INT. J. PROD. RES., 2000, VOL. 38, NO.

18, 4823± 4841

Cycle time reduction by improved MRP-based production planning

ASHUTOSH AGRAWAL{, IOANNIS MINIS{ and

RAKESH NAGI}*

The important practical problem of planning the production of large assemblies

employing an MRP-based system is considered. The objective is to produce
products on-time, with minimal cycle time and low work-in-process costs. The
approach is based on the determination of accurate lead-time estimates and on
the introduction and use of lead-time oŒsets in the solution methodology. An
eŒective Lead-time Evaluation and Scheduling Algorithm (LETSA) is employed
that can perform detailed backward scheduling of operations belonging to a large
assembly on a given facility with an objective of minimizing the cycle time. A
scaling procedure is used to account for capacity sharing eŒects by multiple
products in a common facility. These scaled lead-time estimates are then
employed by an MRP-based system to release work-orders on the shop-¯ oor.
The eŒectiveness of these lead-times and lead-time oŒsets are evaluated by
simulating production using the MRP generated order release times and verifying
on-time completion of the multiple assemblies in the common facility. Numerical
experiments are presented to validate the performance of the approach.
Optimized batch sizes for minimal work-in-process (WIP) costs can also be
obtained using LETSA. Thus, the important objectives of minimizing cycle
time for on-time delivery and minimizing schedule costs can be accomplished
simultaneously.

1. Introduction
The shift from conventional mass production to batch production has
accelerated in recent years. In response to continuously varying customer require-
ments, products are being manufacture d in small batches, each with custom features.
This trend is pervasive in both commercial and defence markets and has severe
implications to the operations of a manufacturing enterprise. The diverse product
mix being manufacture d in a common facility greatly complicates both production
planning and scheduling. Additional pressures on these functions are imposed by
severe on-time delivery and minimal cycle time requirements placed upon
manufacturers by the ® ercely competitive market.
In this study, production planning is addressed in a manufacturing facility that
typically produces large and complex assemblies for which cycle times range between
two months to two years. Although contemporary planning techniques, such as Just-
In-Time (JIT) and optimized production technology (OPT) have gained some

Revision received May 2000

{ Department of Decision Sciences and Engineering Systems, Rensselaer Polytechnic
Institute, Troy, NY 12180, USA.
{ Department of Mechanical Engineering & Institute for Systems Research, University of
Maryland, College Park, MD 20742, USA.
} Department of Industrial Engineering, 342 Bell Hall, SUNY at BuŒalo, BuŒalo, NY
14260, USA.
* To whom correspondence should be addressed. e-mail: nagi@ acsu.buŒalo.edu
International Journal of Production Research ISSN 0020± 7543 print/ISSN 1366± 588X online # 2000 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
4824 A. Agrawal et al.

popularity in such environments, the Materials Requirements Planning (MRP) (and

Manufacturing Resources Planning, MRP II) (Orlicky 1975) philosophy is still
employed by the majority of manufacturing enterprises for production planning.
MRP has been found to be an eŒective way explicitly to consider relationships
between the end items and the various components and sub-assemblies.
Furthermore, it has advantages over JIT in the case of unplanned production
changes, erratic demand and breakdown of key equipment. Many practitioners
and researchers consider the JIT and MRP philosophies not necessarily
contradictory, and seek to develop hybrid systems that take advantage of the posi-
tive aspects of the two (Vollman et al., 1992). Perhaps the major drawback of MRP
is the use of lead-times that are determined a priori from approximate
considerations. These lead-times are either underestimated, causing excessive
resource workloads (and production delays), or overestimated, causing excessively
long cycle-times and associated high work-in-process (WIP) costs.
In this paper, we present an approach to improve MRP-based production
planning targeting minimal product cycle times. The approach is based on the
estimation of accurate part lead-times, and on the introduction of lead-time oŒsets.
The Lead-time Evaluation and Scheduling Algorithm (LETSA) developed by
Agrawal et al. (1996) is used to obtain the values of these parameters. LETSA
performs detailed backward scheduling of operations that are necessary for the
production of one or more large assemblies in a common facility, with an objective
of minimizing the cycle time. It is anticipated that the lead-times and lead-time
oŒsets obtained from the resulting schedule will yield minimal deviation between
product due dates and will limit work-centre overloading, such that the cumulative
production cycle-time is minimized.
The approach has been tested by performing production planning for random
demand based on the estimated lead-times and lead-time oŒsets, and subsequently
by using the simulation package SIMFACTORY to schedule the work orders
released according to the production plan. Extensive numerical experiments were
performed to assess the performance of the approach with respect to the on-time
production of multiple products in a shared facility. The impact of employing lead-
time oŒsets was studied with respect to overall cycle time, workload smoothness, and
WIP costs.
The paper is structured as follows. In section 2, the relevant literature is reviewed
and the problem is formulated. The proposed approach and the scheduling heuristic
are presented in section 3. Software implementation issues are presented in section 4.
The performance of the approach is evaluated in section 5. Finally, the conclusions
of the study are presented in section 6.

