GFGHDGH KK

OPTIMIZATION METHODS FOR
BATCH SCHEDULING
Jaime Cerd
Instituto de Desarrollo Tecnolgico para la Industria Qumica
Universidad Nacional de Litoral - CONICET
Gemes 3450 3000 Santa Fe - Argentina
OUTLINE
Problem definition
Types of scheduling problems
Types of scheduling methodologies
Types of scheduling optimization approaches
Overview of network-type discrete and continuous time models
Comparison of network-type discrete and continuous time
formulations (benchmarking examples)
Overview of batch-oriented continuous time formulations
Conclusions
MAIN REFERENCES
Mndez, C.A., Cerd, J., Grossmann, I.E., Harjunkoski, I., Fahl, M. State-of-the-Art Review of Optimization
Methods for Short-Term Scheduling of Batch Processes. Submitted to Computers & Chemical Engineering
(July, 2005).
Castro, P.M.; Barbosa-Pvoa, A.P.; Matos, H.A. & Novais, A.Q. (2004) I&EC Research,
43, 105 118.
Cerd, J.; Henning, G.P. & Grossmann, I.E. (1997) I&EC Research, 36, 1695 1707.
Floudas, C.A.; Lin, X. (2004) Computers and Chemical Engineering, 28, 2109 2129.
Ierapetritou, M.G. & Floudas, C.A. (1998) I&EC Research, 37, 4341 4359.
Janak, S.L.; Lin, X. & Floudas, C.A. (2004) I&EC Research, 43, 2516 2533.
Kondili, E; Pantelides, C.C. & Sargent, W.H. (1993) Computers and Chemical Engineering, 2, 211 227.
Maravelias, C.T. & Grossmann, I.E. (2003) I&EC Research, 42, 3056 3074.
Mndez, C.A.; Henning, G.P. & Cerd, J. (2001) Computers and Chemical Engineering, 25, 701 711.
Mndez, C.A. & Cerd, J. (2003) Comp. & Chem. Eng., 27, 1247 1259.
Pantelides, C.C. (1994) Foundations of Computer-Aided Process Operations, Cache
publications, New York, 253 274.
Pinto, J.M. & Grossmann, I.E. (1995) I&EC Research, 34, 3037 3051.
Pinto, J.M. & Grossmann, I.E. (1997) Computers and Chemical Engineering, 21, 801 818.
Pinto, J.M. & Grossmann, I.E. (1998). Annals of Operations Research, 81, 433 466.
Reklaitis, G.V. (1992). Overview of scheduling and planning of batch process operations.
NATO Advanced Study InstituteBatch process systems engineering. Turkey: Antalya.
INTRODUCTION
PROBLEM DEFINITION
Scheduling is a decision-making process thay plays an important role in
most manufacturing and service industries
The scheduling function aims to optimally allocate resources, available in
limited supplies, to processing tasks over time.
Each task requires certain amounts of specified resources for a specific
time interval called the processing time
Resources may be equipment units in a chemical plant, runways at an
airport or crews at a construction site
Tasks may be operations in a chemical plant, takeoffs and landings at an
airport, activities in a construction project
SCHEDULING DECISIONS AND GOALS

A proper allocation of resources to tasks enables the company to
achieve its objectives
The objectives may take many forms such as:
- minimizing the time required to complete all the tasks (the makespan)
- minimizing the number of orders completed after their committed due dates
- maximizing customer satisfaction by completing orders in a timely fashion
- maximizing plant throughput
- maximizing profit or minimizing production costs
Two eligible tasks cannot generally use the same required resource
simultaneously but one at a time
Scheduling decisions to be made include:
- allocating resources to tasks
- sequencing tasks allocated to the same resource item
- task timing
AN ILLUSTRATIVE EXAMPLE
29 Tasks - 4 Equipment Units - One-month Period Horizon
4.6
10.7
ROLLING TIME HORIZON

The scheduling rolling horizon ranges from 2 to 6 weeks, depending on
whether task processing times are on the order of a day or a week.
Week 1
Week 2
Week 3
Week 4
Week 5
Week 6
Week 2
Week 3
Week 4
Week 5
Week 6
Week 7
6-week rolling horizon
The full schedule for a 6-week horizon might be updated once a week using
updated order input and plant state.
There will be frequent corrections to the schedule in midweek to account
for unit breakdowns or late order arrivals
The scheduling function has to interact with other decision-making systems
used in the plant like the material requirement planning (the MRP system)
The MRP system provides information on the weekly production order
arrivals (product, arrival time, due date and order size), together with the
tasks required to complete each order.
7
INTERACTION WITH THE MRP SYSTEM

Production Planning
Master Scheduling
MRP-II
Material Requirement /
Capacity Planning
Demand forecasts
Final product orders
Material Requirements
Capacity Requirements
Production Orders
Release Dates
Scheduling &
Rescheduling
Schedule
Dispatching
After the schedule has been developed, all raw materials and resources must be
available at the specified times
MRP-II aims to guarantee that the required raw materials and intermediates will
be available in the right amounts at the right times, and the plant capacity is
enough to process all the required productions orders
8
TYPES OF SCHEDULING PROBLEMS

STATIC vs. DYNAMIC PROBLEMS
In static problems, all the production orders and their arrival times are known
beforehand
In dynamic problems, new production orders can arrive at unexpected times while the
schedule is being executed
FLOW SHOP vs. JOB SHOP PROBLEMS

Assume that the jobs require to perform multiple operations on different machines.
- Flow shop: Every job consists of the same set of tasks to be performed in the same
order. The units are accordingly arranged in production lines to minimize the
movement of materials and manpower (multiproduct plant)
- Compound Flow shop: Each unit in the series may be replaced by a set of parallel
equipment items which may be identical or very different. Each job goes to one unit
in the first stage, then it is transferred to one in the second stage and so on.
- Job shop: Production orders have different routes (require different sequences of tasks)
and some orders may even visit a given unit several times (multipurpose plants)
TYPES OF BATCH PRODUCTION FACILITIES
3
Flow-shop facility
1
2
3
4
5
6
7
8
9
10
11
20
18
19
2
2
21
23
24
25
Compound Flow-shop
facility
Job-shop
facility
10
TYPES OF SCHEDULING PROBLEMS -2

MAKE-TO-STOCK vs. MAKE-TO-ORDER PRODUCTION FACILITIES
MAKE-TO-STOCK FACILITIES:
- A make-to-stock manufacturing facility opt to keep in stock items for which there is a
steady demand and no risk of obsolescence.
- Items that are produced for inventory do not have tight due dates
- The lot size is determined by a trade-off between setup costs and inventory holding
costs
- Make-to-stock manufacturing plants are referred to as open shops
MAKE-TO-ORDER FACILITIES:
- Make-to-order jobs have specified due dates and their sizes are determined by
the customer
- Each order is unique and has a unique routing throughout the plant
- Make-to-order manufacturing facilities are referred to as closed shops
- Many manufacturing plants operate partly as a make-to-stock facility processing
warehouse orders and partly as a make-to-order facility processing customer orders
11
TYPES OF SCHEDULING APPROACHES

HEURISTIC METHODS
- Basic Dispatching Rules
- Composite Dispatching Rules
ALGORITHMS OF THE IMPROVEMENT TYPE

- Simulated Annealing
- Genetic Algorithms
- Tabu Search
OPTIMIZATION APPROACHES
- Discrete Time Models
- Continuous Time Models
Network-oriented Formulations
Batch-oriented Formulations
12
HEURISTIC SCHEDULING METHODS

