International Journal of Production Research

Vol. 49, No. 11, 1 June 2011, 3319–3342

RESEARCH ARTICLE
A hierarchical approach to FMS planning and control with
simulation-based capacity anticipation
Erinç Albey and Ümit Bilge*

Department of Industrial Engineering, Boğaziçi University,

Bebek, 34342 Istanbul, Turkey
(Received 13 August 2009; final version received 26 March 2010)

In this study, a hierarchical production planning and control system (HPPCS-

FMS) framework for a Flexible Manufacturing System (FMS) is developed. The
HPPCS-FMS is a three-level hierarchy embedded in a closed-loop structure where
the main feedback is the anticipated effective capacity. The HPPCS-FMS exploits
the flexibility inherent in the manufacturing system to respond effectively to a
frequently changing demand mix. All levels benefit from flexible process plans as
a means to use capacity effectively. Flexible process plans offer not only
alternative machines for operations but also alternative sequences of operations
for producing the same work piece. To implement the framework, the problems at
each level are identified and modelled; the integration among the levels is defined
and a simulation-based capacity anticipation mechanism is developed. The
proposed HPPCS-FMS is tested under different scenarios. The results demon-
strate that capacity anticipation considerably improves the performance.
Keywords: flexible manufacturing systems; hierarchical production planning;
effective capacity anticipation; mathematical programming and simulation
applications in FMS; FMS loading with alternative routing

1. Introduction
For many production planning and control domains, the hierarchical planning approach is
regarded as more suitable, tractable and practical than a monolithic approach because of
its handling of the inherent set of complex and interrelated decisions with different time
intervals and detail levels. The basic concepts of the hierarchical planning paradigm date
back to Anthony (1965), and an extensive literature has evolved along three main
milestones. The Hierarchical Production Planning (HPP) models, based on Hax and Meal
(1975), offer a pure top-down hierarchy of deterministic optimisation models where the
solution to the higher-level problem constrains the succeeding lower-level problem.
The solution obtained at a higher level based on certain aggregations is disaggregated
in the lower levels. In the hierarchical control approach introduced by Bertrand et al.
(1990), the information flow among levels is not in a single direction and allows the lower
level to improve its objective by negotiating with the higher level concerning acceptable
workloads. The organisational planning hierarchy (OPH) developed by Schneeweiss
(1995), on the other hand, proposes building an anticipation of the lower-level model
within the higher-level model to achieve integration among the hierarchical levels.

*Corresponding author. Email: bilge@boun.edu.tr

ISSN 0020–7543 print/ISSN 1366–588X online

ß 2011 Taylor & Francis
DOI: 10.1080/00207543.2010.482570
http://www.informaworld.com
3320 E. Albey and Ü. Bilge

The concept of anticipation is crucial because the effective capacity of a manufacturing

system is often much lower than its theoretical capacity (i.e. nominal capacity defined
by Elmaghraby (1991)). The effective capacity that is actually available to be utilised
during the execution (i.e. available capacity (Elmaghraby 1991)) is highly dependent on the
operational dynamics (work-in-process, setups, flow patterns, dynamically changing
bottlenecks, etc.), which are in turn dependent on planning decisions such as the
product mix, release quantities, work allocation, batching and sequencing. Ignoring this
circularity in planning and expressing the capacity as fixed exogenous input parameters in
the resource capacity constraints at the aggregate level results mostly in solutions that are
impossible to execute in the lower level. Thus, estimation of the effective capacity becomes
a dilemma that requires knowledge of shop floor status and operational dynamics.
Recognition of this challenge has resulted in several streams of research towards achieving
an anticipation of the lower level. Developing anticipation mechanisms based on
artificial neural networks (Rohde 2004), the exponential smoothing method (Selcuk
et al. 2006), clearing functions (Graves 1986, Karmarkar 1989, Hwang and Uzsoy 2005,
Asmundsson et al. 2006) or simulation (Hung and Leachman 1996, Byrne and Bakir 1999,
Kim and Kim 2001, Byrne and Hossain 2005, Venkateswaran and Son 2005) are some of
these approaches. While these models and frameworks have been developed conceptually,
the relatively few performance evaluation results reported are mostly for classical
manufacturing environments (Pahl et al. 2007).
Flexible Manufacturing Systems (FMSs) have attracted the attention of manufacturers
in recent decades due to their suitability for today’s highly dynamic and customer-oriented
manufacturing environments. Planning and control problems related to FMSs have
attracted researchers since the late 1970s due to their multi-dimensionality, abundance
and complexity. Over the years, several studies have tried to uncover the dynamics of
such systems by offering definitions, models, and solution approaches. However, most of
these studies usually concentrate on solving the short-term planning problems regard-
ing the FMSs (or FMS set-up problems), and very little effort has been expended to fit
them within a hierarchical framework that integrates aggregate planning, short-term
planning and control. Furthermore, the above circularity syndrome has not yet
been studied in the FMS context, although in a manufacturing environment such as the
one targeted by an FMS, with a large variety of items to be produced using several
shared resources along diverse routings, it is a real challenge to anticipate the effective
capacity.
This study aims to develop a generic approach for the overall production planning and
control activity in an FMS possessing several types of manufacturing flexibility, such as
the product-related and process-related flexibilities as defined by Benjaafar and
Ramakrishnan (1996). The proposed three-level hierarchical framework works in a rolling
horizon environment and uses simulation to anticipate the effective capacity of the FMS
under a given plan, thus offering a closed-loop structure.
This architecture is first implemented as a generic test-bed that allows plugging in
different models and solution methodologies at each level, along with different loop and
feedback mechanisms. Specific models and approaches are then proposed for each level, as
well as a specific mechanism that forces convergence of the simulation-based capacity
anticipation methodology. The proposed FMS planning tool is tested under different
scenarios. The experimental results demonstrate the merits of a hierarchical structure with
capacity anticipation capability.
The basic novelties of the study can be summarised as follows.
 International Journal of Production Research 3321

. A new aggregate model specifically developed for an FMS is proposed. The model
recognises tools as critical aggregate resources along with machines, and also
explicitly takes the inherent product-related flexibility (i.e. alternative process
plans) of the FMS into account in utilising the aggregate capacity effectively.
. A new FMS loading model is proposed. The model offers a network-flow
structure to allow handling of alternative process plans as well as the work-in-
process (WIP) so that it can effectively be used in a multi-period planning
environment.
. Using simulation as a planning tool for anticipating the effective capacity in an
FMS context is quite different from a similar approach for the classical
manufacturing domain. The fact that the tool configurations (i.e. loading) of
the machines in an FMS are variable and may change at every anticipation
iteration requires specific attention.
. In defining the closed-loop hierarchical planning structure, particular attention is
given to the integration among the levels, elaborating the timing of decisions that
define flexibilities to be utilised on the shop floor.
. Moreover, the present study is one of the very few that tries to assess the relative
performance of capacity anticipation through experimentation.
The paper continues in Section 2 with a discussion of the hierarchical structures in the
FMS planning and control literature and a review of some simulation-based iterative
approaches to capacity anticipation. In Section 3, after an overview of the proposed
hierarchical framework, detailed descriptions of the models at each level and the proposed
effective capacity anticipation mechanism are provided. Experimental studies presented in
Section 4 compare the performance of hierarchical structures with and without capacity
anticipation. Concluding remarks are presented in Section 5.