2. Background
2.1. Related work
The following discussion of literature focuses on lead-times of multi-level
products within a job-shop. Planned lead-times play a fundamental role in the
operation of MRP systems. They are used both to time-phase purchase and work
orders, and to maintain valid and credible priorities (Peterson 1975), so much so that
the structure and level of lead-times can have a fundamental impact on virtually
every aspect of the logistics system (Kanet 1982). However, as indicated in an earlier
survey by Wemmerlov (1979) scant attention has been paid to setting and revising
 Cycle time reduction 4825

lead-time values. In addition, as acknowledged by Melnyk and Piper (1985), the task
of determining even `reasonable’ planned lead-times, is not easy.
A signi® cant body of literature suggests that the lead-times used in practice are
highly in¯ ated; 90± 95% is the waiting time for shop orders and only 5± 10% of the
time is actual `touch-time’ (Huge 1979, Plossl and Welch 1979, Waliszewski 1979,
Wight 1970). John (1985) has studied the relationship between production costs and
lead-times ranging from no slack time to 95% slack (queue) time. This simulation
study revealed an exponential cost increase beyond the 80± 90% queue time level, and
a 41% increase of cost from no slack to a 95% slack. The study concludes the need
for lead-time reduction, either through downward adjustment of MRP planned lead-
times or by introducing new manufacturing concepts. Waliszewski (1979) presents
Hewlett-Packard’ s experience in reducing lead-times by 70% , which improved
customer service levels by 80% , and amounted to $1.7 million savings in work-in-
process inventory over three months. The above literature clearly indicates the
advantage of reducing lead-times or determining accurate estimates for them, and
recognizes the need for a formal method to determining them.
The diŒerence between the planned and actual lead-times has also been referred
to as lead-time error. Several methods to minimize the error have been proposed in
the literature (Melnyk and Piper 1985): (i) adjusting the master production plan until
the materials plan is consistent with capacity utilization, (ii) lead-time management,
or manual intervention with pre-emption, lot-splitting and a short-term capacity
addition, (iii) tracking lead-time error and updating planned lead-times (Schuchts
1979, Steele 1975), and (iv) adding to the average observed lead-time a multiple of
the standard deviation of the lead-time error. The ® rst two approaches are rather
arbitrary and are not based on a formal methodology. The third method is practised
often in industry. However, extensive data collection and the need for an established
history are essential, while, often, it may be too late to update planned lead-times as
the mistakes have already been made. The last method is formal and is based on
simulation and statistical analysis. Simulation of a multi-product, multi-stage
production environment is used to determine the lot-sizing schemes that minimize
the lead-time error. Although the major objectives of the above mentioned methods
is to minimize the deviations between planned and observed lead-times, they do not,
in general, attempt to determine the lead-times that would result in minimal product
cycle-times. Besides, simulation can be computationally expensive.

2.2. Problem description

A classic job-shop environment, consisting of a large number of work-centres,
each of which may comprise functionally identical machines, is considered. It is
assumed that the facility manufactures complex assemblies in small batches. Each
assembly has numerous levels in its Bill-Of-Materials (BOM), and each
manufactured (or make) part is produced following a unique sequence of operations
(called routing). Production planning is performed by an MRP-based system, for the
given demand and order quantity of each product. It is noted that alternative rout-
ings for make items, although sometimes generated by MRPII, are rarely
implemented, and this methodology does not consider them. The problem is to
determine: (i) a ® xed lot-size for each end-item that minimizes the Work-In-
Process (WIP) costs using a lot-for-lot strategy, (ii) the lead-times and lead-time
oŒsets for all make parts of these end-items, for the above determined lot-sizes
and for an average product mix, such that cycle-times of end-items are minimized.
4826 A. Agrawal et al.

Note that since optimality, or even feasibility, of a production plan can only be
assessed once it is ® nally realized through the shop-¯ oor schedule, we evaluate the
production plans that employ the resulting lead-times and lead-time oŒsets by the
shop performance at the operations scheduling stage. In addition to cycle time,
performance metrics include WIP costs and work-centre smoothness (see
Appendix B for the de® nitions of these metrics).
For the operations scheduling stage, used to evaluate the suggested lead-times
and lead-time oŒsets, the following assumptions are adopted. A batch production
strategy is employed, i.e. (i) the entire batch of a part is processed by a certain
machine before the subsequent operation is commenced; (ii) batches are not
distributed amongst functionally identical machines. This strategy is believed to be
most common in industrial practice. Furthermore, the following assumptions are
made:
. a lot-for-lot strategy is employed for make items;
. extended WIP costs are important compared with production costs (set-up and
processing) (because the assemblies may have signi® cantly long cycle times);
. machines are reliable;
. a machine can perform only one operation at a time;
. an operation can be performed on at most one machine at a given time;
. pre-emption of operations is not permitted; and
. processing times of all operations and due-dates of all ® nal products are
deterministic and known a priori.

3. Approach
As acknowledged earlier, lead-times are dynamic and they consist of the follow-
ing components: (i) set-up time (per batch), (ii) run time (or processing time for a
batch), (iii) move times between operations, and (iv) queue times. While the ® rst two
components are known and the third is somewhat easy to determine (especially when
material handling systems are not the bottlenecks), the queue time is dependent on
the availability of resources and is stochastic in general. The resource availability is,
in-turn, a function of the demand and the scheduling strategy. This study attempts to
assess the queue time for average resource availability such that the cycle time is
minimized while respecting scheduling constraints. Our approach consists of the
following steps.
Step 1. For each end item, determine the optimal batch size, assuming that it is the
only end item in the system.
Step 2. For the batch size computed in step 1, perform detailed scheduling of
operations using LETSA. From this schedule, calculate the lead-times and
lead-time oŒsets of all the make items of the ® nal assembly. This process can
be repeated for each end item.
Step 3. Perform capacity scaling for end items based on average product mix and
their utilization of critical work-centres. This involves determining the
proportion of capacity of critical work-centres that is required by each
end-item, based on the product mix. Average product mix can be obtained
from historical data and forecasts.
 Cycle time reduction 4827