BASIC DISPATCHING RULES
- A basic dispatching rule is a rule that prioritizes all the jobs that are waiting for
processing on a machine
- The prioritization scheme may take into account jobs attributes and machines
attributes as well as the current time
- Whenever a machine has been freed, a dispatching rule inspects the waiting jobs
and selects to process next the job with the highest priority
- Dispatching rules can be classified into STATIC and DYNAMIC RULES.
* A STATIC RULE is not time-dependent but just a function of the job data, the
machine data or both (EDD-earliest due date first, SPT-shortest processing time first)
* DYNAMIC RULES are time-dependent since they also take into account, in addition to
the job and machine data, the current time (Example: MS-minimum slack time-first)
- Dispatching rules can also be categorized into two classes: LOCAL and GLOBAL RULES
* A LOCAL RULE uses only information related to either the queue or the machine /
workcenter to which the rule is applied
* A GLOBAL RULE may use information related to other machines, such as either the
processing times of the jobs or the current queue length on the next machine
13
IMPROVEMENT ALGORITHMS
COMPOSITE DISPATCHING RULES
- Composite dispatching rules combine a number of basic dispatching rules
- Each basic rule in the composite dispatching rule has its own scaling parameter that is
chosen to properly scale the contribution of the basic rule to the final decision
ALGORITHMS OF THE IMPROVEMENT TYPE

- Start with a complete schedule, which may be selected arbitrarily
- Try to obtain a better schedule by manipulating the current schedule
- Use local search procedures which do not guarantee an optimal solution

- Attempt to find a better schedule than the current one in the neighborhood of the
current one.
- Two schedules are said to be neighbors if one can be obtained from the other through
a well-defined modification scheme
- The procedure either accepts or rejects a candidate solution as the next schedule to
move to, based on a given acceptance-rejection criterion
- The four elements of an improvement algorithm are: the schedule representation, the
neighborhood design, the search process within the neighborhood and the
acceptance-rejection criterion.
14
OPTIMIZATION APPROACHES
DISCRETE TIME MODELS OF THE NETWORK TYPE
- State-Task-Network (STN)-based discrete formulation
- Resource-Task-Network (RTN)-based discrete formulation
CONTINUOUS TIME MODELS OF THE NETWORK TYPE

- Global Time Points
* STN-based continuous time formulations
* RTN-based continuous time formulations
- Unit-Specific Time Events

* STN-based unit-specific continuous time formulations
BATCH-ORIENTED CONTINUOUS TIME MODELS

- Time Slot-based formulations
- Precedence-based formulations
* Unit-specific immediate precedence-based models
* Global direct precedence-based models
* Global general precedence-based models
15
KEY ASPECTS IN BATCH SCHEDULING

(1) PROCESS TOPOLOGY:
- Single Stage (single unit or parallel units)
- Multiple Stage (multiproduct or multipurpose)S1
- Network (batch splitting and mixing, recycling)
1
1h 2
Heat
S2
1h
Reaction1
S3
10% 2h
Separation
40%
3
2h
3h
60%
70%
S5
Reaction2
Reaction 3
30%
(2) EQUIPMENT ASSIGNMENT

- Fixed (dedicated)
- Variable
S6
(3) EQUIPMENT CONNECTIVITY

- Partial
- Full
(4) INVENTORY STORAGE POLICIES
- Unlimited intermediate storage (UIS)
- Non-intermediate storage (NIS)
- Finite intermediate storage (FIS): Dedicated or shared storage units
- Zero wait (ZW)
(5) MATERIAL TRANSFER
- Instantaneous (neglected)
-Time consuming (no-resource, pipes, vessels)
16
90%
S4
S7

(6) BATCH SIZE:
- Fixed
- Variable (mixing and splitting operations)
(7) BATCH PROCESSING TIME
- Fixed
- Variable (unit / batch size dependent)
...
0
Due
date 1
Due
date 2
(8) DEMAND PATTERNS

- Due dates (single or multiple product demands)
- Scheduling horizon (fixed, minimum/maximum requirements)
Due
date NO
Due
date 3
Production
Horizon
(9) CHANGEOVERS
i
i
- None
changeover
- Unit dependent
- Sequence dependent (product or product/unit dependent)
(10) RESOURCE CONSTRAINTS

- None (only equipment)
- Discrete (manpower)
- Continuous (utilities)
17

(11) TIME CONSTRAINTS
- None
- Non-working periods
- Maintenance
- Shifts
(12) COSTS
- Equipment
- Utilities (fixed or time dependent)
- Inventory
- Changeovers
(13) Degree of certainty
- Deterministic
- Stochastic
18
ROAD-MAP FOR OPTIMIZATION APPROACHES

(A) TIME DOMAIN REPRESENTATION
- Discrete time
- Continuous time
TASK
TIME
TASK
(B) EVENT REPRESENTATION

EVENTS
DISCRETE TIME
- Global time intervals
CONTINUOUS TIME
- Time slots
- Unit-specific direct precedence
- Global direct precedence
- Global general precedence
- Global time points
- Unit- specific time event
TIME
TASK
TIME
1
2
3
4
5
6
7
8
9
10
11
20
18
19
2
2
21
23
24
25
(C) MATERIAL BALANCES

- Lots (Order or batch oriented)
- Network flow equations (STN or RTN problem representation)
(D) OBJECTIVE FUNCTION
- Makespan
- Earliness/ Tardiness
- Profit
- Inventory
- Cost
19
DISCRETE TIME MODEL FEATURES

- DISCRETE TIME
TASK
TIME
DISCRETE TIME
- Global time interval

- Profit
- Cost
20
STATE-TASK NETWORK (STN) REPRESENTATION

The STN process representation is a directed graph consisting of three elements:
State nodes: standing for the feeds, intermediates and final products and
represented by circles.
Task nodes: representing the process operations which transform material from
one or more input states into one or more output states, and denoted by rectangles
Directed arcs: linking states and tasks to indicate the flow of materials
21
STN-BASED DISCRETE TIME FORMULATION

(Kondili et al., 1993; Shah et al., 1993)
DISCRETE TIME REPRESENTATION

The time horizon is divided into a number of intervals of equal duration (uniform time grid).
The uniform time grid is valid for all shared resources like equipment, utilities or
manpower, i.e. global time intervals.
Events of any type should occur at the interval boundaries.
It can be regarded as events:

- the start or the end of processing tasks
- changes in the availability of any resource
- changes in the resource requirement along the execution of a task
T1
T2
T3
0
8 t (hr)
22

OTHER MAJOR ASSUMPTIONS
The batch processing time for any task is constant, i.e. it does not change with the
batch size.
Dedicated storage tanks for each final/intermediate product are available.
Every batch of state/product s is transferred to the assigned tank (or the next unit)
immediately after finishing the processing task
A processing unit cannot be used as a temporary storage device.
A batch size changing with both the processing task and the assigned unit can be
selected by the model.
INTERESTING PROBLEM FEATURES

Product demands or bounds on product demands (in ton or cubic meters) are given
Batches are to be generated and scheduled by solving the model.
Alternative equipment units for a particular processing task can be available.
The resource requirement may change along the task execution.
Limited resource capacities are available
23

MAJOR MODEL VARIABLES
Wi j t
a) BINARY VARIABLES:
task
time interval
unit
Wi j t = 1 only if the processing of a batch undergoing task i in unit j Ji is

started at time point t.
b) CONTINUOUS VARIABLES:
Bi j t
= size of the batch (i, j, t)
Ss t
= available inventory of state sS at time point t
Rr t
= requirement of resource r (different from equipment)

at time point t
24

Reserved Time
Points for Task i
No task can be assigned to
time points t2 & t3 of unit j1
Equipment
Unit
Task i (Batch a)
j1
Wi, j1, t1 = 1
Just defined at
the start of task i
Task i (Batch c)
Task i (Batch b)
j2
Wi,
0
j2, t1 =
Several tasks can be assigned to

the time point t1 but executed in
different units j1 & j2
25
Time

Equipment
Unit
The batch size of task i
is just defined at the
start of the task
Task i (Batch a)
j1
The batch size of task i is

equal to zero at the other
activated time points
Bi, j1,t1= Ba
Bi, j1, t2 = Bi, j1, t3 = 0
Task i (Batch b)
j2
Bi, j2, t1= Bb

