnitude models the this uncertainty, frequency is the issue concerned by the
tardiness of the uncertainty. practitioners. Since it is difficult to predict which parts
Consider external late supply, a representation of late would be subjected to customer design changes, thus the
delivery from external suppliers. It is modelled with a frequency is varied to give a completely randomised
discrete probability distribution because such supply effect.
tardiness is usually recorded in number of time unit late Machine breakdowns are a representation of machine
with respect to the proportion of batches affected. In failure. In theory, machine failure can be measured with
reality, practitioners recognise and exercise such records. mean time between failures (MTBF) and mean time to
For example, 2% (frequency) of the purchased batches repair (MTTR). An exponential distribution is used to
are on average being delivered late by one day or by model MTBF. It defines the failures interval, which is
three days (magnitude). This method gives a useful also the frequency for which the machine will break-
context to the practitioners. Since it is also simple to down. A gamma distribution is used to model MTTR. It
model, it is employed in some cases. The frequency is defines the failure elapsed time, which is the magnitude
fixed at both levels because its increment will obviously of the machine downtime. These are the suggested
increase late delivery. The magnitude is varied so that distributions to model machine failure [14]. The
different levels of uncertainty can be modelled. frequency of machine breakdowns is fixed because this
The uncertainties that are modelled with a discrete can be estimated from the machine’s life. The magnitude
probability distribution include internal late supply, of machine breakdowns is varied because different
planned set-up time exceeded, labour unavailability, repairman or maintenance programme would affect
tooling unavailability, demand batch size enlargement the elapsed downtime. This could be due to the learning
and customer design changes. Internal late supply is curve of the repairman or the efficiency of the
similar to the external, only with the exception that the maintenance programme. There are only three ma-
internal affects manufactured parts. The delivery of an chines, which are subjected to this uncertainty due to
internal supply is dependent on the promptness of the their high utilisations.
previous workstations. Only machined parts are sub-
jected to planned set-up time exceeded. Assemblies are
subjected to labour unavailability. Machined parts that
5. Results, analysis and discussions
required tooling are subjected to tooling unavailability.
Demand batch size enlargement is a representation of
The results of the simulation experiments are analysed
demand uncertainty. Demand uncertainty is modelled
using analysis of variance (ANOVA). The following
with batch size variation as opposed to moving back-
sections discuss the main effects and the interactions
ward the due date because the latter can be easily
outputs. The outputs significance is determined at a 0.05
tackled by rescheduling in MRP system. The more
level. Further analyses on the practical implications of
interesting question in demand uncertainty is whether
the results will also be discussed.
order batch size enlargement will distort the POR
schedule. The underpinning motive for this is resources
will not be loaded in 100%; therefore the additional 5.1. Main effects
order may be able to process through the manufacturing
system. This uncertainty is modelled with a double batch The main effects output shows the significance level of
size (frequency=100%) and with varied number of the uncertainty to the performance measures used. The
orders being affected (magnitude) in a product. Varying significance signifies that whenever the level of the
the number of orders affected gives a more randomised uncertainty is increased, the performance of the MRP-
effect because in reality we would not know in advanced controlled manufacturing environment would signifi-
what orders are going to be subjected to demand cantly be affected. Table 4 and 5 show the main effects
uncertainty. outputs to PDL and FPDL, respectively.
Customer design changes are modelled with a routing The outputs highlight four uncertainties that have
assignment problem. The design changes are represented significant effects to PDL and FPDL. These are external
with additional operations in their routing. It has been late supply, machine breakdowns, demand batch size
assumed that the resources required and the time enlargement and customer design changes. Nonetheless,
involved to process the additional operations are internal late supply and labour unavailability have
known. This information can be easily identified and significant effects only to FPDL.
166 S.C.L. Koh, S.M. Saad / Robotics and Computer Integrated Manufacturing 19 (2003) 157–171

Table 4
Main effects output to PDL

Source Sum of square Degree-of-freedom Mean square F p (sig.)