Step 4. In¯ ate all part lead-times of an end item employing the capacity scaling
factors computed in step 3.
Steps 1 and 2 are accomplished by using the LETSA algorithm (Agrawal et al.
1996). It uses a precedence network of operations, which incorporates lead-time
oŒsets, and employs an eŒective critical path heuristic to schedule these operations.
Note that backward scheduling is used in an attempt to minimize the WIP costs.
This section is organized as follows. In section 3.1, we present the concept of
lead-time oŒsets and highlight the bene® ts expected from using them. We brie¯ y
describe the Lead-time Evaluation and Scheduling Algorithm in section 3.2. Step 1
of our approach, optimal batch-sizing, using LETSA is described in section 3.3; this
concurrently accomplishes step 2, lead-time and lead-time oŒset estimation for single
assemblies. The capacity scaling and in¯ ation of lead-times (steps 3 and 4) are
described in section 3.4. A ¯ ow chart of our overall approach is presented in
® gure 1. A detailed account of the inputs and outputs of the software system is
presented later in section 4.
This approach has been veri® ed using the predicated lead-times to develop a
production plan given the demand and the due-dates. Subsequently, the shop-¯ oor
simulation package SIMFACTORY is employed to schedule the work-orders that
have been released according to the production plan. This process emulates an
MRP-based planning and the shop-¯ oor schedule resulting from it. Numerical
studies verify the validity of the approach as well as the bene® t of using lead-time

Static Production Data

Bills-Of-Materials
Routings
Work-centers

Batch Size Backward Scheduling

Algorithm (LETSA)

WIP Cost
Plant Independent
Change N
Min. Cost?
Batch Size

Y Optimal Batch Size,

Lead-Times, LTOs
Average Lead-Time Scaling
Product Mix for Capacity

Data Translator

Plant Dependent

Reformatted
MRP Data

Figure 1. Overall system approach.

4828 A. Agrawal et al.

oŒsets, their impact on reducing overall cycle time, improving workload smoothness,
and reducing WIP costs. These results are presented in section 5.

3.1. Lead-time oVsets

Conventionally, the components of an assembly or sub-assembly are available in
inventory before an assembly work-order is released. In contrast, using lead-time
oŒsets permits the manufacture of such an assembly to proceed concurrently with
the manufacture of some of its component parts.
Consider the Bill of Materials (BOM) structure and the routing information for
the ® nal product A shown in ® gure 2. Lead-time oŒsets take advantage of the fact
that production of a component pk of part pi may be delayed until pk is actually
required in a certain operation of the routing of pi . Thus, the lead-time oŒset of pk is
computed as follows: if pk is required for the ® rst time for operation oij of its parent
part pi , its lead-time oŒset is the diŒerence between the start times of operations oij
and oi1 .
In ® gure 2, it is obvious from the routing of part A that availability of C can be
delayed until it is actually required for operation A.20. Thus, operation C.10 may be
performed concurrently with operation A.10 (assuming another machine in work-
centre 1 is available), thereby improving resource utilization as well as reducing the
cycle-time. In the manufacture of large and complex assemblies, a large number of
component lead-time oŒsets may be realized. The introduction of these lead-time
oŒsets into the planning and scheduling functions yields a number of bene® ts, as will
be shown in section 5.2. It is interesting to note that most current MRP systems do
provide the capability of using lead-time oŒsets.

3.2. The lead-time evaluation and scheduling algorithm

The determination of the minimum cycle-time production schedule is a complex
problem; optimal methods can, at best, solve small dimensional cases. To address
practical problems, an eŒective heuristic has been developed by Agrawal et al.
(1996). It exploits the critical path of the network of required manufacturing
operations to generate a near optimal schedule. This network represents the
precedence relationships among operations and may be obtained from the BOM
structure and the routings of all make items of an assembly, as shown below. This
network can also incorporate the component lead-time oŒsets. Figure 3 shows the

Part #: A
Qty: 1
Type: Make Part Operation Components Processing Workcenter
Required Time*

A.10 B 5 WC #1
Part #: B Part #: C
Qty: 1 Qty: 2 A
A.20 B, C 2 WC #2
Type: Purchased Type: Make

C C.10 D, E, F 1 WC #1

Part #: D Part #: E Part #: F D D.10 G 3 WC #1

Qty: 2 Qty: 1 Qty: 1
Type: Make Type: Purchased Type: Make WC #2
F.10 H 2
F
Part #: G Part #: H F.20 H 3 WC #1
Qty: 3 Qty: 3
Type: Purchased Type: Purchased
*Processing time includes the setup and run times for the entire batch

Figure 2. Bill-of-Materials (BOM) of the ® nal product A and routings of its make parts.
 Cycle time reduction 4829

F.10 F.20 C.10 A.20

D.10 A.10

Figure 3. Operation network corresponding to the example of ® gure 2.

precedence network of operations obtained by the BOM and the routing information
for the product structure shown in ® gure 2.
The network of ® gure 3 shows that operation A.20 can be initiated only after
operations C.10 and A.10 are completed. Similarly, operation C.10 can be initiated
only after operations D.10 and F.20 are completed. The above network explicitly
represents all the precedence relationships required to manufacture ® nal product A.
The operation network of ® gure 3 also incorporates lead-time oŒsets. If such oŒsets
were not used, the network would need to be modi® ed to include the dotted arrow to
indicate that operation C.10 must be performed before A.10, which represents an
additional precedence constraint since operation A.10 is required to be performed
before operation A.20.
Given such a network of a ® nal assembly, a continuous sequence of operations
that starts from the ® nal operation of the ® nished item and terminates at a purchased
item is de® ned as a BOM path. The critical path is de® ned as the BOM path along
which the sum of all operation times is maximal. The length of the critical path
would be the time needed to produce the ® nal product if resources were available
when required. However, due to limited resource availability, the actual product
cycle time is, in general, larger than the length of the critical path. This `in® nite
capacity’ critical path is a ® xed property of a given BOM structure and determines
the lower bound of the product’s cycle time.
The LETSA algorithm generates a feasible schedule with a near-optimal cycle
time. It proceeds in a backward scheduling manner similar to MRP II, in which the
last operation is scheduled ® rst and the remaining operations are scheduled
subsequently while respecting all precedence constraints .
The inputs to the algorithm: the delivery schedule, product structures and routing
data, are used to construct an integrated operation network. Given this network, a
set F is de® ned to include those operations that do not have a succeeding operation.
Generating the schedule comprises four phases: (i) select an operation from the set of
feasible operations F, (ii) select a machine from the required work-centre, (iii)
schedule the selected operation, and (iv) update the operation network and the set
of feasible operations.
In step (ii), the operation to be scheduled is selected from F as follows: the
processing times (set-up plus run) that correspond to each path of the existing
network are computed by summing the processing times of all operations along
this path. The critical path is determined and its ® rst operation, which also belongs
to F, is selected for scheduling. The starting and completion times of the selected
operation are determined from: (i) the starting time of its successor operation, or the
due-date; (ii) the ® rst available time of the machine of the corresponding work-
centre. The operation is then scheduled on that machine. Subsequently, the opera-
4830 A. Agrawal et al.