2. Literature review
2.1 Production planning in FMS
Production planning in a classical manufacturing system is typically modelled as a
two-level hierarchy consisting of an aggregate planning level and a detailed scheduling/
control level (see the review of Venkateswaran and Son (2005)). For an FMS, on the other
hand, the classical hierarchy should be adapted to model the specific features of such a
system. Namely, a medium level to handle the problems that are specifically defined for
FMSs is necessary. The problems in this category are related to the management of the
flexibilities the system possesses. These problems can be called FMS set-up problems, and
they are well defined in the milestone paper of Stecke (1983). The most critical of these are
the selection of parts to be processed simultaneously and the configuration of the FMS for
these parts. The latter, which is referred to as the loading problem in the FMS literature,
covers the allocation of tools and operations to machines. The upper and lower levels in
the hierarchy also need some modifications in the FMS context. The upper level should
incorporate modelling features to represent the inherent flexibility. The lower level should
effectively utilise the flexibility given by the FMS configuration through various
operational control decisions.
While the adaptation of aggregate production models to the FMS context is rather
neglected in the academic literature, set-up problems have received considerable attention.
Most of these studies tackle either a single problem, or a subset of these problems.
3322 E. Albey and Ü. Bilge

However, as noted by Nof et al. (1979), among others, these problems are interdependent.
A few studies, such as those of Denizel and Erenguc (1997), Atlihan et al. (1999), and
Sawik (2004), take a monolithic approach and formulate the integrated FMS set-up
problem as a large mixed-integer/linear programming model. In another line of research,
an integration among the sub-problems is sought, either in a sequential or an iterative
form (Bastos 1988, Co et al. 1990, Chung and Chien 1993, Sodhi et al. 1994, Chandra
1995, Lee et al. 1997, Nayak and Acharya 1998).
Bastos (1988) proposes a short-term planning model consisting of batching, routing
and simulation modules. The first module translates a given weekly plan into a batch of
parts to be assigned to the next work shift by means of an LP model. In the second module
an iterative heuristic is proposed. Although not actually implemented in the study, the role
of the simulation module is to execute the plan and it is not used as an anticipation tool.
Any feedback to a higher planning level that gives the initial weekly plan is not
mentioned either.
Co et al. (1990) develop a MIP formulation for batching, loading and tool
configuration problems and propose a four-step heuristic solution procedure, which can
be considered as a pure top-down hierarchy for FMS set-up. Nayak and Acharya (1998)
propose another three-step sequential heuristic procedure: (i) part type selection;
(ii) machine loading; and (iii) part type volume determination. The authors consider
maintaining a high routing flexibility as an important issue in part type selection and
reflect several flexibilities in their objective function. These two studies provide a solution
approach for the set-up problems in FMS, and hence can be used in the implementation of
the medium level in the FMS hierarchy.
The short-term production planning (STPP) system proposed by Chung and
Chien (1993) considers due dates, production requirements, process plans and layout
information. They approximately solve the production planning problem using an
iterative approach. The output of STPP is tested by means of real-time scheduling software
(DOPS) in the presence of some manufacturing flexibilities set by STPP. However,
DOPS does not provide any feedback to the STPP in terms of effective capacity or the
feasibility of the production plan provided by STPP. This issue is suggested as a future
study direction.
Lee et al. (1997) consider a multi-period part selection and loading problem by means
of an iterative heuristic. However, the developed model does not allow the carrying of
inventory or backorder to consecutive periods and employs subcontracting in order to
meet unmet demand, thus the periods are independent of each other.
None of the studies discussed above presents a complete hierarchical structure that
covers all planning problems. In that respect, the study of Sodhi et al. (1994) can be
mentioned. The hierarchical structure presented in their study is composed of four levels
operating in a rolling horizon mode. The time scale of the horizon is decreasing proceeding
down the hierarchy. The aggregate level considers machine and material handling
resources and involves inventory, backlogging, production, and subcontracting costs.
Given the production quantities for the first period, level 2 decides on the tool loading for
each sub-period (shift) and allocates the production requirement among the shifts. Level 3
tries to improve on the configuration of the first shift to obtain an FMS with greater
routing flexibility and operation allocations on machines are determined. The final level
corresponds to short-term control of the FMS under given operational control strategies.
Although a rolling horizon structure is sought, the initial condition (i.e. WIP)
representation is neglected in the models.
 International Journal of Production Research 3323

In a series of papers, Sawik studies flexible flow shops using the mathematical
programming approach. Sawik (1998) investigates the short-term planning problems for a
flexible assembly system. Given a set of assembly jobs, station workloads are balanced
using a linear relaxation-based heuristic and then, given the task allocation among
stations, assembly plans and routes are selected based on a network-flow model. Sawik
(2004) studies the loading and scheduling of a limited-buffer flexible assembly system with
no backtracking. First, using a monolithic approach, a mixed-integer programming (MIP)
model that tackles the two problems simultaneously to minimise the makespan is
proposed. In the second approach, two separate MIP models are used in a sequential
manner to find a balanced machine loading and a corresponding schedule that minimises
the makespan. Sawik (2006) presents a hierarchical framework for a flexible flow shop and
integer programming formulations for the long-term assignment of customer orders to
planning periods, the short-term task–machine assignment, and, lastly, the scheduling of
multi-capacity machines. The existence of a scheduling component within a monolithic
body resolves the capacity anticipation issue on an optimal basis. However, trying to solve
the scheduling problem with the time horizon required by the loading decision inevitably
increases the size of the problem and, in spite of the advances in commercially available
software for MIP, the computational effort required to solve realistic problems using the
monolithic approach is usually prohibitive, as remarked by Sawik (2004). Having a
scheduling module within a hierarchical framework, on the other hand, does not amount
to capacity anticipation unless there is an in-built closed-loop structure to remove the
infeasibility in the lower levels and to improve the bounds on capacity used in the
upper levels.
An interesting study that makes an effort to anticipate the effective capacity using a
closed-loop structure is that of Chandra (1995). The author considers allocating part types
to alternative routes for a given fixed mix of parts and known machine tooling, thus the
setting is actually not different from a classical job shop. The proposed solution procedure
iterates between a mathematical model and a queuing sub-model. At each iteration the
queuing sub-model anticipates bottleneck machines and the average queue lengths for
the release given by the mathematical model. However, this study ignores the aggregate
and the medium level problems such as determination of part types and configuration
of the FMS.

2.2 Simulation-based capacity anticipation

Available capacity is a property of the state of the production facility and consequently
heavily depends on the product mix, workload, and operational decisions such as routing,
sequencing, etc. This dependency is even more complex in unsynchronised shops where the
variety of products follow diverse routings. It is very difficult to express these complex and
often nonlinear dependencies in an analytical form in production planning models.
A natural solution to this is to use simulation to represent, with close conformity, the
complex dynamic operational behaviour of the facility, while retaining the well-behaving
linear structure of production planning models. A number of studies in the literature
propose iterative approaches where the solution to the production planning model is fed
into a simulation model. The estimates obtained from the simulation model are in turn
used to update the production planning model and the iterations continue until
convergence. Although these iterative schemes have been developed for the classical
3324 E. Albey and Ü. Bilge

manufacturing domain, they have inspired our study and they will be briefly
discussed below.
Byrne and Bakir (1999) and Byrne and Hossain (2005) use simulation to estimate
machine capacities. Their procedure stops when the first attainable production plan is
obtained. Although the authors experience convergence within a reasonable number of
iterations in their single test problem, convergence cannot be guaranteed in general.
Moreover, the procedure has the risk of stopping with a too low effective capacity
estimate. Hung and Leachman (1996) collect flow time statistics instead of machine
capacities, while simulating a production plan. Iterations continue until flow time
estimates converge. Irdem et al. (2008) study several stopping criteria for the Hung and
Leachman (1996) method and remark that the convergence behaviour of iterative schemes
that utilise LP models is not well understood and problematic in general. Kim and Kim
(2001) update both machine capacities and flow times by collecting relevant statistics
during simulation, until the output quantities of the simulation results and LP solution
reach agreement. The combined updating mechanism is claimed to perform better than
Byrne and Bakir (1999) in terms of iterations required for convergence and in terms of
number of parts produced. Venkateswaran and Son (2005) present a hybrid simulation-
based hierarchical production planning architecture consisting of system dynamics
components for the aggregate level planning and discrete event simulation components
for shop level scheduling. Feedback control loops are employed at each level to monitor
the performance and update the control parameters based on the cycle time of the
products.