Time
26

MAJOR CONSTRAINTS
ALLOCATION & SEQUENCING
At most a single task can be performed in a particular processing unit at any time
point t.
BATCH SIZE
The size of a batch undergoing task i in unit jJi must be chosen within bounds.
MATERIAL BALANCES
The inventory of state s at time t is equal to that stored at time (t-1), plus the amount of s
produced or received as raw material from external sources, minus the amount of s
consumed in the process or delivered to the market during time interval t.
RESOURCE BALANCES
- The total demand of resource r at time interval t is equal to the sum of the rth-resource
requirements from tasks being executed at time t.
- The overall resource requirement must never exceed the maximum rth-resource capacity
CHANGEOVER TIMES
If unit j starts processing any task of family f at time t, no task i of family f can be
initiated at least (clff + ptij ) units of time before time interval t.
27

(Kondili et al., 1993; Shah et al., 1993)
t
ijt '
t '= t ptij +1
i Ij
j,t
VijminWijt Bijt VijmaxWijt

S st = S s ( t 1) +
i 'I sp
min
s
S st C
Rrt =
i
ijt
iI jf
iI jf '
BATCH SIZE
ij ( t pt is )
is Bijt + st Dst
i 'I sc
jJ i
s , t
ptij 1
max
rt
jJi
max
s
jJ i t '= 0
0 Rrt R
i , j J i , t
B
p
is
ALLOCATION AND SEQUENCING
irt '
Wij (t t ') + virt ' Bij (t t ')
s , t
MATERIAL BALANCE
r , t
RESOURCE BALANCE
r , t
t
Wijt ' 1
t '=t cl f ' f ptij +1
CHANGEOVER TIMES
j , f , f ' , t
28

MAJOR ADVANTAGES
Efficient handling of :
- limited resource availabilities, only monitored at fixed, predefined time
points
- variable resource requirement along the task execution
- other time-dependent aspects without compromising model linearity.
Batch mixing and splitting are allowed
No big-M constraint is required
Good computational performance (lower integrality gap)
Simple problem models accounting for a wide variety of scheduling features
MAJOR DISADVANTAGES
Approximate processing times can lead to sub-optimal or infeasible solutions.
The batch size B is a problem variable despite constant processing times.
Handling of small sequence-dependent changeovers is rather awkward (very fine
time discretization).
Significant increase of the model size for longer time horizons
29
NETWORK-TYPE SCHEDULING EXAMPLE

STATE-TASK NETWORK REPRESENTATION (STN)
EQUIPMENT
-HEATER
- 2 REACTORS
-STILL
30
Task
Unit
Heating
Reaction 1
Heater
Reaction 2
Reaction 3
Separation
Processing
Time (h)
1
2
Reactor 1
Reactor 2
Still
2
1
2
Number of Time Points:

10 (for H = 10 h)
Number of Binary Variables: 80
Heating:
1 x 1 x 10 = 10
Reactions
1, 2 & 3: 3 x 2 x 10 = 60
Separation: 1 x 1 x 10 = 10
Number of Continuous Variables:
B (80), S (60)
Number of Constraints:
Allocation : 40 ; Batch size: 160
Material Balances: 60
Task
Heating
Reaction 1
States Produced
Hot A
Int. AB (60%)+ P1 (40%)
States Consumed
Raw Material A
Hot A (40%) + Int. BC (60%)
Reaction 2
Int. BC
Raw Materials B&C (50/50)
Reaction 3
Impure E
Raw Material C (20%) + Int. AB (80%)
Separation
P2 (90%) + Int. AB(10%)
Impure E
31
SCHEDULING EXAMPLE
STATE-TASK NETWORK REPRESENTATION (STN)
EQUIPMENT
-HEATER
- 2 REACTORS
-STILL
DECISIONS
Heater
Allocation
Reactor 1
Sequencing
Reactor 2
Timing
Still
Heating
Reaction 3
Reaction 1
Reaction 2
Separation
profit = 2805
32
RTN-BASED DISCRETE TIME FORMULATION

(Pantelides, 1994)
MAJOR FEATURES
Similarly to the STN representation, it uses a predefined and fixed uniform time grid
that is valid for all shared resources (global time intervals)
Processing times are assumed to be independent of the batch size
It is based on the Resource-Task-Network (RTN) concept
All resources (equipment, materials, utilities) are treated in the same way
Its major advantage with regards to the STN approach arises in problems involving
identical equipment
It requires to define just a single binary variable rather than multiple ones for a set of
equipment units of similar type
Each task can be allocated to just a single processing unit
Task duplication is then required to handle alternative units and unit-dependent
processing times
Changeovers have to be considered as additional tasks.
33

MAJOR MODEL VARIABLES
a) BINARY VARIABLES:
Wi t
b) CONTINUOUS VARIABLES:
Bi t
(one less subscript)

,
Ri t
Since every task can be assigned to just a single unit, the subscript j can be
eliminated.
MAJOR MODEL PARAMETERS
i r t
i r t
= fixed amount of resource r produced/consumed by an instance of

task i at time t relative to the starting time interval t
= coefficient in the term providing the amount of resource r
produced / consumed by task i at time t that is proportional to the
batch size.
When rR stands for a processing unit, the meaning of parameters irt and irt
is somewhat different.
34

(Pantelides, 1994)
Rrt = Rr (t 1) +
iI r
0 Rrt Rrtmax
(
pti
t '= 0
irt '
Wi (t t ') + virt ' Bi (t t ') ) ) + rt
RESOURCE BALANCE
r , t
Virmin Wit Bit Virmax Wit
r , t
J
i , r Ri , t
BATCH SIZE
If resource r corresponds to a processing unit and task i requires pti units of time, then:
i r t
i r t
i r t
and:
i r t
= -1
for t =0 (Rrt decreases by one if Wit = 1)
= +1
for t =pti (Rrt increases by one if Wit = 1)
for any other t (Rrt remains unchanged even if Wit = 1)
0 for any t
35

MAJOR ADVANTAGES
Resource constraints are only monitored at predefined and fixed time points
All resources are treated in the same way
Saving in binary variables for problems involving identical equipment units
Efficient handling of limited resource availabilities
Good computational performance (lower integrality gap)
Very simple models and easy representation of a wide variety of scheduling features
MAJOR DISADVANTAGES
Model size and complexity depend on the number of time intervals
Constant processing times independent of the batch size
Sub-optimal or infeasible solutions can be generated due to the use of
approximate processing times
Changeovers have to be considered as additional tasks
36
NETWORK-TYPE GLOBAL TIME CONTINUOUS MODELS

- CONTINUOUS TIME
TASK
TIME
EVENTS
CONTINUOUS TIME
- Global time points

- Makespan
- Profit
- Cost
37
STN-BASED CONTINUOUS TIME FORMULATION

[Schilling & Pantelides (1996); Zhang & Sargent (1996); Mockus & Reklaitis (1999);
Maravelias & Grossmann (2003)]
MAJOR FEATURES
(Maravelias & Grossmann, 2003)
A common time grid that is variable and valid for all shared resources (global
time points)
A predefined maximum number of time points (N) (a model parameter)
The time points will occur at a priori unknown times (model decisions)
Every event including the start and the end of a task must occur at a time point
The start of several tasks can be assigned to the same time point n but at different
units and, therefore, all must begin at the same time Tn.
The end time of a task assigned to time point n does not necesarily occur exactly at Tn
They can finish before except those tasks following a zero wait policy (ZW)
For storage policies other than ZW, the equipment can be used as a temporary
storage device from the end of the task to time Tn
Each task can be allocated to just a single unit. Task duplication is required to
handle alternative equipment units
38

Continuous Time Representation II
Continuous Time Representation I

T1
T2
T3
T1
T2
T3
0
8 t (hr)
Schilling & Pantelides, 1996
8 t (hr)
Maravelias & Grossmann, 2003
MAJOR PROBLEM VARIABLES

a)
BINARY VARIABLES:
Wsin = denotes allocation of the start of task i to time point n

Wfin = denotes allocation of the end of task i to time point n
b) CONTINUOUS VARIABLES: Tn
= time for events allocated to time point n