tion network is modi® ed by deleting the operation just scheduled. Its predecessors
are included in the feasible set of operations F, and the process is repeated until F is
empty, i.e. the schedule is complete. The ¯ owchart of the algorithm is presented in
Appendix A (see also Agrawal et al. 1996). Later, Anwar and Nagi (1997) showed
that an equivalent, yet more e cient way to perform this backward scheduling is ® rst
to compute early ® nish times of the operations network (similar to the critical path
method in project networks). Then, in step (i), select from F, the operation with the
highest early ® nish time.

3.3. Optimal batch sizing using LETSA

LETSA is used to perform the trade-oŒanalysis between processing costs and
WIP costs in batch sizing decisions implemented within an overall lot-for-lot
production strategy. The function of LETSA in this mode is to evaluate the WIP
cost, and once the batch size corresponding to the minimal WIP costs has been
determined, the scheduling results of LETSA also provide lead-times and lead-
time oŒsets of each make item of the assembly. To this end, the algorithm is run
iteratively, varying only the batch size and the resulting WIP cost is evaluated from
the schedule of operations as described in Appendix B. The optimum batch size is
selected to be the one that corresponds to the lowest cost per unit. A typical plot of
the relationship between batch size and cost per unit is shown in ® gure 4. The
assembly tested in this case is a large industrial product. It includes 200 make
items and the lead-times ranged from 57 days (batch size of one) to 522 days
(batch size of 15). Additional details of the run and set-up costs terms are similar
to those in section 5.1. The two curves in the ® gure correspond to the conditions
during scheduling; the top curve was obtained when no lead-time oŒsets were used,
while the lower curve was obtained using lead-time oŒsets. Figure 4 clearly illustrates
the bene® ts of introducing lead-time oŒsets towards lowering unit cost.

3.4. Capacity scaling approach

Based on the schedule generated, the lead-time of each make part can be
determined. Obviously, if production planning is performed using these lead-times,
the schedule provided by LETSA can be replicated in push production. However,

58000
Cost/Unit with LTOs
Cost/Unit without LTOs
56000

54000
Cost/Unit

52000

50000

48000

46000
0 10 20
Batch Size
Figure 4. Implementation of LETSA to optimize batch sizing.
 Cycle time reduction 4831

these lead-time estimates would be valid if the assembly under consideration was the
only product in the system. The presence of other products is considered by in¯ ating
the lead-times of parts to account for the resource capacity consumed by these
products.
In order to determine a robust scaling method that can provide realizable lead-
times and, at the same time, keep cycle-times minimal, we attempted two capacity
scaling procedures to account for multiple products within a shared facility: (i) Exact
Capacity Scaling (ECS), and (ii) Bottleneck Capacity Scaling (BCS). To de® ne these
we will use the following indexes, i for product, j for a make part of product i, and k
for work-centre. For the ECS method, the workload requirements of each product
on each work-centre were computed by summing, for all operations (requiring that
work-centre) belonging to the routings of all make parts of the product, the
processing times (derived from set-up and run times) for the given long-term product
demand. The total workload requirement for each work-centre was determined by
summing the workload requirements over all the products. The ratio of the total
workload requirement of a work-centre to a product’ s workload on that work-centre
was de® ned as the scaling factor Sik . The routing length of each make part was
computed by adding, over its operations (on diŒerent work-centres), the processing
times for the given long term demand, and the ratio of this part’s processing on
work-centre k to the part’s routing length wasPdetermined as Ljk . Then the lead-time
of the make part j was in¯ ated by a factor k …Sik £ Lik †.
For the BCS method, the work-centre with the highest workload requirement for
each product for the given long-term demands was determined individually as the
bottleneck for that product. Then the lead-time of each part of a particular product
was in¯ ated by the ratio of the cumulative workload (over all products) of the
bottleneck to the product’s workload requirement from the bottleneck.
As indicated in the results in section 5.3, we recommend the BCS method to
be superior to the ECS. BCS suggested shorter lead-times that resulted in corre-
sponding shorter cycle-times than ECS in all cases. However, in 3% of the cases
it underestimated the lead-time. In these cases the actual lead-times were 1.5%
longer, and in no case were they more than 2.5% longer than the suggested lead-
time. ECS-suggested lead-times were 21% higher than that of BCS, but they were
realizable in all cases (with an acceptable minor elongation of less than 0.03% in
a few cases).

4. Software implementation
The overall approach for the determination of accurate lead-times and lead-time
oŒsets, and optimal batch sizes, has been implemented in a software system. The
inputs and outputs of this system are detailed in this section. These data were
obtained from an MRP II system and translated automaticall y to conform to the
system input requirements. Some degree of automation has been incorporated into
the user interface to facilitate data input by providing a menu to select the necessary
components for each operation.

4.1. System inputs

Two types of inputs are provided to the system: (i) MRP data, and (ii) user
data.
4832 A. Agrawal et al.