3. Proposed generic HPPCS-FMS architecture

3.1 Description of the FMS
The FMS under consideration comprises one or more groups of machines where the
machines in a group are identical (i.e. CNCs). Each CNC has independent and local input
and output buffers, automatic tool changing and material handling capabilities. Tool
magazines are of moderate size so that tool seek times can be considered negligible relative
to processing times. Each processing operation requires a designated tool type. For each
tool type a number of copies is available. There is no on-line tool transfer system, so the
tool magazines are loaded with tools during periodic set-ups according to the next batch of
product types to be produced. Tool magazine configurations remain unchanged until the
next period and the new batch (batching mode FMS set-up, see Grieco et al. (2001)). The
material handling system that transfers parts to and from the CNCs consists of a fleet of
Automated Guided Vehicles (AGVs) and/or a closed-loop conveyor. The part types that
are processed in this FMS have flexible process plans such as the one depicted in Figure 1.
The nodes in the figure represent different operations on the process plan, and at each
node, candidate machines that can carry out the operation are indicated. Upon completion
of an operation, an on-line part routing (or part dispatching) decision is required to choose
an appropriate operation–workstation pair among the alternatives offered by the flexible
process plan. The flexible process plan shown in Figure 1 follows the product-related
flexibility definition given by Benjaafar and Ramakrishnan (1996). Product flexibility
refers to the variety of manufacturing options associated with a product. The product
flexibility is further classified into three types: operation flexibility is defined as the
possibility of performing an operation on more than one machine; sequencing flexibility
 International Journal of Production Research 3325

Figure 1. A sample flexible process plan demonstrating operation, sequencing and processing
flexibilities (from Bilge et al. (2008)).

relates to the possibility of interchanging the sequence of operations; and processing

flexibility is defined as the possibility of producing the same work piece with alternative
sequences of operations. The process plan shown in Figure 1 demonstrates all three types
of product flexibility. Note that, before the tool set-up (the loading problem solution) is
designated, only sequence and processing flexibilities for a given product type are known.
The operation flexibility (machines linked to each operation node in Figure 1) can only be
assessed after solving the loading problem. The product flexibility is a potential flexibility,
while its utilisation during execution is often called routing flexibility, i.e. the ability of a
manufacturing system to use multiple alternate routes to produce a set of parts. The
FMS is run under computer control by a real-time Shop Floor Control (SFC) system
that recognises and utilises different dimensions of the potential product flexibility. The
SFC system inputs parts into the system according to the part release plan given by
a higher planning level and controls the flow of parts throughout the system by taking
on-line operation control decisions based on the dynamic shop status (i.e. resource
conditions), static shop data (i.e. machine configuration) and product data (i.e. flexible
process plans for the parts). The basic decisions involve operation and machine selection
(part routing), part sequencing on machines, and the management of the material handling
system (i.e. AGV matching, dispatching, routing, traffic control at junctions, blockage
resolving, etc.).

3.2 The overall HPPCS-FMS architecture

The proposed generic hierarchical approach basically addresses a multi-period production
planning problem in an FMS. Finding satisfactory answers to multi-period production
planning problems in FMSs requires explicit handling of FMS set-up problems. Although
it is possible to construct the hierarchy in various ways, the generic framework depicted in
3326 E. Albey and Ü. Bilge

Figure 2 is proposed in this study. The proposed HPPCS-FMS hierarchy is composed of

three levels: (i) the Aggregate Planning Level (APL); (ii) the Short Term Planning Level
(STPL); and (iii) the Shop Floor Execution Level (SFEL).
The Aggregate Planning Module (APM) in the Aggregate Planning Level (APL) works
over a medium-length rolling planning horizon to produce a production plan that optimises
certain company goals, while satisfying future demands dictated by some outer agent. APL
also models the FMS product flexibility. The Loading Module (LM), employed in the Short
Term Planning Level, works for the upcoming planning period to determine tool and
operation allocation to machines in order to accomplish the production plan given by the
APM in the most efficient way. The Detailed Planning Module (DPM) in the STPL
determines the part release plan for the period and arranges the feasible alternative process
plans of parts according to the LM result. The FMS Simulator in the Anticipation Module
executes the proposed plan. If released parts cannot be completed within the allowed time
period, some feedback information is provided in order to learn the capacity-related
parameters in the upper levels and a new iteration is started. The inner loop within the
HPPCS-FMS in Figure 2 is iterated in this manner until an executable plan for the upcoming
period is finally obtained. The finalised plan is executed in a Shop Floor Control and
Execution System environment (i.e. by a real-time FMS controller). In this study, the shop
floor control system is also modelled by simulation. FMS.NET (Gonen 2005) is used as the

Figure 2. A conceptual framework for hierarchical production planning and control for FMS.
 International Journal of Production Research 3327

simulation tool for both simulation models. The models are constructed at each iteration
according to the input generated by DPM, including part release plans, arranged process
plans and selected set of real-time decision rules. After the execution of the plan for one
week, the end-of-period status of the shop floor (i.e. WIP levels, etc.) is fed as the initial
condition to the APM and the horizon is rolled on. The execution of the finalised plan and
feedback to APM can be traced in the outer loop in Figure 2. When using simulation as an
anticipation tool, the model is based on the expected behaviour of the real-world shop floor
system, i.e. expected processing times, breakdown occurrences and repair times for
resources are used. These should be obtained from up-to-date statistics collected from
SFEL. On the other hand, the SFEL and the upper levels should have certain protocols to
deal with situations calling for plan revisions due to events causing substantial variation
within a period. The development of such protocols is outside the scope of this study. Thus,
the outer feedback loop in the current implementation covers only the end-of-period state of
the SF.
In the rest of this section, particular mathematical models for APM and LM and a
heuristic for DPM are presented. At the end of the section, the effective capacity
anticipation approach is described in detail.

3.3 A mathematical model for APL

The main features of the aggregate planning mathematical model (APMM) proposed for
APM are as follows.
(1) Aggregation: The aggregation is carried out over machines and tools. All machines
of the same type are regarded as a single machine resource and the tools are
aggregated similarly. Without loss of generality, the presented model represents a
case with a single machine group. Time is aggregated into time buckets such that
the period length is long compared with lead times. Alternative process plans are
modelled explicitly, since an aggregation over routes is not applicable in an FMS
context. At this level, a route is defined as an operation sequence without any
reference to machines since tool allocations of the machines are not determined.
(2) Critical resources: Tools and machines are the critical resources. Tool wear is not
modelled. Material handling times are assumed to be negligible. Pallets and fixtures
are not considered critical.
(3) Deterministic demand: The effect of demand uncertainty is not investigated in this
study.
Both mathematical models presented in this study use all or some of the basic sets
defined below.
Part types i ¼ 1, 2, . . . , I
Routes j ¼ 1, 2, . . . , J
Time periods t ¼ 1, 2, . . . , T
Machines w ¼ 1, 2, . . . , W
Tool types k ¼ 1, 2, . . . , K
Operations n ¼ 1, 2, . . . , N
Alias of operations n0 ¼ 1, 2, . . . , N
3328 E. Albey and Ü. Bilge