Corrected Model 57225.029 92 622.011 98.463 0.000


Intercept 149609.592 1 149609.592 23682.779 0.000
External late supply 3705.849 1 3705.849 586.626 0.000n
Internal late supply 19.618 1 19.618 3.105 0.078
Planned set-up time exceeded 6.953 1 6.953 1.101 0.294
Machine breakdowns 50881.286 1 50881.286 8054.365 0.000n
Labour unavailability 0.240 1 0.240 0.038 0.846
Tooling unavailability 0.532 1 0.532 0.084 0.772
Demand batch size enlargement 378.375 1 378.375 59.896 0.000n
Customer design changes 53.634 1 53.634 8.490 0.004n
Error 9103.130 1441 6.317
Total 256965.068 1534
Corrected Total 66328.159 1533

Key: * indicates po0.05.

Table 5
Main effects output to FPDL

Source Sum of square Degree-of-freedom Mean square F p (sig.)

Corrected Model 1441466.307 92 15668.112 408.894 0.000


Intercept 5335016.622 1 5335016.622 139228.98 0.000
External late supply 140079.803 1 140079.803 3655.690 0.000n
Internal late supply 494.355 1 494.355 12.901 0.000n
Planned set-up time exceeded 140.508 1 140.508 3.667 0.056
Machine breakdowns 1192396.041 1 1192396.041 31118.195 0.000n
Labour unavailability 323.295 1 323.295 8.437 0.004n
Tooling unavailability 74.567 1 74.567 1.946 0.163
Demand batch size enlargement 817.075 1 817.075 21.323 0.000n
Customer design changes 670.795 1 670.795 17.506 0.000n
Error 55216.658 1441 38.318
Total 8086941.040 1534
Corrected Total 1496682.965 1533

Key: * indicates po0.05.

There are two phenomenons that have been observed which makes the uncertainty to be significant, this effect
in the simulation: (1) knock-on effect, and (2) com- would naturally increase the risk on finished products
pound effect, which explain how the uncertainties delivery. Therefore, FPDL is likely to be significantly
significantly affect PDL and FPDL. Fig. 3 illustrates affected.
the knock-on and compound effects phenomenon using External late supply does not merely result in the
external late supply and machine breakdowns as purchased parts subjected to this uncertainty being
examples. recorded in PDL, but all of their parent will also be
Knock-on effect can be defined as the delay resulted recorded in PDL and FPDL due to the knock-on effect.
from uncertainty from a lower level to a higher level in a Within a typical BOM, purchased parts tend to be
BOM. Within a multi-level dependent demand system, located at the lower levels; hence it has resulted in the
parts tardiness at the lower level BOM chain will have a maximum knock-on effect. Thus, external late supply is
knock-on effect to the higher level. The propagation of found to be significant to both measures.
the knock-on effect is dependent on the level at which Visualisation on the parts that are affected by
the subjected part is located with respect to its entire uncertainty can sometimes be convoluted. This situation
subsequent parent. In another word, the greater is the arises when the resources loading profile has changed.
where-used of a subjected part, the greater is the knock- For example, if machine breakdowns occur during
on effect and hence the larger is the PDL. production, the parts that required machining by this
It is expected that the uncertainty that has significant machine will be accumulated in the queue. These parts
effects to PDL will behave likewise to FPDL. Due to a could be from a variety of products. Hence, the effect
large number of parts affected as recorded in the PDL, can be expanded across product range.
S.C.L. Koh, S.M. Saad / Robotics and Computer Integrated Manufacturing 19 (2003) 157–171 167

Knock-on effects from Compound effects


external late supply from machine
breakdowns

Product A Product B

Assembly c Assembly d
Assembly j Assembly k

Part e Part f
Part l Part m

Purchased part g Part h


Part n Part o Part p

If purchased part g is If part f is subjected to machine


subjected to external breakdowns, whilst part l is at the queue
late supply: of the machine waiting to be processed:

PDL = Part g, f, PDL = Part f, assembly d, part l,


assembly d assembly j

FPDL = Product A FPDL = Product A, product B

Fig. 3. Examples of the knock-on and compound effects phenomenon.