MRP II data
MRP data include complete production information for manufacture d products
as well as detailed information about the manufacturing environment. Speci® cally,
the inputs are:

. item master records;

. bills-of-materials (BOM) for all ® nal products;
. routing information for all make items;
. manufacturing facility information, including a list of work-centres, the
number of identical machines per work-centre, work-centre e ciency and
capacity, set-up and run rates for each work-centre.
User inputs
In addition to the MRP II data, the user is required to provide:

. the batch sizes of all end-items;

. the percentage workload contributed by each product to the manufacturing
resources. This is provided as the long-term production requirements of each
product, and accounts for sharing of the resources amongst multiple products.
Note that it can be computed from historical data and projected forecasts;
. the interest rate at which the work-in-process (WIP) costs are to be com-
pounded;
. component requirements for each operation in a routing. This information is
necessary primarily for the accurate computation of lead-time oŒsets and WIP
costs.

4.2. System outputs

. Operations dispatch list: generated for each work-centre, it provides the start
and ® nish times of each operation performed at the work-centre within the
scheduling horizon.
. Schedule of operations by make item: generated for each make item, it
provides the start and ® nish times for each operation in the corresponding
routing.
. Constrained capacity lead-times and lead-time oŒsets for all make items: these
are calculated from the schedule of operations.
. Cost of each make item: it combines raw material, labour and WIP costs.
. Order times and quantities for raw materials: since LETSA operates in a back-
ward scheduling mode, it provides the appropriate release dates of purchase
orders for raw materials in order to meet the due date of the end items.
. Daily workload for each machine: it is computed in a straightforward manner
from the generated schedule of operations. The work-centre smoothness
pro® les (see Appendix B) can then be calculated from this output.
. Scaled lead-times and lead-time oŒsets for all manufactured parts to be
employed in MRP.
 Cycle time reduction 4833

5. Performance
This section evaluates the eŒectiveness of the proposed method. The ® rst section
examines the bene® ts of using lead-time oŒsets. In the second section, the production
plans that are generated based on the system’ s lead-times are examined under
realistic production conditions, assuming more than one ® nal assembly is being
manufactured concurrently in the facility. For the performance of LETSA and its
eŒectiveness in minimizing cycle time, the reader is referred to Agrawal et al. (1996).

5.1. Bene® ts of using lead-time oVsets

A numerical study was conducted to assess the bene® ts of using lead-time oŒsets
using an example set of 200 randomly generated BOMs. The part, routing, work-
centre, and BOM structure data for all examples were generated randomly. For each
BOM structure, there was an of average of four work-centres in the facility, each
with one to two identical machines. This particular set of examples constituted large
and wide BOMs, characteristic of large assemblies. Average parameters for the
BOMs for these examples were 5 levels, 130 make parts, 300 operations, and
about 200 branches. Routings for the make parts were made up of one to four
operations, chosen uniformly. Each operation had a set-up time per batch between
10 and 30 minutes, and a processing or run time per part between 5 and 10 minutes,
both chosen uniformly. Since the labour rates for set-up and processing were
comparable, the ratio of set-up cost to run cost was between 1 and 6. For lead-
time oŒsets, if a make part was assembled from two or more components, its last
operation in the routing required all component parts, but the preceding operations
required one or two less components. The oŒset for a component would then be
generated from the time diŒerence between the ® rst operation of the parent make
part and the ® rst operation where this speci® c component was required.
LETSA was applied to determine the schedule of operations with and without
LTOs and the results were compared with respect to the following production per-
formance metrics of Appendix B. When LTOs are not used, all components of a
make part are made to be required at the ® rst operation of its routing. It is noted that
the above parameters are conservative and that improved performance characteris-
tics can be obtained by increasing the number of identical machines, operations per
routing, and the number of levels of the BOM.
The following measure was used to quantify the bene® ts of LTOs in terms of
product cycle time or makespan (m) and cost/unit (c). In v* below, the asterisk is m
and c for percentage improvement in cycle time and cost respectively.
…Result†…Without LTOs† ¡ …Result†…With LTOs†
Percentage variation ˆ v* ˆ £ 100: …1†
…Result†…Without LTOs†
For workload smoothness equation (B.1) was used; see Appendix B.
…WS†…With LTOs† ¡ …WS†…Without LTOs†
Percentage variation ˆ vws ˆ £ 100: …2†
…WS†…With LTOs†
P* ˆ Efv*g (mean percentage variation) was evaluated considering all examples
within the example set. Note that for both cases the larger this measure, the greater
the bene® t of using LTOs.
In ® gures 5 through 7, the relationship between P* ˆ Efv*g versus the batch size
is shown. Note that the batch size was used as the independent variable, since it is
4834 A. Agrawal et al.

provided by the user and directly impacts upon the bene® ts resulting from the use of
LTOs. The relationship P* ˆ P* (batch size) is represented by the continuous line in
® gures 5 through 7. In addition, the bounds within which 95% of the cases were
contained (i.e. 95% con® dence intervals) are represented by the closed and open
circles, above and below the continuous line respectively.
Figure 5 shows the eŒect of LTOs on product cycle time. It can be seen that for
small batch sizes, the improvement is higher and stabilizes to about 7.2% for the
large batch size cases. The improvement in cycle time with the use of LTOs is
expected because operations with LTOs are delayed with respect to the due-dates
speci® ed by conventional planning, without aŒecting the product cycle-time. This
increases the degree of ¯ exibility while assigning the operations to work-centres,
facilitating better resource utilization. The eŒect of batch size on Pm is attributed
mainly to the fact that smaller batch sizes require lower processing times. In this
case, there exists a higher degree of ¯ exibility while assigning operations to work-
centres, since the shorter operations can better ® ll idle times between scheduled
work-centre operations. However, as the batch size increases, processing times
% Makespan Improvement (P )
m

9.0

8.5

8.0

7.5

7.0

6.5

6.0
0 5 10 15 20 25 30 35 40 45 50 55
Batch Size

Figure 5. EŒect of using lead-time oŒsets on product cycle time: Pm versus batch size.