The notation and the LP model for APMM are given below.
Variables
Qijt Amount of part type i produced following route j in period t
TQit Amount of part type i produced in period t (aggregated over routes)
Akt Number of tool type k used in period t
Bit Unsatisfied demand of part type i in period t
Hit Ending inventory of part type i in period t
Sets
PR( i ) Set of process route of part type i
RO( j ) Set of operations for each route j
RT( j, k) Set of operations of route j requiring tool k
Parameters
bit Backorder cost for part type i in period t
hit Holding cost for part type i for period t
cit Cost of producing part i in period t
dit Demand for part type i in period t
takt Available time of tool k in period t
atk Available number of copies of tool type k
wipi Amount of WIP of part type i at the beginning of period t ¼ 1
lk Tool slot requirement of tool type k
l Total tool magazine capacity of all machines
ptn Unit processing time of operation n
rt Available time of aggregate machine resource in period t
W Number of machines
ek Capacity correction factor for tool type k, to account for finished
operations of the WIP at the beginning of period t ¼ 1
o Capacity correction factor for aggregate machine resource, to
account for finished operations of the WIP at the beginning of
period t ¼ 1
kt Capacity coefficient for tools
t Capacity coefficient for aggregate machine resource
APMM:
T X
X I
min ðcit TQit þ hit Hit þ bit Bit Þ, ð1Þ
t¼1 i¼1

s.t.
X
K
lk Akt  l, 8t, ð2Þ
k¼1

Akt  atk , 8k, t, ð3Þ

X
Qijt ¼ TQit , 8i, t, ð4Þ
j2PRð i Þ
 International Journal of Production Research 3329

TQi1  wipi , 8i, ð5Þ

TQi1 þ Hi0  Hi1 þ Bi1  Bi0 ¼ di1 þ wipi , 8i, ð6Þ

TQit þ Hit1  Hit þ Bit  Bit1 ¼ dit , 8i, t 4 1, ð7Þ

X
I X X
Qij1 ptn  ðAk1 tak1 þ ek Þk1 , 8k, ð8Þ
i¼1 j2PRðiÞ n2RTð j,kÞ

X
I X X
Qijt ptn  ðAkt takt Þkt , 8k, t 4 1, ð9Þ
i¼1 j2PRðiÞ n2RTð j,kÞ

X
I X X
Qij1 ptn  ðWðr1 Þ þ oÞ1 , ð10Þ
i¼1 j2PRðiÞ n2ROð j Þ

X
I X X
Qijt ptn  ðW Þðrt Þðt Þ, t 4 1, ð11Þ
i¼1 j2PRðiÞ n2ROð j Þ

Akt , Qijt , TQit , Hit , Bit  0, 8i, j, k, t:

The objective function minimises the production, holding and backorder costs. Constraint
set (2) ensures that the tool slot requirement of mounted tools does not exceed the total
magazine slot capacity of the aggregate machine resource for each planning period.
Constraint set (3) ensures that the number of tool copies used is less than the available tool
copies at hand, for each tool type and planning period. Constraint set (4) defines the total
amount produced for each part, aggregated over all of its possible process routes, in each
planning period. Constraint set (5) is an important constraint, reflecting the current shop
floor (SF) condition in the model as initial condition. This constraint forces the model to
give first priority, in allocating resources, to the production of WIP already waiting in the
SF. Constraint set (7) is the classical demand and balance constraint. It ensures that the
backorder and inventory of two successive periods are netted with the demand and
production amount of a later period. Constraint (6) is for the first period (t ¼ 1) and is
exactly the same as (7), except for the parameter wipi, which is used to account for the WIP
in SF. Constraint set (9) ensures that the processing time for all operations requiring a
specific tool type should be less than the total available time (the period length times the
number of copies of the tool type that is used) of that tool type, corrected by the feedback
parameter, kt, obtained from SF. Constraint set (8) is for t ¼ 1 and contains an additional
parameter, ek, to reflect the initial status of the SF. Since in (5) the WIP amounts are
forced to be released to SF like new orders, the available capacities should be corrected by
restoring the amount of time consumed by the already finished operations of the WIP. The
actual time used by each tool for finished operations of the WIP is recorded during SF
execution and used as input, ek, at the beginning of the next period. Constraint sets (10)
and (11) are very similar to the availability constraints of tool resources explained above.
They ensure that the cumulative processing time requirement of all operations cannot
exceed the capacity limit of the aggregate machine resource, again corrected by a factor t,
3330 E. Albey and Ü. Bilge

obtained from SF. The additional parameter o in constraint set (10) is used for the same
purpose with ek.

3.4 A mathematical model for the loading problem in STPL

Grieco et al. (2001) provide an elaborate analysis of the characteristics of loading problems
along with a categorisation and review of mathematical modelling approaches considering
several types of objective functions and constraints. The most common objectives involve:
costs (manufacturing, inventory, profit), flow times, makespan, number of alternative
routes, changes in tool magazine configurations and workload balancing among machines.
Grieco et al. (2001) report that few articles deal with how the loading model is integrated
with the upper and lower levels in the planning hierarchy. Furthermore, they report that
the objective function should be selected based on how the loading model fits into the
whole hierarchy. Since the higher level considers the achievement of the goals of the firm
by minimising the cost for the overall horizon, it is clear that the medium level should aim
at a good fit to upper level plans along with good tool allocation, allowing a smooth flow
of parts in the SF. In that respect, workload balancing can be seen as an appropriate
choice at this level. A nice study on comparing workload balancing objectives is presented
by Kumar and Shanker (2001). The review of Grieco et al. (2001) also reveals that most
studies acknowledge tool magazine capacity and processing time availability on machines
as the most critical resources. Another important remark by Grieco et al. (2001) is that
alternative process plans are totally ignored in available loading models despite the fact
that they could allow better exploitation of the resources in the system. In this respect, the
formulation presented by Guerrero (1999) is a rare example.
In this study a new MIP model for the loading problem that focuses on flexible process
plans and communicates with the upper and lower levels is proposed. The primary
objective of our Loading Mathematical Model (LMM) is achieving a setup that provides
the maximum possible consistency with the production amounts announced by APL. The
APMM determines production quantities for each alternative process plan for each part.
However, rather than forcing the LMM to follow quantities produced for each alternative
process plan, the APMM passes on the aggregated quantity for the part type to LMM.
The idea behind this aggregation is to allow LMM to utilise routing flexibility in such a
way as to optimise its own objective function. As a secondary objective, LMM tries to
balance the workload among machines in order to minimise the possibility of creating
bottlenecks on the SF and benefits from routing flexibility. Since the loading model is a
part of a rolling horizon scheme, the WIP remaining from the previous period in
intermediary stages should be explicitly considered to guarantee that the new shop
configuration allows their production. The proposed LMM is provided below.
Parameters
bi Backorder cost for part type i
qi Amount of part type i required by APMM
qmax Maximum of qi
tw Available time of machine w
tmax Maximum of tw
atk Available number of tool copies of type k
ptn Unit processing time of operation n
 International Journal of Production Research 3331

lk Tool slot requirement of tool type k

scw Slot capacity of machine w
wipn Amount of WIP having operation n as the last completed operation
in the previous planning period
w Smoothed capacity coefficient of machine w
Variables
Xnw Quantity of operation n processed on machine w
Fnn0 Quantity of flow from operation node n to n0 on the process plan
network
Dkw 1 if tool k is placed on machine w, 0 otherwise
Si Unsatisfied requirement of part type i
Umax Workload of machine with maximum utilisation
Sets
O Set of all operations
LO(i) Last operations of part type i on its process plan network
FO(i) First operations of part type i on its process plan network
S(n) Set of immediate successors of operation n on the network
P(n) Set of immediate predecessors of operation n on the network
T(n) Set of tools required for processing operation n
LMM:
( !  )
X
I
bi Si Umax
min qmax þ , ð12Þ
i¼1
qi tmax