Those parts that are in the queue at the time of variations for processing the revised routing are also
breakdown can easily be visualised but prediction on unpredictable. Their compound effects have resulted in
parts that are going to be in the queue when the machine larger PDL and FPDL; therefore demand batch size
is repaired cannot be made. Their deliveries may be enlargement and customer design changes are found to
affected because the machine may not be idle when the be significant.
parts are ready to be released. It is assumed that the It is found that internal late supply and labour
machine will process the parts in a first-in-first-out unavailability do not significantly affect PDL. This may
(FIFO) basis. When the machine is repaired, it will be be due to the fact that only the manufactured or
busy processing the delayed parts in that sequence. The assembled parts are being affected, which majority of
changes in the planned resources loading profile may them are located at the higher levels BOM. Hence, a
further delay the parts. Such delays will also create a large number of parts at the lower levels have skipped
secondary knock-on effect to their parent. the PDL records. However, those affected parts that are
This phenomenon can be defined as the compound at the higher levels BOM do increase the risk of finished
effect because PDL and/or FPDL are significantly products delivery. These uncertainties create compound
affected when one bottleneck machine or resource effect owing to the unpredictability of the resources
disrupts the synchronisation of the entire manufacturing affected and the time required for processing the parts
system. The compound effect is more immense than the tardiness. Leading to the final level, more and more
knock-on effect. parts are pushed to assembly. Whilst labour unavail-
The compound effect has resulted in machine break- ability affects the assemblies, this has resulted in larger
downs, demand batch size enlargement and customer FPDL. Therefore internal late supply and labour
deign changes to significantly affect PDL and FPDL. unavailability are found to be significant to this
The phenomenon of machine breakdowns has already measure.
been discussed. Demand batch size enlargement changes
the resources loading profile by randomly affecting the
number of orders with increased order size. The 5.2. Interactions
randomness in the orders affected and the increased in
order size have resulted in the resources to be seized for The interactions output shows which uncertainties
an extended period. The resources affected and the time when jointly affecting a part create additional level to
variations are unpredictable, hence queue may build-up the performance measures used. The significance sig-
at the resources and it creates compound effect. nifies the frequent encounter of such interactions. When
Customer design changes also create compound effect such frequency is encountered, the PDL and FPDL
due to the additional operations required. Since the would significantly be affected. Tables 6 and 7 show the
orders that are subjected to this uncertainty are significant interactions from the interactions outputs to
randomised, the resources affected and the total time PDL and FPDL, respectively.
168 S.C.L. Koh, S.M. Saad / Robotics and Computer Integrated Manufacturing 19 (2003) 157–171

Table 6
Significant interactions outputs to PDL

Source Sum of square Degree-of-freedom Mean square F p (sig.)

Corrected Model 57225.029 92 622.011 98.463 0.000


Intercept 149609.592 1 149609.592 23682.779 0.000
External late supply  Machine breakdowns 238.565 1 238.565 37.764 0.000n
Machine breakdowns  Demand batch size enlargement 38.791 1 38.791 6.140 0.013n
External late supply  Planned set-up time exceeded  Labour unavailability 25.076 1 25.076 3.970 0.047n
Internal late supply  Planned set-up time exceed  Demand batch 43.305 1 43.305 6.855 0.009*
size enlargement
Internal late supply  Machine breakdowns  Labour unavailability 29.125 1 29.125 4.610 0.032n
Labour unavailability  Tooling unavailability  Demand batch 30.846 1 30.846 4.883 0.027*
size enlargement
Error 9103.130 1441 6.317
Total 256965.068 1534
Corrected Total 66328.159 1533
*
Key: indicates po0.05.

Table 7
Significant interactions outputs to FPDL

Source Sum of square Degree-of-freedom Mean square F p (sig.)