Tfi n = end time of task i assigned to time point n
Tsi n = start time of task i assigned to time point n
Bsi n = batch size of task i at the start time point n
Bpi n = batch size of task i at the intermediate time point n
Bfi n = batch size of task i at the completion time point n
Ss n = inventory of state s at time point n
Rr n = availability of resource r at time point n
39
STN-BASED UNIT-SPECIFIC TIME-EVENT FORMULATION

MAJOR CONSTRAINTS
ALLOCATION CONSTRAINTS:
- At most a single task can be performed in unit j at the event time n
- A task will be active at event time n only if it stars before or at event
time n, and it finishes before time event n
- All tasks that start must finish
- An occurrence of task i can be started at event point n only if all previous
instances of task i beginning earlier have finished before n
- An occurrence of task i can finish at event point n only if it starts before n and
ends not before time point n
BATCH SIZE CONSTRAINTS
MATERIAL BALANCES
TIMING AND SEQUENCING CONSTRAINTS
STORAGE CONSTRAINTS
RESOURCE CONSTRAINTS
40

(GLOBAL TIME POINTS)
(Maravelias and Grossmann, 2003)
ALLOCATION CONSTRAINTS
Ws
iI j
in
j , n
iIj n ' n
Ws
in
in '
= Wf in
iI sp
s, n
s, n > 1
MATERIAL AND RESOURCE

BALANCES
Rrn = Rr ( n 1) irc Wsi n + irc Bsi n + irpWfi n + irp Bfi n r , n
in
in '
iI sc
S sn C smax
Wf 1 j, n
(Ws Wf ) 1
iIj
S sn = S s ( n 1) isc Bsin + isp Bfin
j , n
TIMING AND SEQUENCING

CONSTRAINTS
Tf in Tn + iWsin + i Bsin H (1 Wsin ) i, n
i , n
V iminWf in Bf in V imaxWf in i, n
Vi min Wsin ' Wf in ' Bpin

n ' n
n '< n
Vi max Wsin ' Wf in ' i, n

n ' n
n '< n
Bsin 1 + Bpi ( n 1) = Bpin + Bf in
Tn+1 Tn
Tf in Tn + iWsin + i Bsin + H (1 Wsin ) i, n
V iminWsin Bsin V imaxWsin
i, n > 1
Tf
i(n 1)
+ H ( 1 Wf
in
i, n > 1
Tf i ( n 1) Tn H (1 Wfin ) i I ZW , n > 1
Tsi 'n Tf i ( n 1) + clii '
sS j
jsn
j , i Ij , i ' Ij , n
1 j J T , n
SHARED STORAGE TASKS
S sjn CjVjsn j J , s Sj , n
T
S sn =
jJ sT
sjn
s S T , n
41

EXAMPLE
Problem Size (8 effective tasks, 8 time points)

a) Binary Variables:
it requires less time points
Ws (64) + Wf (64) = 128
b) Continuous Variables: T(64) + Tf(64) + Bs (64) + Bf (64) + Bp (64) + S (48) = 368

c) Constraints: Allocat (104) + BSize (568) + Time (264) + InvS (56)
= 992
42

MAJOR ADVANTAGES
- Significant reduction in model size by predefining a minimum number of time
points much lower than that required by discrete formulations
- Handling of processing times which vary with the batch size
- Consideration of a range of scheduling aspects
- Monitoring of resource availabilities just at the time points
MAJOR DISADVANTAGES
- Need of computing the minimum number of time points
- Model size and complexity both depending on the number of predefined time points
- Suboptimal or infeasible schedules can be generated if the number of points is
smaller than required
43
RTN-BASED CONTINUOUS TIME FORMULATION

(Castro et al., (2001, 2004))
MAIN ASSUMPTIONS
A common time grid for all shared resources
The maximum number of time points is predefined
The time at which each time point occurs is a model decision (continuous time domain)
Tasks allocated to a certain time point n must start at the same time
Only zero wait tasks must finish at a time point, others may finish before
Continuous Time Representation II
Continuous Time Representation I

T1
T2
T3
T1
T2
T3
0
8 t (hr)
8 t (hr)
ADVANTAGES
Significant reduction in model size when the minimum number of time points is predefined
Variable processing times
Resource constraints are only monitored at each time point
A wide variety of scheduling aspects can be considered in a very simple model
DISADVANTAGES
Definition of the minimum number of time points
Model size and complexity depend on the number of time points predefined
Sub-optimal or infeasible solution can be generated if the number of time points is smaller than
required
44
RTN-BASED CONTINUOUS FORMULATION

(GLOBAL TIME POINTS)
(Castro et al., 2004)
Tn ' Tn (i Winn ' + i Binn ' )
r R J , n, n' , (n < n' )

TIMING CONSTRAINTS
iI r
Tn ' Tn H 1 Winn ' + ( i Winn ' + i Binn ' )

iI ZW
iI ZW
r
r
Vi minWinn ' Binn ' Vi maxWinn '
r R J , n, n' , (n < n' )
i, n, n' , ( n < n' )
BATCH SIZE
Rrn = Rr ( n 1) + irp Win 'n + irp Bin 'n irc Winn ' + irc Binn '
iI r n '< n
n '> n
min
r
Rrn R
Vi min Win ( n + 1)
r , n
max
r
rt
p
ir
Wi ( n 1) n irc Wi ( n + 1)
Vi max Win ( n + 1)
r , n > 1
i I s , n, (n | N |)
STORAGE CONSTRAINTS
rt
rRiS
RESOURCE
BALANCE
iI S
rRiS
Vi minWi ( n 1) n
) +
Vi maxWi ( n 1) n
i I s , n, (n 1)
45

(Ierapetritou and Floudas, 1998; Vin and Ierapetritou, 2000; Lin et al., 2002; Janak et al., 2004).
MAIN ASSUMPTIONS
It is a STN-based formulation but the global time representation has been relaxed
Different tasks assigned to the same event point but performed in different units can be
started/finished at different times
The number of event points is predefined (a model parameter)
The time points will occur at a priori unknown times (model decisions)
The start and the end of a task must occur at an event point
Each task can be allocated to just a single unit. Task duplication is required to
handle alternative equipment units
It considers processing tasks i and storage tasks ist
Event-Based Representation
J1
J2
J3
2
2
3
2
3
4
8 t (hr)
46

(Janak et al., 2004)

A. BINARY VARIABLES:
Wsi n (start) , Wi n (active) ,

8 tasks x 10 event points
B. CONTINUOUS VARIABLES:
Wfi n (end)
8 x 10 x 3 = 240
Bsi n (start) , Bi n (active) , Bfi n
Tsi n , Tfi n , Tsr n , Tfr n , Ss n , Ri r n ,

8 tasks x 10 event points x 1 resource
6 states x 10 event points
(8 x 10 x 6) + 30 = 510
6 x 10
= 60
(end)
RAr n
570
47

MAJOR CONSTRAINTS
ALLOCATION CONSTRAINTS:
- At most a single task can be performed in unit j at the event time n
- A task will be active at event time n only if it stars before or at event
time n, and it finishes before time event n
- All tasks that start must finish
- An occurrence of task i can be started at event point n only if all previous
instances of task i beginning earlier have finished before n
- An occurrence of task i can finish at event point n only if it starts before n and
ends not before time point n
MATERIAL BALANCES
STORAGE CONSTRAINTS
RESOURCE CONSTRAINTS
48
STN-BASED CONTINUOUS FORMULATION