s.t.
X
K
lk Dkw  scw , 8w, ð13Þ
k¼1

X
W
Dkw  atk , 8k, ð14Þ
w¼1

Xnw  qmaxDkw , 8n, w, k 2 TðnÞ, ð15Þ

X
N
ptn Xnw  tw w , 8w, ð16Þ
n¼1

X
W X
Xnw þ wipn ¼ Fnn0 , 8n 2 OnLOðiÞ, ð17Þ
w¼1 n0 2SðnÞ

X X
W
Fnn0 ¼ Xn0 w , 8n0 2 OnFOðiÞ, ð18Þ
n2Pðn0 Þ w¼1
3332 E. Albey and Ü. Bilge

X
N
Umax  ptn Xnw , 8w, ð19Þ
n¼1

X X
W
Xnw þ Si ¼ qi , 8i, ð20Þ
n2LOðiÞ w¼1

Dkw 2 Binary, 8k, w,

Xnw , Fnn0 2 Integer, 8n, n0 , w,
Si  0, 8i,
Umax  0:
The multi-criteria objective function (12) handles two criteria, the unsatisfied demand and
the workload balance, through a weighted sum. In the first component, the unsatisfied
ratio for the requirement of each part type is weighted by the part’s backlogging cost and
the sum is weighted by the maximum required amount over part types. In order to balance
the workloads assigned to each machine, the second component of the objective function
minimises the scaled workload of the machine with maximum usage. In this manner, the
relative weights of the two criteria are based on problem-specific data and the first
criterion is assigned more weight than the second. This choice of relative weights is verified
through preliminary experimentation. Constraint set (13) ensures that the total slot
requirements of the tools allocated to each machine cannot exceed its total slot capacity.
Constraint set (14) asserts that the number of tool copies that can be mounted on machines
is limited by tool availability. Constraint set (15) ensures that an operation is not allocated
to a machine if the required tools are not mounted. Constraint set (16) ensures that the
total processing time of the operations loaded on a machine cannot exceed the total
available processing time of that machine. The alternative routes for each part type
are represented as a network defined by the sets S(n), P(n), LO(i) and FO(i) as shown in
Figure 3, where nodes are operations and arcs are Fnn0 . Constraint sets (17) and (18) are
network flow constraints that ensure the parts flow through the routes defined in the
network. Constraint (17) asserts that the total production amount of an operation and
the WIP already available at that stage is equal to the outflow of that operation to the
succeeding operations, while in constraint (18) the total inflow to an operation is equated
to the production amount at that stage, ensuring the completion of each part in the system.
Constraint set (19) determines the maximum used machine resource time, which is used in
the objective function in order to balance the workload among machines. Constraint set
(20) is the definition representing the quantity of unsatisfied demand.

Figure 3. An alternative process plan network for a sample part type and constraints (17) and (18)
for n ¼ 3 in LMM.
 International Journal of Production Research 3333

3.5 Detailed Planning Module (DPM)

The DPM issues instructions to prepare the SFEL for a given production plan. Similar
instructions are also issued to the FMS Simulator when experimenting for anticipation
purposes. These instructions include the new tool allocation, the part process plans revised
according to the results of LM by eliminating the routes whose set of tools are not selected,
and the part release schedule. The part release schedule is particularly important
because, as Sabuncuoglu and Karapinar (1999) point out, order review and release (ORR)
may be regarded as a capacity management tool which performs finer capacity
adjustments prior to a dispatching function by controlling the WIP level over time.
Several studies in the literature, including the recent work of Ebadian et al. (2009),
study the order release problem within a HPP framework. The ORR strategy that is used
in the current implementation of DPM determines a periodic pattern of order release
sequences to smooth production using the heuristic procedure described by Askin
and Goldberg (2002).

3.6 Shop Floor Anticipation Module

The Anticipation Module tries to anticipate the behaviour of the shop floor for a given
production plan and FMS configuration by means of simulation and accordingly updates
the capacity coefficients that are used in the upper-level modules. Capacity anticipation in
the FMS context can be more complicated than for conventional production because the
manufacturing environment is dynamic. Each time LMM is solved with a new set of
resource capacity coefficients, a different FMS loading may result. However, the
production amounts determined by APMM are shaped according to the resource capacity
coefficients learned for the previous FMS configuration. This may cause fluctuations and
poor convergence for the anticipation process. The effective capacity in the SF is also
affected by the operational policies (e.g., routing strategy) used during execution. Thus, the
role of the anticipation module is not only to estimate the effective capacity, but also to
select a good operational setup that results in a high-performance shop floor. Several
strategies have been developed to deal with this nervousness and to obtain good
coefficients. These are basically concerned with the selection of the loading solution to be
implemented in SFEL and the synchronisation of the SF execution with the selected
loading solution. The details of these strategies and the procedure used for updating
capacity coefficients are explained below.

3.6.1 The loading strategy

Using a new FMS loading at every iteration may cause some nervousness in the system, as
explained above. An extreme approach would be to solve the loading only once at the first
iteration for the period, then keep this configuration unchanged unless a change in the part
mix requested by APMM renders it incapable of producing the new mix. However,
particularly for drastic changes in the relative quantities of part types, this approach may
perform poorly because the workload balance in the fixed configuration deteriorates. The
loading strategy adopted in the anticipation module compares the loading solution from
the previous iteration and that obtained for the new set of coefficients through two distinct
simulation runs. The one yielding the shortest makespan is retained. If simulating a
scenario is not computationally expensive as in our case (on the order of seconds), this is a
3334 E. Albey and Ü. Bilge

reasonably good strategy to follow. An extensive numerical study reports that this strategy
performs better than fixing the loading or changing it at each anticipation iteration.
The results of these experiments are not reported in this paper due to space limitations,
however they can be found in Bilge and Albey (2008).

3.6.2 Part routing synchronisation

The solution to the LMM actually gives detailed decisions about which routes for
each part type will be used. Whether and how this information will be used during
SF execution, in other words the level of synchronisation among LM and SFEL, would be
an interesting issue to investigate at this point.
During SF execution, whenever an operation is completed the next operation and the
next machine to perform it should be selected for this part among the eligible alternatives
in its alternative routes. This part routing (PR) is made by means of an on-line PR rule.
While the mathematical model arranges routes by considering the whole period, a PR rule
considers only a local snapshot. If the PR rule totally ignores the results of LMM, then the
global information already created will be wasted. It seems, at first glance, that PR should
directly follow the routing decisions of the model. However, this choice results in a loss of
real-time operational flexibility, reducing the ability to respond to unforeseen events. How
much flexibility should be allowed at the SFEL and how much of it should be fixed at
STPL is an issue that needs to be addressed in defining the FMS planning and control
hierarchy. As a means of addressing this issue we propose regulating the set of eligible
route alternatives over which a PR rule works.
In this study, the Smallest Work in the Next Queue (SWINQ) rule is used as the
PR rule. As the name implies, this rule selects the machine having the minimum workload
on its input buffer among those that can perform the next operation. SWINQ is a widely
used, simple and efficient rule that aims to distribute work to machines in a balanced way
(for more sophisticated PR rules working under alternative routes, see Bilge et al. (2008)).
Depending on the way the eligible set for the PR rule is defined, it is possible to have
different synchronisation levels. In the most relaxed case (SWINQ-0) the eligible set of the
next operation and machine pairs are defined as the whole set of feasible candidates under
the given tool allocation, ignoring route selections suggested by LMM. Thus SWINQ-0
does not pursue any synchronisation among the two levels in terms of PR. We define two
more implementations of SWINQ having different degrees of synchronisation. SWINQ-1
is allowed to use only those alternative operation routes which are used by LMM. This
approach restricts the selection of candidate operations. However, machine selection
among feasible candidates is not restricted, allowing us to utilise full operation flexibility.
SWINQ-2 is the most restricted SWINQ version, forcing full synchronisation in route
selection decisions. In addition to the restriction of candidate operations, machine
selection is also restricted according to the results of LMM. In other words, even if a
machine has the proper tooling for an operation, LMM may prefer not to use it for
balancing reasons, and, subsequently, SWINQ-2 is prohibited to consider such a machine
as an eligible candidate for processing the operation under consideration. However, the
amount of parts to be allocated on each allowed route is still left as an on-line decision.
Our experimentation shows that strategies with synchronisation dominate SWINQ-0
significantly, although the synchronisation versions are not statistically different from each
other. These results, which are omitted here, are reported by Bilge and Albey (2008).
In our anticipation module, each SWINQ version is tested in a separate simulation run
 International Journal of Production Research 3335

and the one that gives the smallest makespan is selected as the PR rule to be used during
SF execution of that period.