Corrected Model 1441466.307 92 15668.112 408.894 0.000


Intercept 5335016.622 1 5335016.622 139228.98 0.000
External late supply  Machine breakdowns 88833.234 1 88833.234 2318.298 0.000n
Internal late supply  Machine breakdowns 434.667 1 434.667 11.344 0.001n
Machine breakdowns  Tooling unavailability 242.396 1 242.396 6.326 0.012n
Machine breakdowns  Demand batch size enlargement 235.789 1 235.789 6.153 0.013n
Machine breakdowns  Customer design changes 767.939 1 767.939 20.041 0.000n
Demand batch size enlargement  Customer design changes 156.146 1 156.146 4.075 0.044n
Error 55216.658 1441 38.318
Total 8086941.040 1534
Corrected total 1496682.965 1533

Key: * indicates po0.05.

Some of these two-way interactions are found to be delivered late. Thus, this interaction is found to be
creating significant additional level to PDL and FPDL. significant to PDL and FPDL.
No significant three-way interaction to FPDL or The interaction between machine breakdowns and
significant higher order interactions for both measures demand batch size enlargement occur when the addi-
are found. The interactions that are not significant are tional parts in the orders are due to be processed by a
excluded in this paper. However, this does not mean machine, which is modelled to be subjected to machine
that they do not create additional level to PDL or breakdowns at that time. Such interaction is found to be
FPDL, but the levels are found not to significantly affect significant because of the high frequency of the crossing
the delivery performance. point where the orders affected by demand batch size
The interaction between external late supply and enlargement need to be processed by the broken down
machine breakdowns can be explained with the sequence machines. In this case, the compound effect from both
of events in the manufacturing system. When the uncertainties would multiply and thus more additional
purchased parts that are subjected to external late parts and finished products are recorded in PDL and
supply have ultimately arrived, the resources loading FPDL.
profile would have changed. This may cause the parts The delays would be amplified when a broken down
that are affected by external late supply to be delayed by machine blocks a part that has been subjected to
machine breakdowns, which are modelled at that time. internal late supply, tooling unavailability or customer
The crossing point of such events results in the design changes. A reversed sequence of this event could
interaction between the uncertainties. The combination also result in such delays. These explain the interactions
of knock-on effect from external late supply and between internal late supply and machine breakdowns;
compound effect from machine breakdowns have machine breakdowns and tooling unavailability; and
resulted in many additional parts and finished products machine breakdowns and customer design changes.
S.C.L. Koh, S.M. Saad / Robotics and Computer Integrated Manufacturing 19 (2003) 157–171 169

Tooling unavailability creates compound effect owing delivered late and thus the interaction is found to be
to the unpredictability of the total tardiness involved at significant to PDL.
parts that require such tools for production. Parts that The interactions between internal late supply, planned
require tooling would be manufactured parts, which set-up time exceeded and demand batch size enlarge-
tend to be at higher levels BOM. It has been assumed ment; internal late supply, machine breakdowns and
that customer design changes would only affect manu- labour unavailability; and labour unavailability, tooling
factured parts. Internal late supply would also affect unavailability and demand batch size enlargement are
manufactured parts. Hence, it is expected that the found to significantly affect PDL. These interactions
crossing points of these interactions would be at a later occur in a similar way with the above interaction. To
stage of the production. summarise, the delivery of the same part could be
The compound effects have resulted in these interac- affected sequentially by the changes in the resource
tions to be significant because more orders are affected loading profile that are caused by these uncertainties
by such delays. It is found that these interactions are during the production. These uncertainties individually
only significant to FPDL but not to PDL. These results create compound effects, but when interacted the joint
support our expectation for the time of the crossing effects result in many additional parts to be delivered
points of such events, which have resulted in more late.
finished products to be delivered late.
The significance effect to FPDL from the interaction
5.3. Practical implications
between demand batch size enlargement and customer
design changes could also be explained with the above
Further analyses have also been carried out to
expectation. Demand batch size enlargement would
investigate how do the levels of the uncertainties affect
affect the order size of all parts in the entire product of
PDL and FPDL; and what are the benefits if MRP-
the affected order. However, only the schedules of
controlled manufacturing enterprises know which un-
the manufactured parts would be affected because the
certainty is significant to their performance. Figs. 4
operations time would be doubled. The schedules of
and 5 show the PDL and the FPDL plot for the
the purchased parts would not be affected because their
significant uncertainties at both high and low levels.
releases and deliveries are independent of batch size.
It is found that the higher is the level of the significant
This situation results in a higher risk on the finished
uncertainties, the higher is the level of the PDL and the
products delivery as more manufactured parts are
FPDL. These results clearly show that the levels of the
affected due to the joint compound effects from this
significant uncertainties must be reduced to minimise
interaction.
PDL and FPDL. An average of 18.20% PDL is
All significant three-way interactions are found to
achieved when the significant uncertainties are at their
affect PDL but not to FPDL. Owing to the complexity
high levels. An average of 1.87% PDL is achieved when
of such crossing point, which the delivery of a part
the significant uncertainties are at their low levels. In
involves three uncertainties simultaneously; this may
contrast, the average levels of FPDL are 92.26% and
have subdued the effect to higher levels BOM or at later
12.35%, respectively, when the significant uncertainties
stages of production. Therefore, significant additional
are at their high and their low levels.
level of FPDL is not found.
These results indicate that the customer delivery
Following with the discussion for the inter-
performance (which is denoted by the FPDL level)
action involved external late supply, it is also possible
is unacceptably poor even when the levels of the
that when the subjected purchased parts have ultimately
arrived, the changes in the resources loading profile
could cause further delay to process their parent. 25
It is found that the changes in the resources loading
profile are also due to planned set-up time exceeded 20
and labour unavailability. The interaction occurs when
PDL (%)