25.0
% Cost Improvement (P )
c

20.0

15.0

10.0

5.0

0.0
0 5 10 15 20 25 30 35 40 45 50 55
Batch Size

Figure 6. EŒect of using lead-time oŒsets on product cost/unit Pc versus batch size.
 Cycle time reduction 4835

increase almost proportionally , making it more di cult to utilize inter-operation idle
time.
Figure 6 shows that for a batch size of 5, the improvement in product cost/unit is
approximately 2.5% but increases to about 21% for a batch size of 50. Component
parts with LTOs can be introduced after the assembly of the parent part has started.
Thus, expensive parts do not have to wait for extended periods of time in inventory,
reducing WIP costs. As seen from the ® gure, savings in WIP costs realized in our
experiments continue to increase proportionately with batch size. This is due to the
compounding of interest accrued on delayed value addition. When LTOs are
utilized, early savings in oŒsetting a component’s value addition continue to be
compounded over the remaining duration of the makespan. Further, the makespan
increase is roughly proportional to batch size because the processing times of
operations as well as the work-centre idle times increase almost proportionally
(NB processing time for an operation is set-up time ‡ batch size £ run time).
Consequently, the savings in extended WIP costs continue to increase with batch
size. It has to be noted that the slope of the savings curve would depend on the raw-
material costs and rate of compounding. Our experiments use some exaggerated
values for material costs and 20% annual percentage rate compounded hourly,
which may be high for some industrial situations. This is simply for illustrative
purposes and to amplify the bene® t to WIP costs when using LTOs. Finally, an
alternative view of these increasing savings in unit cost due to LTOs can be seen in
the diverging right halves of the unit cost curves shown in ® gure 4.
Figure 7 shows that the introduction of lead-time oŒsets improves the smooth-
ness of work-centre load. This is due to staggering the requirements of the
components of an assembly such that they are not all due at the same time (which
causes an unnecessary peak in the workload). Note that with increasing batch size
the processing times increase and, thus, the LTOs increase almost proportionally.
Consequently, the ¯ exibility in ® lling the idle times, which results from the
introduction of LTOs, is unchanged. Figure 7 con® rms this hypothesis, showing
that despite the variation in product batch size, the advantage of using LTOs
remains almost constant.
From the above numerical results and discussion it can be clearly seen that the
introduction of lead-time oŒsets in planning and scheduling results in signi® cant
advantages in terms of product cycle-time and cost per unit. In addition, the result-

9.0

8.5
Workload Smoothness
% Improvement (P )
ws

8.0

7.5

7.0

6.5

6.0
0 5 10 15 20 25 30 35 40 45 50 55
Batch Size
Figure 7. EŒect of using lead-time oŒsets on work-centre load pro® le: Pws versus batch size.
4836 A. Agrawal et al.

ing improvement of workload pro® les is invaluable in the case of alleviation of peaks
and valleys on bottleneck resources.

5.2. Overall system evaluation: validation of proposed lead-times

The proposed approach has been veri® ed by performing production planning
based on the derived lead-times, and then using a discrete-event simulation tool
(SIMFACTORY) to release work-orders into the system according to the suggested
lead times. This production planning and scheduling system emulates MRP II-based
planning. To restate this in detail, this is what our simulation study attempts to do.
We obtain our lead time and oŒset information from the approach developed in this
work. We then generate an arbitrary shipping schedule for a mix of end products.
Due dates in this shipping schedule are generated randomly, while the delivery
quantities match the respective optimal batch sizes. Random due-dates create a
real-life situation of arbitrary shop loading in response to customer orders and
permit a study of the robustness of our derived lead times. Now, an MRP run is
emulated using the BOM, routing and lead time information. Material is released
into the simulation system using the backward MRP calculated release dates.
Manufacturing operations on the work-centres are scheduled using the commonly
used ® rst-come ® rst-serve (FCFS) shop-¯ oor dispatching rule. We let the production
proceed in time as the simulation clock advances, and observe the completion times
of the diŒerent end items, and compare them to the desired due dates given in the
original shipping schedule.
It is important to note that discrete-event simulation involves construction of an
elaborate model, which is a labour intensive and error-prone process. To overcome
this limitation, a translator routine was developed to automatically generate the
model from the available input data. This is of signi® cant practical importance,
rendering simulation a useful tool in large job-shop applications. Further, this dis-
crete event simulation approach can be readily used to study additional stochastic
events like machine failures and rework/scrap rates. We will, however, remain
focused on our validation under a deterministic shop operation.
This numerical study was performed considering 16 sets of examples, with each
set consisting of 10 unique end products. The ® rst step of our overall approach,
determining optimal batch sizes, was performed by the simple enumeration pro-
cedure detailed in section 3.3. In the second step, each product was considered
individually and LETSA was employed to provide independent lead-times and
lead-time oŒsets for the make parts. The two capacity scaling procedures described
in section 3.4 were employed to derive estimated lead-times for the parts in a shared
facility; this concludes steps 3 and 4 of our procedure. We considered the shipping
schedule to be generated arbitrarily where one batch of each product in its respective
optimal batch size was assumed due at a speci® c due-date. An MRP II-type
algorithm was employed to schedule the release of purchased orders and make
orders in our simulated factory. Push simulation was performed using the FCFS
rule, but respecting order release times. The completion times of the 10 end-products
were observed from the resulting simulation run, and compared with the planned
lead-times to assess the on-time performance.
Table 1 presents the results of example set 7, consisting of 10 end products.
Column 1 contains the example product identi® cation number. Column 2 contains
the lead-times that are determined from LETSA’ s schedule when all products are
considered in a combined manner. Column 3 contains the estimated lead-time using
 Cycle time reduction 4837