3.6.3 Effective capacity update procedure

The effective capacity of the manufacturing system is represented through the coefficients
kt, t and wt on the right-hand sides of the resource availability constraints (8)–(11) in
APMM and (16) in LMM. The Effective Capacity Update Procedure (ECUP) offers a
methodology for estimating and dynamically updating these coefficients. The basic aspects
of ECUP are as follows.
(1) The driving resource coefficient is t for the aggregate machine resource. Since it is
assumed that all the machines are identical, wt ¼ t. Furthermore, for the sake of
simplicity, kt is set equal to t.
(2) The key parameter in computing  is the final completion times, RTw, of individual
machines in the simulation. Individual machine coefficients are given by the ratio
of the period length to RTw.  is then computed as their average.
(3) The main stopping condition is having all RTw values within a specified allowable
range of the planning period length.
(4) ECUP is designed in two phases to deal with the fluctuation behaviour
described at the beginning of this section and ensures convergence to a feasible
solution.
The main aim of Phase 1 is to find a  value such that all machines complete their
workload before the end of the period. If such a  results in completion times within the
allowable range of the period length, then ECUP stops. Otherwise, to make sure that the
procedure will not stop with a low capacity estimate, Phase 2 starts. Phase 2 basically
makes a half-interval search to find the highest possible  value that gives a feasible plan,
thus eliminating the shortcoming in the work of Byrne and Bakir (1999). The basic
difference between the two phases is in the way  is updated.
At every iteration when ECUP returns with a new for trial, a new main execution loop
for the HPPCS-FMS is started. ECUP stops with the final effective capacity coefficient
final, thus the order release plan with the machine loading solution is finalised for the
planning period. The notation and pseudo-code of ECUP are as follows.
NMC Nominal machine capacity (the planning period length)
AB Allowable bandwidth of makespan. When all RTw fall inside AB,
then the current plan is accepted as satisfactory
RTw Final completion times of machines in the execution
ULAB Upper limit of allowable bandwidth. Assumed to be equal to NMC
LLAB Lower limit of allowable bandwidth. Computed by NMC  p. Based
on preliminary experiments, p is set to 0.98
RCw Individual machine coefficients. Initially set to one. If RTw 4 NMC,
then RCw is set to NMC/RTw, otherwise
P it is left as one
ARC Average of resource coefficients, RCw/W
new Updated cumulative resource coefficients
prev Previous cumulative resource coefficient. Initially set to one
ULC, LLC Upper/lower end points for the current search interval for , in
Phase 2
3336 E. Albey and Ü. Bilge

SS A real number between zero and one indicating step size, used to
speed up Phase 1 when two consecutive ARC values do not differ by
more than SS. Based on preliminary experiments, SS is set to 0.05
CIL The number of allowable Phase 2 iterations. Based on preliminary
experiments, CIL is set to five
ECUP
– Take RTw as input from simulation results.
– If iteration count ¼ 1 and all RTw 5 LLAB, then
Set final ¼ 1 and STOP.
– If all RTw 2 AB (main stopping condition), then
Set final ¼ prev and STOP.
– Otherwise:
If Phase 2 flag is true, then execute Phase 2.
Else execute Phase 1.
Phase 1:
– If RTw 5 ULAB for all w, then
Set Phase 2 flag to true and GO TO Phase 2.
– Compute RCw for each machine.
– Calculate ARC.
– Update :
If two consecutive iterations result in ARC values which differ by less than SS,
then (speed up)
new ¼ prev  (1  SS).
Else
new ¼ prev  ARC.
– Record prev as ULC and return new for new iteration.
Phase 2:
– If RTw 5 ULAB for all w, then
Set LLC as prev.
Else
Set ULC as prev.
– If iteration count 4 CIL, then
Set final ¼ LLC and STOP.
Else (half-interval search)
Set new ¼ (LLC þ ULC )/2.
Increment iteration count and return new for new iteration.

4. Numerical study
A set of numerical experiments is conducted to examine the relative performance
improvement that can be gained via the iterative capacity anticipation described in
Section 3.6. For this purpose, the closed-loop hierarchical production planning system
(A1) shown in Figure 2 is compared with a single-pass hierarchical production planning
system (A0). The basic difference in A0 is that it does not have the capacity anticipation
 International Journal of Production Research 3337

module and the inner feedback loop in Figure 2, hence it does not employ any of the
strategies discussed in Section 3.6.
In the test scenarios generated, the properties of the material handling system and
number of machines is held fixed (at six), whereas all the remaining data (i.e. number and
definition of part types, process plans, demand over a five-period planning horizon,
number of tool types, copies and slot requirements) are varied. The data for the test
problems are summarised in Table 1. In each scenario, the total magazine capacity
is determined by controlling the ratio ofPtotal slot requirements of all tool copies to total
magazine capacity using the formula ( k lk atk)/(W)(scw) (Denizel and Erenguc 1997,
Kumar and Shanker 2001). This ratio is referred as the Magazine capacity tightness in
Table 1. While generating the demand data, emphasis is given to creating a varying
workload over the planning periods. In addition, the mix of products and their alternative
process plans are generated in such a way so as to allow manufacturing flexibility. The
complete data set for these problems can be found at www.bufaim.boun.edu.tr/
HPPCSFMSDataSets.aspx.
Models are solved using CPLEX 7.0. APMM requires negligible CPU time. However,
the integer variables in LMM make it intractable, so LMM is allowed a 10% optimality
gap. With this policy, the longest LMM run in our experiments takes less than 120
seconds. A full experiment to solve a five-period problem with anticipation takes between
10 and 50 minutes on a 2 GHz Pentium processor with 1 GB RAM.
In the experiments, solution quality is assessed using two performance
measures. The first is the Satisfied Demand, denoting the total demand satisfied within
the planning horizon. The second performance measure is the Final Total Cost after rolling
the horizon for a fixed number of periods (i.e. until the end of the execution of the whole

Table 1. Summary of experimental data (see www.bufaim.boun.edu.tr/HPPCSFMSDataSets.aspx).

Number of Total tool Average Number Magazine Workload

Problem Problem product magazine number of of tool capacity range/period
group name types capacity process plans types tightness (%)

1 P1 3 54 2.3 10 0.72 (106–123)

P2 3 54 2.3 10 0.72 (128–150)
P3 3 54 2.3 10 0.72 (108–126)
2 P4 3 48 2.3 8 1.20 (119–145)
P5 3 48 2.3 8 1.20 (124–183)
P6 3 48 2.3 8 1.20 (89–187)
3 P7 4 30 1.3 9 0.59 (100–100)
P8 4 30 1.3 9 0.59 (120–120)
P9 4 30 1.3 9 0.59 (100–100)
4 P10 4 36 2.5 7 0.86 (107–127)
P11 4 36 2.5 7 0.86 (99–158)
P12 4 36 2.5 7 0.86 (69–143)
5 P13 5 24 1.8 7 0.48 (101–129)
P14 5 24 1.8 7 0.48 (128–161)
6 P15 5 54 1.6 8 0.75 (91–136)
P16 5 54 1.6 8 0.75 (99–106)
P17 5 54 1.6 8 0.75 (123–154)
7 P18 7 54 1.3 11 0.58 (90–122)
8 P19 8 54 2.4 10 0.64 (122–182)
9 P20 12 90 1.8 18 0.66 (80–165)
3338 E. Albey and Ü. Bilge

planning horizon). Figures 4 and 5 present the results related to these measures.
The reference line in Figure 4 shows the normalised total demand and the two vertical
bars for each problem show the Satisfied Demand as a percentage of total demand for A1
and A0. On the other hand, the reference line in Figure 5 indicates the Initial Total Cost
obtained by solving APMM for the first time at the beginning with nominal capacity. The
bars indicate the Final Total Cost (incurred upon the complete execution of the rolling
horizon) as a percentage of the Initial Total Cost with each strategy. As revealed by the
figures, A1 outperforms A0 with respect to both performance measures for all problem
instances. The dominance of strategy A1 over A0 is also verified in Table 2, which provides
detailed information including the cost components for these experiments. Column 4 shows
the Satisfied Demand as a percentage of the total demand. Column 10 (Cost Increase)
demonstrates the increase in the Final Total Cost relative to the Initial Total Cost. The last
column presents the ratio of the difference in the Final Total Cost of A1 and A0 to that of A0.

