You are on page 1of 3

Chapter 2

Analysis
In this chapter it will be defined the FSSPLB and presented the related works ...

2.1 Flow shop scheduling problem with limited capacity buffers


In the following section we will analyze the FSSPLB. Firstly we will define the problem, after
that it will be explained a mathematical model and finally the first version of the algorithm will be
presented.

2.1.1

Problem definition

The FSSPLB is a generalization of the classical FSSP and it can be formulated as follows:
There are m machines M1 , ...Mm and m 1 buffers Bi between consecutive machines Mi and
Mi+1 with the capacity of bi units (i = 1, ..., m 1). There are n independent jobs ( j = 1, ..., n) that
should be processed on these machines. The order of machines is given and the jobs are processed
on m machines in the same sequence. Processing time of job j on machine i is shown by pi, j . Each
job consists of m operations Oi, j (i = 1, ..., m) and operation Oi, j has to be processed on machine Mi
without preemption for pi, j 0 time units. Between the operations of each job there are precedence
relations in form of a chain O1, j O2, j O3, j ... Om, j .
When a particular type of machine is busy, the incoming parts will be moved to the intermediate
buffer (if the buffer is not full) or will be blocked on the same machine.
A feasible schedule of the jobs is given by an assignment of the starting times Si, j (and, thus,
completion times Ci, j = Si, j + pi, j ) to operations Oi, j (i = 1, ..., m; j = 1, ..., n) such that:
the precedence relations within the jobs are respected (Ci, j Si+1, j ),
at any time, each machine can processes at most one job and each job can be processed only
on one machine. ([Si, j ,Ci, j ) [Si,k ,Ci,k ] = for j 6= k)
in each buffer Bi may be stored at most bi jobs at the same time, and
the blocking-restrictions are respected (explained later).
Lets continue with detailed description of the buffers and blocking restrictions for feasible schedules:

2.1. FLOW SHOP SCHEDULING PROBLEM WITH LIMITED CAPACITY BUFFERS

2.1.2

The mathematical model

The mathematical model is represented based on the research of Nowicki [16] and Leisten
[12]. In order to provide a formal mathematical model, we apply the notation for the set of jobs
J = 1, 2, ..., n, the set of machines M = 1, 2, ..., m and the job processing order represented by the
permutation = ((1), (2), ..., (n)) on the set J, where ( j) denotes the element of J which is in
position j in . A feasible schedule is defined by start times Si j , i M, j J, of job j on machine
i such that the constraints (1), (2) and (3) are satisfied. The problem is to find a feasible schedule
that minimizes the makespan max jJ (Sm j + pm j ). Let denote the set of all permutations on the
set J and let Ci j = Si j + pi j denote the time of completion of job j on machine i. Each permutation
generates many feasible schedules. For providing a feasible schedule for the given if and only if
the following constraints are satisfied for any i M, j J:
Si, ( j) Si1, ( j) + pi1, ( j) = Ci1, ( j) ,

(2.1)

Si, ( j) Si, ( j1) + pi, ( j1) = Ci, ( j1) ,

(2.2)

Si, ( j) Si+1, ( j fi+1 1) ,

(2.3)

where ( j) = 0 if j 0, Si,0 = 0, pi,0 = 0, i M {m + 1}, Fm+1 = 0, Sm+1, j = 0, j J, S0, j = 0,


p0, j = 0, j J.
Condition (Eq. 2.1) represent the route of jobs through machines, condition (Eq. 2.2) follow
from the given sequence of processing jobs on each machine, whereas conditions (Eq. 2.3) are
associated with limited buffer capacities. The schedule satisfying constraints (Eq. 2.1 - 2.3), which
minimize the makespan (for the sake of simplicity we will denote it also by Si, j and Ci, j ) we can use
the recursive formula:
Si, ( j) = max(Si1, ( j) + pi1, ( j) , Si, ( j1) + pi, ( j1) , Si+1, ( j fi+1 1) )
f or i = 1, 2, ..., m, j = 1, 2, ..., n.

(2.4)

Now the stated problem can be formulated as that of finding a job processing order which
minimizes the makespan Cmax () = Sm, (n) + pm, (n) , where Sm, (n) follows from equation (Eq. 2.4).
Let Ci, j denote the time of releasing the machine i after the completion of job j. So, it has to be:
Ci, ( j) = max(Ci, ( j) , Si+1, ( j fi+1 ) ), i M, j J.

2.1.3

(2.5)

The proposed algorithm

The algorithm presented in this section is proposed by Nowicki [16] [17] and it uses the local
search method based on tabu search (TS) technique. We will try to use different heuristics to analyze
if we can find a better one than the heuristic proposed by Nowicki. It has been shown that tabu search
technique is very efficient in finding the solution close to optimal solutions of many optimization
problems including scheduling field. The basic approach of the TS has fundamentals elements move,
neighborhood, initial solution, searching strategy, memory, aspiration criterion and stopping rule.
In continuation we will present the preparing steps of the algorithm that should be done before the
proper Nowickis tabu search algorithm.

You might also like