15
the affected parts are processed by a machine that
is subjected to planned set-up time exceeded and at 10
some stages in the production, the affected parts are
also delayed by labour unavailability. This could 5
happen before or after the affected parts are
0
delayed by planned set-up time exceeded. In
13
19
25
31
37
43
49
55
61
67
73
79
85
91
1
7

all circumstances, the combination of the knock-on


Experiment Number
effect from external late supply and the compound
Low level High level
effects from planned set-up time exceeded and
labour unavailability have resulted in more parts to be Fig. 4. PDL plot of the significant uncertainties.
170 S.C.L. Koh, S.M. Saad / Robotics and Computer Integrated Manufacturing 19 (2003) 157–171

100 enlargement and customer design changes. Internal late


supply and labour unavailability were significant only to
80
FPDL. Several significant two-way and three-way
FPDL (%)

60 interactions were also found.


The knock-on and the compound effects cause more
40 parts or finished products to be delivered late. The
knock-on effect can be estimated because the parts that
20
are affected by uncertainties can be visualised using the
0
BOM. The compound effect is unpredictable because it
affects parts or products deliveries indirectly. This has
13
19
25
31
37
43
49
55
61
67
73
79
85
91
1
7

Experiment Number led to the conclusion that machine breakdowns, demand


Low level High level batch size enlargement, customer design changes,
internal late supply and labour unavailability must be
Fig. 5. FPDL plot of the significant uncertainties.
tackled with extra care. The deliveries of parts or
finished products, which are affected by significant
interactions of uncertainties, must also be recognised.
uncertainties are low. Therefore, when the levels of the Therefore, further research is required to develop a
uncertainties are high, the FPDL average denotes the method or a strategy to capture the knock-on and the
worst case. It needs to be reiterated that buffer or slack compound effects.
is not available in the manufacturing system that we This research has also shown that the levels of the
modelled. These performances represent the net effect. significant uncertainties must be reduced to minimise
It is useful for MRP-controlled manufacturing PDL and FPDL. This will lead to further work on the
enterprise to know which uncertainty significantly types of buffer or slack to tackle these uncertainties. It is
affects their delivery performance so that they can suggested that if the MRP-controlled manufacturing
optimise the use of their buffer or slack. Such net effect enterprise knows which uncertainty significantly affects
encourages the use of buffer or slack to surmount the their delivery performance, they can optimise the use of
level of late delivery. It will become the baseline of the their buffer or slack. Hence, it is important that this
delivery performance where comparison will be taken. enterprise would diagnose for the significant uncertainties.
After buffer or slack is employed, it is expected that the It can be concluded that FPDL provides a more
delivery performance will be improved. transparent measure than PDL. The level of PDL could
The average PDL levels are lower than the FPDL still be reduced with buffer or slack before it reaches the
averages due to a relatively greater denominator in the later stages of production. Since the ultimate objective
total parts as compared to the total number of finished of many MRP-controlled manufacturing enterprise is to
products. If only the PDL is measured, the effect may be able to deliver to their customer on time, the level of