2 9
1 LETSA 3 4 5 6 7 8 PUSH
Ex. ID combined LT/EC SF/EC/PP SF/EC/MP LT/BC SF/BC/PP SF/BC/MP SF/0

7.1 12 904 18 881 17 679 18 882 16 638.9 16 220.5 16 637.9 16 684.1

7.2 15 997 18 268 17 975 18 268 17 279.6 17 139.3 17 279.8 16 275.4
7.3 12 984 21 818 21 454 21 818.2 22 270.7 21 907 22 271.1 16 573.3
7.4 15 253 18 806 18 358 18 806.9 16 939.1 16 636.6 16 939.2 15 344
7.5 16 182 16 907 16 746 16 906.6 16 203.1 16 045.4 16 203 16 755.5
7.6 5563 20 462 18 982 20 458.5 18 391.8 17 335.1 18 391.6 16 655.2
7.7 16 132 22 159 18 092 22 158.8 16 802 15 816.7 16 802 16 582
7.8 16 058 23 987 21 655 23 987.4 16 558.1 15 840.1 16 558.2 15 832
7.9 16 113 20 503 19 970 20 503.1 17 065.7 16 534.3 17 066.6 16 400.9
7.10 16 420 19 849 19 115 19 849 15 168.2 14 876.7 15 168.3 16 619.2
Makespan 17 190.9 23 826.6 23 987.4 22 130.6 22 271.1 16 755.5

Table 1. Estimated lead-times and realized lead-times by MRP planning with FCFS dis-
patching.

the exact capacity scaling procedure. Column 4 contains the lead-times that are
realized by SIMFACTORY with FCFS dispatching when purchased parts (raw
materials) are released according to the lead-time estimates of column 3. Column
5 contains similar information to column 4, but with both purchased as well as make
part orders released according to column 3 lead-times. This emulates a true MRP
schedule. Columns 6 through 8 have similar information as columns 3 through 5
respectively, only that the bottleneck capacity scaling procedure is employed in lead-
time estimation. Finally, column 9 contains a push schedule when all purchased
parts are released at time `0’ .
Similar results were obtained for all 16 sets of examples. Without presenting the
details of these rather similar results, we summarize our ® ndings as follows. LETSA
combined and PUSH resulted in very close makespans (within 0.1% on the average),
which were the smallest among the production scheduling methods. This is again
consistent with the results found by Agrawal et al. (1996). In these examples, LETSA
combined was never more than 5.6% higher than PUSH, while, in one case, PUSH
was about 48% higher than LETSA. More importantly, even with comparable
makespans, PUSH resulted in signi® cantly higher WIP costs, making it an undesir-
able production strategy. This is further substantiated from the results that the
individual lead-times in a PUSH schedule were about 20% higher than the ones
derived from a LETSA schedule, on average.
The estimated lead-times due to the exact capacity scaling (ECS) method were
always realized very accurately, when purchased as well as make part orders were
released on time (column 5) and were never underestimated by more than 0.02% .
When these estimated lead-times were used to release only purchased parts while
make orders were released as early as possible due to FCFS and PUSH production
(column 4), the realized lead-times were always smaller, and smaller by 6% on
average. This indicates that this capacity scaling method is too conservative .
Along similar lines, the estimated lead-times due to the bottleneck capacity
scaling (BCS) method were always realized very accurately when purchased as
well as make part orders were released on time (column 8). In addition, the estimated
lead-times were only underestimated by 0.07% on average and never more than
2.4% . When these estimated lead-times were used to release only purchased parts
(column 7), the realized lead-times were generally smaller, and smaller by 3.5% on
4838 A. Agrawal et al.

average. The realized lead-time was never more than 2.6% higher than the estimated
one. We believe that the BCS capacity scaling method is not overly conservative
because any further stringency will only cause realized lead-times to exceed estimated
ones. Further, given that the lead-times estimated by the BCS method were 21%
lower than the ECS method, BCS is clearly superior to the ECS method.
Even while we have presented an accurate estimator for lead-times that
are closely realizable and not overly conservative, it has to be recognized that
MRP planned production in this manner results in overall makespans that are
about 28% higher than that of LETSA combined (or PUSH). This is not surprising
in that static lead-time-based planning and FCFS dispatching is not expected to
outperform a specialized scheduling method such as LETSA, yet the diŒerence
may be signi® cant. In summary, since MRP-based planning and scheduling remains
a popular tool in industry, we have proposed a systematic method to determine
accurate lead-time data that will result in low product cycle times.

6. Conclusions
Facing global competitiveness, manufacturers are hard pressed to minimize
product cycle times and costs, and meet deliveries on time. In the batch production
environment for assembled products, MRP/MRP II still remains a popular tool in
industry to accomplish production planning and scheduling. In addition, most
MRP systems perform planning and scheduling using a priori lead-times (although
some recent systems include limited ® nite scheduling capabilities). Underestimating
and overestimating these lead-times can lead to a variety of undesirable circum-
stances.
In this paper we have revisited this classical and industrially relevant problem of
determining accurate estimates for item lead-times because it is at the heart of a
successful MRP schedule. At the backbone of our proposed approach is a recent JIT
production methodology for assembled products, LETSA (Agrawal et al. 1996) that
performs detailed backward scheduling of one or more large assemblies in a common
facility, with an objective of minimizing the cycle time. It is obvious that, now, if
lead-times for items are determined from this detailed schedule and employed in an
MRP system to plan the release of raw-materials and manufacturing orders, push
production can accomplish on-time delivery at minimal cycle time. That is, a back-
ward generated schedule can be replicated in a forward production run. Despite this,
there is an important impediment: dealing with multiple end-items in a dynamic
shipping schedule while the lead-times in MRP are static. It would obviously not
be practical to repeat this process for every shipping schedule. To overcome this
remaining problem, we proposed a capacity scaling approach that in¯ ates, according
to a long-term capacity utilization, the LETSA lead-times determined for each end-
item separately. Many numerical studies demonstrate the e cacy of this approach.
Apart from a methodology to determine realistic lead-times that minimize prod-
uct cycle times, this paper makes the following contributions. It demonstrates, in
quantitative terms, the bene® ts of introducing lead-time oŒsets in their ability to
minimize cycle time and cost, and to result in more balanced workloads on resources.
It also presents an enumeration approach to determine the optimal batch sizes of
end-items for minimal WIP costs using LETSA. These lot sizes strike a balance
between the set-up cost and WIP inventory costs. Thus, important objectives of
minimizing cycle time for on-time delivery and minimizing schedule costs can be
accomplished simultaneously.
 Cycle time reduction 4839