(UNIT-SPECIFIC TIME EVENT)
(Janak et al., 2004)
W
iIj
1 j , n
in
Wsin ' Wfin ' = Win
n ' n
in
Ws
in
Bin Bi ( n 1) Vi
i, n
n '< n
Ws
= Wf in
1 Wsin' + Wf in'
n' < n
n '< n
i , n
n' <n
Wf in Wsin ' Wf in '
i, n
Win
(1 W
(1 W
i ( n 1)
st
I sST
Bi st ( n1) isc Bsin

iI sc
+ Wf i ( n 1) )
st
I sst
i st n
i , n > 1
i , n > 1
i , n
Bf in Bin Vi max (1 Wf in )
i , n
i, n
Bf in Bin + Vi maxWf in
MATERIAL BALANCE
iI sp
max
i
max
Bin Bi ( n 1) Vi max
i ( n 1) + Wf i ( n 1) )
Bsin Bin
i, n
Bsin Bin + Vi maxWsin i, n
Bsin Bin Vi max (1 Wsin ) i, n
Bf in Bin
n '< n
S sn = S s ( n1) + isp Bf i ( n1) +
Win Bin V
min
i
i , n
s, n
STORAGE CAPACITY
Bi st n C max
s
s, i st I sst , n
49
STN-BASED UNIT-SPECIFIC TIME EVENT FORMULATION

(PROCESSING TASKS)
Tf in Tsin
i, n
Tf in Tsin + H Win
i, n
Tsin Tf i ( n1) + H (1 Wi ( n1) + Wf i ( n1)
Tf in ' Tsin iWsin + i Bin + H (1 Wsin
i, n > 1
) + H (1 Wfin' ) + H Wfin''
Tf in ' Tsin iWsin + i Bin + H (1 Wsin

Tsi n Tf i ( n1)
i, n > 1
Tsi n Tf i '( n1) + cli 'i + H (1 Wfi '( n1) Wsin )

Tsi n Tf i '( n1) + H (1 Wfi '( n1) )
nn ''n '
i, n, n' , (n n' )
) + H (1 Wfin' ) + H Wfin''
nn ''n '
i I ZW , n, n' , (n n' )
i, i ' , i i ' , j J ii ' , n > 1
s, i I sc , i ' I sp , j J i , j ' J i ' , j j ' , n > 1
Tsi n Tf i '( n1) + H (2 Wf i '( n1) Wsin )

s S ZW , i I sc , i ' I sp , j J i , j ' J i ' , j j ' , n > 1
50

Tf i st n Tsi st n
i st , n
Tsi st n Tf i ( n1) H (1 Wfi ( n1) ) s, i I sp , i st I sST , n > 1
Tsi st n Tf i ( n1) + H (1 Wfi ( n1) ) s, i I sp , i st I sST , n > 1
Tsi n Tf i st ( n1)
s, i I sc , i st I sST , n > 1
Tsi n Tf i st ( n1) + H (1 Wsin ) s, i I sc , i st I sST , n > 1
Tsi st n = Tf i st ( n1)
i st , n > 1
Rirn = irc Win + irc Bin
r , i Ir , n
R
R
r , n = 1
iIr
iIr
irn
+ RrnA = Rrmax
irn
+ RrnA = Rir ( n1) + RrA( n1)

CONSTRAINTS (STORAGE TASKS)
RESOURCE BALANCE
r , n > 1
iIr
Tfr n Ts rn
r , n
Tf i ( n1) Tsrn H 1 Wi ( n1) + Wfi ( n1)
r , i I r , n > 1
Tf i ( n1) Tsrn H 1 Wi ( n1)
r , i I r , n > 1
Tsrn Tsin H (1 Win ) r , i I r , n
Tsrn Tsin + H (1 Win ) r , i I r , n
Tsrn = Tf r ( n1) r , n > 1
(
(
TIMING AND SEQUENCING OF

RESOURCE USAGE
51

MAJOR ADVANTAGES
- More flexible time decisions
- Less number of event points
MAJOR DISADVANTAGES
- Definition of event points
- More complicated models
- Model size and complexity depend on the number of time points predefined
- Sub-optimal or infeasible solution can be generated if the number of time points is
smaller than required
- Additional tasks for storage and utilities
52
COMPARISON OF DISCRETE AND STN CONTINUOUS TIME FORMULATIONS
CASE STUDY: Westenberger & Kallrath (1995)

Benchmark problem for production scheduling in
chemical industry
0.31
12
Tasks
Tasks
10-12
10-12
U6/U7
U6/U7
0.5
13
zw
Task
Task 11
U1
U1
Task
Task 22
U2
U2
Task
Task 33
U3
U3
Tasks
Tasks
4-7
4-7
U4
U4
16
7
zw
0.5
6
0.2 0.7
15
14
Tasks
Tasks
13-17
13-17
U8/U9
U8/U9
17
18
Tasks
Tasks
8,9
8,9
U5
U5
10
zw
zw
19
11
53

PROBLEM FEATURES
The process includes flexible proportions of output states 3 & 4 (task 2), material
recycles from task 3, and five final states (S15, S16, S17, S18, S19)
There is enough stock of raw material (S1) and unlimited storage for the required
raw material (S1) and the final products (S15-S19).
Different intermediate storage modes are considered:
- Zero-Wait transfer policy for states (S6, S10, S11, S13)
- Finite dedicated intermediate storage (FIS) policy for the other intermediate
states
Problem data involves only integer processing times
Two alternative problem objectives are considered:
- Minimizing makespan (Case I)
- Maximizing profit (Case II)
Options A,B:
Option C:
product demands of 20 tons just for three final states have to be

satisfied
minimum product demands of (10, 10, 10, 5,10) tons for states (15, 16,
17, 18 and 19)
54
SUMMARY OF PROBLEM FEATURES
17 processing tasks, 19 states, 5 final products

9 production units
37 material flows
Batch mixing / splitting
Cyclical material flows
Flexible output proportions
Non-storable intermediate products
No initial stock of final products
Unlimited storage for raw material and final products
Sequence-dependent changeover times
0.31
Task
Task 11
U1
U1
Task
Task 22
U2
U2
Task
Task 33
U3
U3
Tasks
Tasks
4-7
4-7
U4
U4
12
9
Tasks
Tasks
10-12
10-12
U6/U7
U6/U7
zw
15
14
16
7
zw
0.5
0.2 0.7
Tasks
Tasks
8,9
8,9
U5
U5
0.5
13
Tasks
Tasks
13-17
13-17
U8/U9
U8/U9
17
18
10
zw
zw
19
11
55
CASE I: MAKESPAN MINIMIZATION

20
20
20
20
20
20
Instance
Formulation
time points
binary variables
continuous variables
constraints
LP relaxation
objective
iterations
nodes
CPU time (s)
relative gap
Discrete
30
720
3542
6713
9.9
28
728
10
1.34
0.0
A
Continuous
8
9
384
432
2258
2540
4962
5585
24.2
24.1
28
28
78082
27148
1180
470
108.39
51.41
0.0
0.0
Discrete
30
720
3542
6713
9.9
28
2276
25
4.41
0.0
B
Continuous
7
8
336
384
1976
2258
4343
4964
25.2
24.3
32
30
58979
2815823
1690
63855
66.45
3600.21
0.0
0.067
56
CASE I.B: MAKESPAN MINIMIZATION
Discrete model
Time intervals: 30
Makespan: 28
Continuous model
Time points: 7
Makespan: 32
57
CASE II.C : PROFIT MAXIMIZATION

10
10
10
H = 24 h
10
Instance D
Formulation
time points
binary variables
continuous variables
constraints
LP relaxation
objective
iterations
nodes
CPU time (s)
relative gap
240
5760
28322
47851
1769.9
1425.8
449765
5580
7202
0.122
Discrete
LB
24
576
2834
4794
1383.0
1184.2
3133
203
6.41
0.047
Continuous
UB
24
576
2834
4799
2070.9
1721.8
99692
4384
58.32
0.050
14
672
3950
8476
1647.1
1407.4
256271
1920
258.54
0.042
58
CASE II.C : PROFIT MAXIMIZATION
H = 24 h
Discrete model
Time intervals: 240
Profit: 1425.8
Continuous model
Time points: 14
Profit: 1407.4
59
COMPARISON OF DISCRETE AND STN-CONTINUOUS FORMULATIONS

COMMENTS
For Case I, instances comprising a larger number of demands were not possible to solve
in a reasonable time
Case I - Minimizing makespan:

- Both formulations reach the same objective value of 28 h
- 30 time points for the discrete model vs. 8 points for the continuous formulation
- 1.34 s (discrete model) vs. 108 s (continuous model)
- The number of time points is increased by one in each iteration until no improvement is
achieved and the reported CPU time corresponds to the last iteration
Case II - Maximizing profit:

- A fixed horizon length of 24 hours was defined (longer periods cannot be solved in a
reasonable time)
- 240 time points for the discrete model vs. 14 points for the continuous formulation
- The solution found through the discrete time model was slightly better
- With 14 points the continuous approach is faster
60
COMPARING DISCRETE VS. STN-BASED CONTINUOUS MODELS

SOME PRELIMINARY CONCLUSIONS
(1) Discrete time formulations are usually larger, but its simpler model structure
tends to reduce the CPU time if a reasonable number of time points is proposed.
(2) Discrete time models may generate better and faster solutions than the
continuous ones whenever the time discretization is a good approximation to the
real data.
(3) The complex structure of continuous time models makes them useful only for
problems that can be solved with a reduced number of time points (less than 15
time points).
(4) The model objective function selected may have a notable influence on the
computational cost.
(5) Serious limitations for solving large-scale problem instances requiring a large
number of fixed/variable time points were observed.
61
SLOT-BASED UNIT-SPECIFIC CONTINUOUS TIME FORMULATION

(Pinto and Grossmann (1995, 1996); Chen et. al. ,2002; Lim and Karimi, 2003)
MAIN ASSUMPTIONS
One of the first contributions on batch-oriented scheduling methodologies
The notion of time slots stands for a set of predefined time intervals of unknown duration
A different set of time slots is predefined for each processing unit
Batches to be scheduled are defined a priori (problem data)
Every batch is to be allocated to at most a single time slot
No mixing and splitting operations are allowed
It can be applied to a multistage sequential process with several parallel units at each stage
Batches can start and finish at any time during the scheduling horizon
task
unit
U1
U2
U3
slot
Time
62
SLOT-BASED UNIT-SPECIFIC CONTINUOUS FORMULATION

A MULTISTAGE SEQUENTIAL BATCH PROCESS
4
1
job
3
7
reaction
drying
packing
Neither the batch sizes nor the equipment capacities are model parameters
A batch size feasibility test is not required
Only batch processing times and setup times for each product at each stage are
problem data
Batch processing times can vary with the selected equipment unit
63

PROBLEM FEATURES
The process includes flexible proportions of output states 3 & 4 (task 2), material
recycles from task 3, and five final states (S15, S16, S17, S18, S19)
There is enough stock of raw material (S1) and unlimited storage for the required
raw material (S1) and the final products (S15-S19).
Different intermediate storage modes are considered:
- Zero-Wait transfer policy for states (S6, S10, S11, S13)
- Finite dedicated intermediate storage (FIS) policy for the other intermediate
states
Problem data involves only integer processing times
Two alternative problem objectives are considered:
- Minimizing makespan (Case I)
- Maximizing profit (Case II)
Options A,B:
Option C:
product demands of 20 tons just for three final states have to be

satisfied
minimum product demands of (10, 10, 10, 5,10) tons for states (15, 16,
17, 18 and 19)
64
SLOT-BASED UNIT-SPECIFIC CONTINUOUS FORMULATION

MAJOR CONSTRAINTS
BATCH ALLOCATION:
- The stage l of batch i must be allocated to just a single time slot
SLOT ALLOCATION:
- A time slot (j,k) can at most be assigned to a single task (stage l of batch i)
MATCHING CONSTRAINTS:
- If task (i,l) has been assigned to slot (j,k), then the start/end time of task (i,l)
and the start/end time of slot (j,k) must be the same
SLOT SEQUENCING:
- The slot (k+1) at every unit j cannot be started before ending the slot (j,k). No
overlap of time slots is permitted
STAGE SEQUENCING:
- The processing stage l+1 on batch i cannot be started before completing stage l
SLOT TIMING:
- The duration of slot (j,k) is given by the sum of the processing time & the setup time
for the assigned task (i,l) , if Wijkl = 1.
65
TIME-SLOT CONTINUOUS TIME FORMULATION

(Pinto and Grossmann (1995)
=1
ijkl
j kK j
lLi
ijkl
i , l Li
BATCH ALLOCATION
j , k K j
SLOT ALLOCATION
Tf jk = Ts jk + Wijkl ( pijl + suijl )

lLi
Tf il = Tsil +
W ( p
j kK j
Tf jk Ts j ( k +1)
Tf il Tsi ( l +1)
ijkl
ijl
+ suijl )
j , k K j
i , l Li
j , k K j
M (1 Wijkl ) Tsil Ts jk
BATCH TIMING
SLOT SEQUENCING
j , k K j
M (1 Wijkl ) Tsil Ts jk
SLOT TIMING
STAGE SEQUENCING
i , j , k K j , l Li
SLOT-BATCH MATCHING
i , j , k K j , l Li
66
SLOT-BASED CONTINUOUS TIME FORMULATION

ADVANTAGES
Significant reduction in model size when a minimum number of time slots is
predefined
Good computational performance
Simple model and easy representation for sequencing and allocation scheduling
problems
DISADVANTAGES
Resource and inventory constraints are difficult to model
Model size and complexity depend on the number of time slots predefined
Sub-optimal or infeasible solution can be generated if the number of time
slots is smaller than required
67
UNIT-SPECIFIC DIRECT PRECEDENCE CONTINUOUS MODEL

(Cerd et al., 1997).
MAIN ASSUMPTIONS
Batches to be scheduled are defined a priori
No mixing and splitting operations are allowed (multistage sequential processes)
UNITS
J
X 2,3,J = 1
2
X 3,5,J = 1
3
X 4,6,J =1
X 1,4,J =1
Time
The position of a batch in the processing sequence is defined in terms of its

immediate predecesor & its immediate successor and the assigned unit
Definition of time slots is not required
Sequence-dependent setup times are explicitely considered
A single-stage sequential batch process was studied
68
SLOT-BASED CONTINUOUS TIME FORMULATION

ADVANTAGES
Significant reduction in model size when a minimum number of time slots is
predefined
Good computational performance
Simple model and easy representation for sequencing and allocation scheduling
problems
DISADVANTAGES
Resource and inventory constraints are difficult to model
Model size and complexity depend on the number of time slots predefined
Sub-optimal or infeasible solution can be generated if the number of time
slots is smaller than required
69
UNIT-SPECIFIC DIRECT-PRECEDENCE CONTINUOUS MODEL

MAJOR VARIABLES
Xfi j
batch
,
unit
Xi i j
batch
unit
batch
6 batches, 1 stage
2 units per stage
(6 x 2) + (6 x 5 x 2) = 72 variables
slot-based approach
36
Xfi j = denotes that batch i is first processed in unit j

Xi i j = denotes that batch i is processed immediately after batch i in unit j
Tsi, Tfi = start/end time of batch i
Tsi
Tfi
6 x 2 = 12 variables
Slot-based approach
(3 slots per unit)
24
C. MODEL PARAMETERS:
tpi j = processing time of batch i in unit j
cli i j = setup time between batches i & i
70

MAJOR CONSTRAINTS
BATCH ALLOCATION AND SEQUENCING
- Only one batch can be first processed in a particular unit
- A batch is first processed or it has a single direct predecessor
- A batch has at most a single direct successor in the processing sequence
- A batch shares the same unit with its direct predecessor and its direct successor
BATCH TIMING AND SEQUENCING

- A batch i cannot be started before ending the processing of its direct predecessor i
- The completion time of a batch can be computed from its starting time by adding both
the sequence-dependent setup time and the unit-dependent processing time
71

(Cerd et al., 1997)
XF
ij = 1
FIRST BATCH IN THE PROCESSING SEQUENCE
iI j
XF + X
jJ i
ij
jJ i i 'I j
ii ' j 1
i 'ij
=1
FIRST OR WITH ONE PREDECESSOR
AT MOST ONE SUCCESSOR
i 'I j
XFij +
X + X
i 'ij
i 'I j
j 'J i i 'I j
j j'
ii ' j ' 1
i, j J i
SUCCESSOR AND PREDECESSOR IN