not be as strong as this. Since many MRP-controlled finished product delivered late (FPDL) do provide a
manufacturing enterprise is keen on improving their clear direction for this enterprise.
customer delivery performance, the use of FPDL would
provide a more transparent result. The level of the PDL
could still be reduced with buffer or slack before it Acknowledgements
reaches the later stages of the production.
The research was partially funded by the ORS Award
from the Council of Vice Chancellors and Principals
6. Conclusions and further work (CVCP). The authors wish to acknowledge the con-
structive suggestions received from the editor and the
An MRP-controlled batch manufacturing simulation reviewers of this paper.
model that has represented the multi-level dependent
demand system and the MRP order release timing logic
was developed and modelled with a unique method References
called the tagging configuration, which is conceptualised
from the parent and child in MRP systems. The MRP [1] Enns ST. MRP performance effects due to lot size and planned
order release timing logic was modelled with an lead-time settings. Int J Prod Res 2001;39(3):461–80.
evaluation of the potential releases scenario. [2] Koh SCL, Saad SM, Jones MH. Uncertainty under MRP-
planned manufacture: review and categorisation. Int J Prod Res
Simulation experiments were carried out on eight
2002;40(10):2399–421.
uncertainties to PDL and FPDL. Four significant [3] Guide VDR, Srivastava R. A review of techniques for buffering
uncertainties to these measures were found: external against uncertainty with MRP systems. Prod Planning Control
late supply, machine breakdowns, demand batch size 2000;11:223–33.
S.C.L. Koh, S.M. Saad / Robotics and Computer Integrated Manufacturing 19 (2003) 157–171 171

[4] Vollmann TE, Berry WL, Whybark DC. Manufacturing planning [10] Homem-de-Mello T, Shapiro A, Spearman ML. Finding optimal
and control systems. 4th ed. Homewood, IL: Irwin Inc., 1997. material release times using simulation-based optimisation. Int
[5] Koh SCL, Saad SM. Design and implementation of ERP- J Manage Sci 1999;45(1):86–102.
controlled manufacturing in simulation. Proceedings of 2nd [11] Kanet JJ, Sridharan SV. The value of using scheduling informa-
International Responsive Manufacturing Conference, 2002. Ga- tion in planning material requirements. Int J Decision Sci
ziantep, Turkey: University of Gaziantep. 1998;29:479–96.
[6] Koh SCL. Development of a business model for diagnosing [12] Matsuura H, Tsubone H, Kataoka K. Comparison between
uncertainty in MRP environments. Ph.D. Thesis. University of simple infinite loading and loading considering a workload status
Hertfordshire, UK, 2001. under uncertainty in job operation times. Int J Prod Econ
[7] Minifie JR, Davis RA. Interaction effects on MRP nervousness. 1995;40:45–55.
Int J Prod Res 1990;28:173–83. [13] Yang CO, Pei HN. Developing a STEP-based integration
[8] Brennan L, Gupta SM. A structured analysis of material environment to evaluate the impact of an engineering change on
requirements planning systems under combined demand and MRP. Int J Adv Manufacturing Technol 1999;15(11):769–79.
supply uncertainty. Int J Prod Res 1993;31:1689–707. [14] Pegden D, Shannon RE, Sadowski RP. Introduction to simula-
[9] Ho CJ, Carter PL. An investigation of alternative dampening tion using siman/c, 2nd ed.. New York: Mc Graw-Hill, 1995.
procedures to cope with MRP system nervousness. Int J Prod Res [15] Saad SM. Design, analysis of a flexible hybrid assembly system.
1996;34:137–56. Ph.D. Thesis. University of Nottingham, UK, 1994.

You might also like