Figure 4. Satisfied demand percentages for strategies A0 and A1.

Figure 5. Final total cost percentages for strategies A0 and A1 compared with the initial total cost.
 International Journal of Production Research 3339

5. Conclusion
In contrast to the extensive hierarchical planning literature for conventional manufactur-
ing environments, the literature dealing with this paradigm in the FMS domain is still very
sparse. In spite of the vast amount of research that address FMSs, most of these studies
focus only on short-term planning problems, ignoring their integration with the upper or
lower levels, and do not offer upper and lower-level models that incorporate specific FMS
characteristics, thus failing to provide full hierarchical structures. Moreover, the recent

Table 2. Comparison of cases with anticipation (A1) and no anticipation (A0).

Satisfied Back Final Initial Cost Relative

Total Total ratio Prod. Hold. order total total increase gain
Problems demand satisfied (%) cost cost cost cost cost (%) (%)

P1–A1 9351 9351 100 19,711 0 258 19,969 19,711 1 42

P1–A0 9351 8823 94 18,463 0 15,922 34,384 19,711 74
P2–A1 11,626 10,430 90 22,054 0 26,713 48,767 43,474 12 31
P2–A0 11,626 7819 67 18,675 0 52,120 70,795 43,474 63
P3–A1 9753 9753 100 29,183 0 0 29,183 29,183 0 49
P3–A0 9753 8248 85 24,728 0 32,017 56,744 29,183 94
P4–A1 10,459 9232 88 18,346 0 23,916 42,262 38,409 10 37
P4–A0 10,459 8203 78 15,826 0 51,448 67,274 38,409 75
P5–A1 12,765 9704 76 20,061 369 53,168 73,599 67,755 9 17
P5–A0 12,765 8710 68 18,041 76 70,201 88,318 67,755 30
P6–A1 8832 7894 89 21,690 677 12,789 35,156 31,251 12 35
P6–A0 8832 6942 79 18,997 318 34,536 53,851 31,251 72
P7–A1 9448 9261 98 10,762 441 451 11,654 11,586 1 8
P7–A0 9448 9027 96 10,452 406 1776 12,634 11,586 9
P8–A1 11,340 10,215 90 11,101 296 6023 17,420 15,375 13 4
P8–A0 11,340 10,081 89 10,962 250 6844 18,055 15,375 17
P9–A1 8899 8899 100 8222 0 237 8458 8422 0 42
P9–A0 8899 7596 85 6974 0 7717 14,691 8422 74
P10–A1 9706 9359 96 10,281 3 2331 12,615 10,787 17 43
P10–A0 9706 7718 80 8284 0 13,687 21,970 10,787 104
P11–A1 11,932 9932 83 10,032 43 16,662 26,737 19,767 35 24
P11–A0 11,932 8189 69 8129 0 27,153 35,282 19,767 78
P12–A1 8586 8586 100 9513 167 274 9955 9633 3 37
P12–A0 8586 7301 85 7889 81 7717 15,686 9633 63
P13–A1 11,480 10,168 89 10,024 200 9713 19,937 15,369 30 17
P13–A0 11,480 9501 83 9387 121 14,566 24,074 15,369 57
P14–A1 15,081 12,262 81 13,229 0 17,499 30,728 28,733 7 17
P14–A0 15,081 11,079 73 11,978 0 25,023 37,001 28,733 29
P15–A1 8900 8634 97 9549 71 2307 11,926 10,319 16 13
P15–A0 8900 8359 94 9200 62 4377 13,638 10,319 32
P16–A1 9720 9570 98 6378 3 1019 7400 6514 14 23
P16–A0 9720 9307 96 6204 1 3421 9626 6514 48
P17–A1 11,729 9973 85 10,507 0 11,310 21,817 20,890 4 13
P17–A0 11,729 9453 81 10,018 0 15,034 25,052 20,890 20
P18–A1 9959 9077 91 9432 240 7770 17,442 15,716 11 12
P18–A0 9959 8920 90 9293 318 10,137 19,748 15,716 26
P19–A1 15,135 13,294 88 15,695 202 15,969 31,867 29,909 7 20
P19–A0 15,135 12,216 81 14,273 161 25,641 40,075 29,909 34
P20–A1 10,645 10,645 100 10,732 135 388 11,256 10,960 3 6
P20–A0 10,645 10,645 100 10,732 135 1117 11,982 10,960 9
3340 E. Albey and Ü. Bilge

vigorous stream of research dealing with circularity in planning and capacity anticipation
has not yet turned to FMSs.
In this paper, a complete and closed-loop hierarchical production planning frame-
work designed specifically for flexible manufacturing environments is presented. There are
two main issues emphasised. First, the models proposed at each level represent several
FMS flexibilities explicitly and exploit them as a means of using capacity effectively. It is
also recognised that the integration among levels requires an explicit definition of the level
and the way each type of flexibility will be fixed during a planning/control epoch.
Secondly, a simulation-based effective capacity anticipation approach is developed. The
experimental evaluation of the proposed framework demonstrates that anticipation
considerably improves the quality of planning and reveals the importance of research
in this direction.
The proposed HPPCS-FMS is generic because of its modular and multi-level structure
that allows plugging in several different models and solution methodologies dealing with
one or more problems embedded in the levels. Obviously, different models and solution
techniques can be suggested depending on the nature of the FMS environment, the level
of modelling detail, the relevant objectives and the time available for solution. Those
included in this study form a complementary set that constitute a complete production
planning system for a reasonable problem context.
The following issues are interesting future research directions to investigate different
facets of the overall FMS production planning problem.
(1) The current study uses simulation as a tool for anticipation. The implementation
of other approaches such as clearing functions in the FMS domain is an open
question.
(2) The current implementation of STPL assumes an FMS working under batching
mode for set-up. STPL can be redesigned for incremental set-up mode, where the
tool configurations are updated upon completion of the production requirements
for each part type. This brings forth a new and interesting hierarchical framework
for the FMS, which requires some changes in LMM as well as the redesign of the
inner and outer loops in Figure 2.
(3) Different and additional types of feedback from SFEL to the planning levels can be
developed. Specifically, in the case of substantial variation from the plan during
execution, re-planning calls might be issued. Similarly, substantial variation in the
demand within a period may require re-planning. The timing or the triggering
events for re-planning calls, the extent of re-planning and the protocols to manage
re-planning should be investigated as a means of coping with uncertainties within
the facility.
(4) DPM in STPL is specifically designed to manage the part release to SF
during a given period. Investigation and integration of more sophisticated
part release policies can further enhance the performance of the hierarchical
system.