Appendix A: Lead-time Evaluation and Scheduling Algorithm (LETSA)

Input BOM, routing, workcenter and part data

Input delivery schedule

Convert BOM and routing data

into an operation network

Determine set of feasible operations

Compute critical path from current network

Select operation to be scheduled

Identify resource required

and select specific machine

Schedule operation on resource

Delete operation from network

and update set of feasible operations

Is
set of N
feasible operations
empty
?

Compute raw-material, labor and WIP costs

END

Figure A1. LETSA ¯ ow chart.

4840 A. Agrawal et al.

Appendix B: Performance metrics

Product cost per unit
In general, a product’ s cost is composed of material cost, direct labour, work-in-
process costs, and overhead costs. In this metric, we consider a subset of production
costs that are a function of the operations schedule, i.e. work-in-process costsÐ we
disregard costs that cannot be controlled by the production planning and scheduling
function. The material and direct labour costs can also be computed, but they are
constants in our case because both the material and work (labour) content of parts
remains unchanged with the actual production schedule.
The cost of production as well as work-in-process (WIP) costs are also computed
from the schedule of operations. For calculating WIP costs, a rate of interest per
annum and a certain compounding period is assumed. Figure B1 illustrates the
method used to compute the cost of a part processed through two operations with
idle time between them. Note that the accumulation of labour cost during an
operation is assumed to be linear, and interest during the operation is compounded
assuming that the principal consists of the mean value of the part(s) and labour
costs. Thus, for operation 10 shown in ® gure B1 the total cost comprises three
components, the initial raw material cost, the labour cost (linear component), and
the interest (non-linear component). During the idle time between operations 10 and
20, the value of the part is incremented by the amount of interest accrued on the
value of the part at the end of operation 10, for the duration of the idle time. This
procedure is repeated for all parts to obtain the total WIP cost.
Work-centre utilizationÐ smoothness of work-centre load pro® le
The measure of workload smoothness can be evaluated from the load pro® le of
each work-centre. Figure B2 shows such a pro® le over the scheduling horizon. The
daily load requirement is plotted against the number of days in the scheduling
horizon and the average workload requirement is also indicated.
From this pro® le, the workload smoothness is calculated as

Final Part WIP Cost

Cost

Total Cost

Labor Cost

WIP Cost
WIP Cost

Cost Total Cost

Labor Cost

WIP Cost
Initial
Part Cost

Operation 10 Idle Time Operation 20 Idle Time

Time

Figure B1. Method used for computing part cost (material, labour and WIP costs).
 Cycle time reduction 4841

24
Average Workload
Load requirement (hrs/day)

Requirement
20

15

10

0
0 20 40 60 80 100
Days

Figure B2. Work-centre load pro® le.

Area above average workload

WS ˆ 1 ¡ : …B1†
…Average workload†*…Length of scheduling horizon†

The closer this value is to 1, the smoother the work-centre loading.

References
A GRAWAL, A ., H ARHALAKIS, G ., M INIS, I . and N AGI, R ., 1996, Just-In-Time production of
large assemblies. IIE Transactions on Scheduling and Logistics, 28, 653± 667.
A NWAR, M . F . and N AGI, R ., 1997, Integrated lot-sizing and scheduling for just-in-time
production of complex assemblies with ® nite set-ups. International Journal of
Production Research, 35, 1447± 1470.
H UGE, E . C ., 1979, Managing manufacturing lead-times. Harvard Business Review, 57, 116±
123.
J OHN, R . S., 1985, The cost of in¯ ated planning lead-times in MRP systems. Journal of
Operations Management, 5, 119± 128.
K ANET, J . J ., 1982, Towards understanding lead-times in MRP systems. Production and
Inventory Management, 23, 1± 14.
M ELNYK, S. A . and P IPER, C . J ., 1985, Leadtime errors in MRP: the lot-sizing eŒect.
International Journal of Production Research, 23, 253± 264.
O RLICKY, J ., 1975, Material Requirements Planning (New York: McGraw Hill).
P ETERSON, L . D ., 1975, Design considerations for improving the eŒectiveness of MRP.
Production and Inventory Management, 16, 48± 68.
P LOSSL, G . W . and W ELCH, W . E ., 1979, The Role of Top Management in the Control of
Inventory (VA: Reston Publishing).
SCHUCHTS, D . R ., 1979, Give MRP a kiss: keep it smooth and simple. Production and
Inventory Management, 20, 68± 80.
STEELE, D . C ., 1975, The nervous MRP system: how to do battle. Production and Inventory
Management , 16, 83± 88.
VOLLMAN , T . E ., BERRY, W . L . and W HYBARK , D . C ., 1992, Manufacturing Planning and
Control Systems, Third edition (Richard D. Irwin).
W ALISZEWSKI, D . A ., 1979, Lead-time reduction in multi-¯ ow shops. APICS 22nd Annual
Conference Proceedings, 209.
W EMMERLOV, U ., 1979, Design factors in MRP systems, a limited survey. Production and
Inventory Management, 20, 15± 35.
W IGHT, O ., 1970, Input/output control: a real handle on lead-time. Production and Inventory
Management , 11, 9± 31.