THE SAME UNIT
Tf i = Tsi + tpij XFij + X i 'ij i PROCESSING TIME
jJ i
i
'
I
j
Tsi Tf i ' + cli 'ij X i 'ij M 1 X i 'ij i , i ' SEQUENCING

jJ ii '
jJ ii '
72
MAJOR ADVANTAGES
Sequencing is explicitly considered in model variables
Sequence-dependent changeover times and costs are easy to implement
MAJOR DISADVANTAGES
Larger number of sequencing variables compared with the slot-based approach
Resource and material balances are difficult to model
73
GLOBAL DIRECT PRECEDENCE CONTINUOUS FORMULATION

(Mndez et al., 2000; Gupta and Karimi, 2003)
MAIN ASSUMPTIONS
UNITS
J
X =1
2,3
2
X 3,5 =1
3
X 4,6 =1
X 1,4 =1
Allocation variables
W 2,J = 1;W 3,J = 1 ;W 5,J = 1
W 1,J = 1; W4,J = 1 ;W 6,J = 1

Time
The formulation is still based on the immediate or direct precedence notion

The batch location in the processing sequence is given in terms of the immediate
predecessor
74

Allocation and sequencing decisions are separately handled through two different
sets of binary variables
The general precedence notion is not associated to a specific unit, i.e. it is a
global one
No time slots are predefined

(SINGLE-STAGE BATCH PROCESS)
Wi j
batch
,
unit
Xi i
batch
Xfi j
batch
batch
6 batches, 1 stage
2 units per stage
(6 x 5) + (6 x 2 x 2) =
slot-based approach
unit
54 variables
36
Xi i = denotes that batch i is processed before batch i in the same unit

Xfi j = denotes that batch i is first processed in unit j
Wi j = denotes that batch i is processed in unit j but not in first place
75

Tsi
Tfi
Slot-based approach
(3 slots per unit)
24

MAJOR CONSTRAINTS
- At most one batch i can be the first processed in unit j
ALLOCATION-SEQUENCING MATCHING
- Whenever a pair of batches are related through the immediate precedence
relationship, both batches must be allocated to the same unit
SEQUENCING CONSTRAINTS
- Every batch should be either the first processed or directly preceded by
another batch
- Every batch has at most only one successor
TIMING CONSTRAINTS
- The ending time of batch i can be computed from its starting time and the
sum of its processing time and the setup time in the allocated unit
- A batch can be started after its direct predecessor has been completed
76

(Mndez et al., 2000)
XF
ij
iI j
XF + W
ij
jJ i
jJ i
AT MOST ONE FIRST BATCH IN THE PROCESSING

SEQUENCE
j
ij
=1
ALLOCATION CONSTRAINT
XFij + Wij Wi ' j X ii ' + 1 i, i ' , j J ii '
XFij + Wij 1 X ii '
XF + X
ij
jJ i
i 'i = 1
i, i ' , j ( J i J ii ' )
SEQUENCING-ALLOCATION
MATCHING
FIRST OR WITH ONE PREDECESSOR
i'
ii '
AT MOST ONE SUCCESSOR
i'
Tf i = Tsi + tpij (XFij + Wij )
i
jJ i
Tsi ' Tf i + (clii ' j + sui ' j )Wi ' j M (1 X ii ' )
jJ i
77
MAJOR ADVANTAGES
Sequencing is explicitly considered in model variables
Changeover times and costs are easy to implement
Lower number of sequencing variables compared with the immediate precedence
approach
MAJOR DISADVANTAGES
Resource and material balances are difficult to model
Large number of sequencing variables compared with the slot-based approach
78
GLOBAL GENERAL PRECEDENCE CONTINUOUS FORMULATION

(Mndez et al., 2001; Mndez and Cerd, 2003; Mndez and Cerd, 2004))
MAIN ASSUMPTIONS
UNITS
J
X 2,3 =1
X2,5 =1
5
X 4,6 =1
X 1,4 =1
X 3,5 = 1
Allocation variables
Y2,J = 1; Y3,J = 1 ; Y5,J = 1
Y1,J = 1;Y4,J= 1 ; Y6,J = 1
X 1,6 =1
Time
The generalized precedence notion extends the immediate precedence concept

The batch location is given in terms of not only the immediate predecessor but also of
all the batches processed before in the same unit
The general precedence notion is not related to a specific unit but a global one
No time slots are predefined
79

INTERESTING FEATURES OF THE APPROACH
Allocation and sequencing decisions are divided into two different sets of binary
variables
Only one sequencing variable is required to define the relative location of any pair
of batch tasks that can be allocated to the same resource
Different types of shared renewable resources such as processing units, storage
tanks, utilities and manpower can be treated in the same way
Such renewable resources can be efficiently handled through the same set of
sequencing variables without compromising the solution optimality
80
Unit
One-Stage
batch a
batch c
Unit 3
Wa3 = Wb3 = Wc3 = 1

Xab = Xac = Xbc = 1
batch b
Time
Batches a and b are the (direct/non-direct)
predecessors of batch c
81

MAJOR VARIABLES
(SINGLE-STAGE BATCH PROCESS)
Wi j
batch
Xi i
unit
batch
i < i
batch
6 batches, 1 stage
2 units per stage
(6 x 5)/2 + (6 x 2) = 27 variables
slot-based approach
36
Xi i = denotes that batch i is processed before batch i in the same unit

Tsi

Tfi
Slot-based approach
(3 slots per unit)
24

82

MAJOR PROBLEM VARIABLES (MULTISTAGE BATCH PROCESS)
Wi l j
batch
,
unit
Xi l i l
batch
stage
6 batches, 3 stages
2 units per stage
3 x (6 x 5)/2 + (6 x 2 x 3)
slot-based approach
i < i
stage
stage batch
= 81 variables
108
Xi l i l = 1 , if task (i,l) is processed before task (i,l) in the same unit

Xi l i l = 0 , if task (i,l) is processed after task (i,l) in the same unit
Tsi l
Tsil, Tfil = start/end time of batch task (i,l)

Tfi l
6 x 3 x 2 = 36 variables
Slot-based approach
72
(3 slots per unit)
tpi l j = processing time of batch task (i,l) in unit j

cli l i l j = setup time between batch tasks (i,l) & (i,l)
83

MAJOR CONSTRAINTS
Every batch task should be allocated to only one processing unit
- Defined for every pair of tasks that can be allocated to the same unit
- Assume that tasks (i,l) & (i,l) were allocated to the same unit. If task (i,l) is
processed before, then task (i,l) starts after completing task (i,l).
- Otherwise, task (i,l) is ended before starting task (i,l)
TECHNOLOGICAL CONSTRAINT
- The stage (l+1) of batch i can be started only if stage l has been completed
TIMING CONSTRAINTS
- The completion time of task (i,l) can be computed from its starting time by adding
to it both the task processing time and the sequence-dependent setup time
84

(Mndez and Cerd, 2003)
i, l Li
Tfil = Tsil +
tp
ilj = 1
jJ il
iljWilj
ALLOCATION CONSTRAINT
i, l Li
PROCESSING TIME
jJ il
Tsi 'l ' Tf il + clil ,i 'l ' + sui 'l ' M (1 X il ,i 'l ' ) M (2 Wilj Wi 'l ' j )
i, i ' , l Li , l ' Li ' , j J il ,i 'l '
Tsil Tf i 'l ' + cli 'l ',il + suil M X il ,i 'l ' M (2 Wilj Wi 'l ' j )
Tsil Tf i ( l 1)
i, l Li , l > 1
i, i ' , l Li , l ' Li ' , j J il ,i 'l '
STAGE PRECEDENCE
85