Acknowledgement
The work reported in this paper is supported by Bogazici University Research Fund under Grant
No. 05A301.
 International Journal of Production Research 3341

References

Anthony, R.N., 1965. Planning and control systems: a framework for analysis. Cambridge, MA:
Harvard University Press.
Askin, R.G. and Goldberg, J.B., 2002. Design and analysis of lean production systems. New York:
Wiley.
Asmundsson, J., Rardin, R., and Uzsoy, R., 2006. Tractable nonlinear production planning models
for semiconductor wafer fabrication facilities. IEEE Transactions on Semiconductor
Manufacturing, 19 (1), 95–111.
Atlihan, M.K., Kayaligil, S., and Erkip, N., 1999. A generic model to solve tactical planning
problems in flexible manufacturing systems. International Journal of Flexible Manufacturing
Systems, 11 (3), 215–243.
Bastos, J.M., 1988. Batching and routing: Two functions in the operational planning of flexible
manufacturing systems. European Journal of Operational Research, 33 (3), 230–244.
Benjaafar, S. and Ramakrishnan, R., 1996. Modelling, measurement and evaluation of sequencing
flexibility in manufacturing systems. International Journal of Production Research, 34 (5),
1195–1220.
Bertrand, J.W.M., Wortmann, J.C., and Wijngaard, J., 1990. Production control: a structural and
design oriented approach. New York: Elsevier.
Bilge, U. and Albey, E., 2008. A hierarchical approach to FMS planning and control with capacity
anticipation. Bogazici University, FBE-IE-12/2008-13.
Bilge, U., Firat, M., and Albey, E., 2008. A parametric fuzzy logic approach to dynamic part routing
under full routing flexibility. Computers & Industrial Engineering, 55 (1), 15–33.
Byrne, M.D. and Bakir, M.A., 1999. Production planning using a hybrid simulation-analytical
approach. International Journal of Production Economics, 59 (1–3), 305–311.
Byrne, M. and Hossain, M., 2005. Production planning: an improved hybrid approach. International
Journal of Production Economics, 93/94, 225–229.
Chandra, P., 1995. Production planning model for a flexible manufacturing system. In: A. Raouf
and M. Ben-Daya, eds. Flexible manufacturing systems: recent developments. Netherlands:
Elsevier, 157–171.
Chung, S.H. and Chien, W.L., 1993. Building a short term production planning system for FMS: an
integration viewpoint. Production Planning and Control, 4 (2), 112–127.
Co, H.C., Biermann, J.S., and Chen, S.K., 1990. A methodical approach to the flexible-
manufacturing-system batching, loading and tool configuration problems. International
Journal of Production Research, 28 (12), 2171–2186.
Denizel, M. and Erenguc, S.S., 1997. Exact solution procedures for certain planning problems
in flexible manufacturing systems. Computers and Operations Research, 24 (11),
1043–1055.
Ebadian, M., et al., 2009. Hierarchical production planning and scheduling in make to order
environments: reaching short and reliable delivery dates. International Journal of Production
Research, 47 (20), 5761–5789.
Elmaghraby, S., 1991. Manufacturing capacity and its measurement: a critical evaluation. Computers
and Operations Research, 18 (7), 615–627.
Gonen, M., 2005. BUILD.NET: a graphical application generator for object-oriented software and
sample applications. Thesis (Master’s), Bogazici University.
Graves, S.C., 1986. A tactical planning model for a job shop. Operations Research, 34 (4),
522–533.
Grieco, A., Semeraro, Q., and Tolio, T., 2001. A review of different approaches to the FMS loading
problem. International Journal of Flexible Manufacturing Systems, 13 (4), 361–384.
Guerrero, F., 1999. Machine loading and part type selection in flexible manufacturing systems.
International Journal of Production Research, 37 (6), 1303–1317.
3342 E. Albey and Ü. Bilge

Hax, A.C. and Meal, H.C., 1975. Hierarchical integration of production planning and scheduling;
studies in management sciences. In: M.A. Geisler, ed. Logistics. New York: North Holland-
American Elsevier.
Hung, Y.F. and Leachman, R.C., 1996. A production planning methodology for semiconductor
manufacturing based on iterative simulation and linear programming calculations. IEEE
Transactions on Semiconductor Manufacturing, 9 (2), 257–269.
Hwang, S. and Uzsoy, R., 2005. A single stage multi product dynamic lot sizing model with WIP and
congestion. West Lafayette: Laboratory for Extended Enterprises at Purdue, School of
Industrial Engineering, Purdue University, Technical report.
Irdem, D.F., Kacar, N.B., and Uzsoy, R., 2008. An experimental study of an interative simulation-
optimization algorithm for production planning. In: S.J. Mason, R.R. Hill, L. Mönch,
O. Rose, T. Jefferson and J.W. Fowler, eds. Proceedings of the 2008 winter simulation
conference, 2176–2184.
Karmarkar, U.S., 1989. Capacity loading and release planning with work-in-progress (WIP) and
leadtimes. Journal of Manufacturing and Operations Management, 2, 105–123.
Kim, B. and Kim, S., 2001. Extended model for a hybrid production planning approach.
International Journal of Production Economics, 73 (2), 165–173.
Kumar, N. and Shanker, K., 2001. Comparing the effectiveness of workload balancing objectives in
FMS loading. International Journal of Production Research, 39 (5), 843–871.
Lee, D.H., et al., 1997. Multi-period part selection and loading problems in flexible manufacturing
systems. Computers & Industrial Engineering, 33 (3/4), 541–544.
Nayak, G.K. and Acharya, D., 1998. Part type selection, machine loading and part type volume
determination problems in FMS planning. International Journal of Production Research,
36 (7), 1801–1824.
Nof, S.Y., Barash, M.M., and Solberg, J.J., 1979. Operational control of item flow in versatile
manufacturing systems. International Journal of Production Research, 17 (5), 479–489.
Pahl, J., Voss, S., and Woodruff, D.L., 2007. Production planning with load dependent lead times:
an update of research. Annals of Operations Research, 153 (1), 297–345.
Rohde, J., 2004. Hierarchical supply chain planning using artificial neural networks to anticipate
base-level outcomes. OR Spectrum, 26 (4), 471–492.
Sabuncuoglu, I. and Karapinar, H.Y., 1999. Analysis of order review/release problems in production
systems. International Journal of Production Economics, 62 (3), 259–279.
Sawik, T., 1998. Simultaneous loading, routing, and assembly plan selection in a flexible assembly
system. International Journal of Production Research, 28 (9), 19–29.
Sawik, T., 2004. Loading and Scheduling of a flexible assembly system by mixed integer
programming. European Journal of Operational Research, 154 (1), 1–19.
Sawik, T., 2006. Hierarchical approach to production scheduling in make-to-order assembly.
International Journal of Production Research, 44 (4), 801–830.
Schneeweiss, C., 1995. Hierarchical structures in organisations: a conceptual framework. European
Journal of Operational Research, 86 (1), 4–31.
Selcuk, B., Fransoo, J.C., and Kok, A.G.D., 2006. The effect of updating lead times on the
performance of hierarchical planning systems. International Journal of Production Economics,
104 (2), 427–440.
Sodhi, M., Askin, R.G., and Sen, S., 1994. A hierarchical model for control of flexible
manufacturing systems. Journal of the Operational Research Society, 45 (10), 1185–1196.
Stecke, K.E., 1983. Formulation and solution of nonlinear integer production planning problems for
flexible manufacturing systems. Management Science, 29 (3), 273–288.
Venkateswaran, J. and Son, Y.J., 2005. Hybrid system dynamic discrete event simulation-based
architecture for hierarchical production planning. International Journal of Production
Research, 43 (20), 4397–4429.
Copyright of International Journal of Production Research is the property of Taylor & Francis Ltd and its
content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's
express written permission. However, users may print, download, or email articles for individual use.