ADVANTAGES
General sequencing is explicitly considered in model variables
Changeover times and costs are easy to implement
Lower number of sequencing decisions
Sequencing decisions can be extrapolated to other resources
DISADVANTAGES
Material balances are difficult to model
86
CASE STUDY FOR COMPARING BATCH-ORIENTED APPROACHES
1
Extruders
Batches to be
Processed
B1
B2
B3
....
....
B11
B12
Time Horizon = 30 days
One operator crew per extruder
3
Limited Manpower:
Case 1: 4 operator crews
87
CASE STUDY FOR COMPARING BATCH-ORIENTED APPROACHES-2

OPTIMAL SCHEDULES
(a) without manpower limitation
(b) 3 operator crews
88

OPTIMAL SCHEDULES
(b) 3 operator crews
(c ) 2 operator crews
89

Case
Study
Event
representation
Binary vars, cont. vars,

constraints
Objective
function
CPU time
2.a
Time slots &

preordering
100, 220, 478
1.581
67.74
(113.35)*
General precedence
82, 12, 202
1.026
0.11b
Time slots &

preordering
General
precedence
289, 329, 1156
2.424
127, 12, 610
1.895
2224
(210.7)*
7.91b
Time slots &

preordering
289, 329, 1156
8.323
76390
(927.16)*
General precedence
115, 12, 478
7.334
35.87b
2.b
2.c
90
BATCH REACTIVE SCHEDULING

Roslf, Harjunkoski, Bjrkqvist, Karlsson and Westerlund (2001); Mndez and Cerd
(2003, 2004)
An industrial environment is dynamic in nature and the proposed schedule must usually
be updated in midweek because of unexpected events.
Different types of unexpected events may happen like:
- changes in batch processing/setup times
- unit breakdown/startup
- late order arrivals and/or orders cancellations
- reprocessing of batches
- delayed raw material shipments
- modifications in order due dates and/or customer priorities
To prevent rescheduling actions from disrupting a smooth plant operation, limited
changes in batch sequencing and unit assignment are just permitted.
The goal is to meet all the production requirements still to achieve under the new
conditions by making limited low-cost changes in batch sequencing and assignment.
91
BATCH REACTIVE SCHEDULING

Roslf, Harjunkoski, Bjrkqvist, Karlsson and Westerlund (2001); Mndez and Cerd
(2003, 2004)
A minimum deterioration of the problem objective (minimum makespan, minimum
tardiness) is pursued.
PROBLEM DATA
- the schedule in progress
- the present plant state
- current inventory levels
- present resource availabilities
- current time data
- unexpected events
- allowed rescheduling actions
- The criterion to be optimized
RESCHEDULING ACTIONS
Proper adjustments to the current schedule may include:
- simultaneous local reordering of old batches at some equipment units
- reassignment of certain old batches to alternative equipment items due to
unexpected unit failures
- insertion of new batches
- batch time shifting
92
BATCH REACTIVE SCHEDULING CONTINUOUS APPROACH

Mndez and Cerd (2003, 2004)
It is based on the global general precedence notion
It is a batch-oriented iterative approach
It allows simultaneous insertion and reallocation of new/old batches as well as the
resequencing of old batches at each iteration
It considers sequence-dependent changeovers and limited renewable resources
A limited number of rescheduling actions can be applied in order to reduce the problem
size as much as the scheduler wants
At each iteration of the rescheduling algorithm, two steps are sequentially executed:
- the assignment step during which new batches are inserted and a limited number of old
batches can be simultaneously reallocated
- the sequencing step where neighboring batches in the same queue can exchange
locations, and the procedure is repeated until no improvement in the objective function
is observed
93
REACTIVE SCHEDULING
Problem Variables
New batch to
be inserted
94
REACTIVE SCHEDULING EXAMPLE

A single-stage multiproduct batch plant with four units working in parallel
Forty batches are to be processed within a 30-day scheduling horizon.
An unexpected 3-day maintenance period for unit U3 at time t = 14.6 d makes

necessary to perform a rescheduling process
At the rescheduling time, 25 batches are still to be processed
95

UNEXPECTED 3-DAY MAINTAINANCE OF UNIT U3
BATCH RESEQUENCING
SELECTED BATCH REALLOCATION
SCHEDULE IN PROGRESS
96

NEW IMPROVED SCHEDULE
Only batch time shifting in unit U3:

New improved schedule:
Total Tardiness
13.55 d
8.84 d
97
CONCLUSIONS
Current optimization models are able to solve moderate-size batch processes
Small examples can be solved to optimality
Discrete-time models are computationally more effective than continuous-time models of the
network type
Difficult selection of the number of time or event points in network-oriented continuous time
formulations
Network-oriented continuous-time models become quickly computationally intractable for
scheduling of medium complexity process networks.
Problems with more than 150 time intervals are difficult to solve using discrete time models
Problems with more than 15 time or event points appear intractable using network-oriented
continuous time models.
Depending on the objective function, different computational performances are observed
Batch-oriented continuous approaches are computationally more efficient but usually
require to first solve the batching problem (a decomposition approach)
Combining other approaches with mathematical programming (hybrid methods) for solving
large scale problems looks very promising
98

GFGHDGH KK

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

GFGHDGH KK

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

GFGHDGH KK

Uploaded by

Copyright:

Available Formats

OPTIMIZATION METHODS FOR

SCHEDULING DECISIONS AND GOALS

ROLLING TIME HORIZON

6-week rolling horizon

INTERACTION WITH THE MRP SYSTEM

TYPES OF SCHEDULING PROBLEMS

FLOW SHOP vs. JOB SHOP PROBLEMS

TYPES OF BATCH PRODUCTION FACILITIES

TYPES OF SCHEDULING PROBLEMS -2

TYPES OF SCHEDULING APPROACHES

ALGORITHMS OF THE IMPROVEMENT TYPE

HEURISTIC SCHEDULING METHODS

ALGORITHMS OF THE IMPROVEMENT TYPE

- Use local search procedures which do not guarantee an optimal solution

CONTINUOUS TIME MODELS OF THE NETWORK TYPE

- Unit-Specific Time Events

BATCH-ORIENTED CONTINUOUS TIME MODELS

KEY ASPECTS IN BATCH SCHEDULING

(2) EQUIPMENT ASSIGNMENT

(3) EQUIPMENT CONNECTIVITY

KEY ASPECTS IN BATCH SCHEDULING

(8) DEMAND PATTERNS

(10) RESOURCE CONSTRAINTS

KEY ASPECTS IN BATCH SCHEDULING

ROAD-MAP FOR OPTIMIZATION APPROACHES

(B) EVENT REPRESENTATION

(C) MATERIAL BALANCES

DISCRETE TIME MODEL FEATURES

(B) EVENT REPRESENTATION

(C) MATERIAL BALANCES

- Network flow equations (STN or RTN problem representation)

STATE-TASK NETWORK (STN) REPRESENTATION

STN-BASED DISCRETE TIME FORMULATION

DISCRETE TIME REPRESENTATION

Events of any type should occur at the interval boundaries.

It can be regarded as events:

STN-BASED DISCRETE TIME FORMULATION

INTERESTING PROBLEM FEATURES

STN-BASED DISCRETE TIME FORMULATION

Wi j t = 1 only if the processing of a batch undergoing task i in unit j Ji is

= size of the batch (i, j, t)

= available inventory of state sS at time point t

= requirement of resource r (different from equipment)

STN-BASED DISCRETE TIME FORMULATION

Several tasks can be assigned to

STN-BASED DISCRETE TIME FORMULATION

The batch size of task i is

Bi, j1, t2 = Bi, j1, t3 = 0

Bi, j2, t1= Bb

STN-BASED DISCRETE TIME FORMULATION

STN-BASED DISCRETE TIME FORMULATION

VijminWijt Bijt VijmaxWijt

ALLOCATION AND SEQUENCING

Wij (t t ') + virt ' Bij (t t ')

t '=t cl f ' f ptij +1

STN-BASED DISCRETE TIME FORMULATION

NETWORK-TYPE SCHEDULING EXAMPLE