Copyright © IFAC Management and Control of Production and

Logistics, Campinas, SP, Brazil, 1997

SOLVING A HOIST SCHEDULING PROBLEM AS A SEQUENCING

PROBLEM

C. Caux and H. Pierreval

Equipe de recherche en systemes de production de l'IFMA

Laboratoire inter-universitaire d'lnformatique, de Modeiisation et
d'Optimisation des Systemes
IFMA - Campus des Cezeaux - BP 265
F-63175 AUBIERE cedex
FRANCE
e-mail: caux@iftna.ifmafr and pierreva@iftna.ifmajr

Abstract: We propose in this paper a method to minimize the makespan of a set of jobs in
a no-wait jobshop where transports are made by a hoist. Our approach consists in solving
the no-wait jobshop scheduling problem as a classical sequencing problem. For this, two
heuristics are provided to transform the scheduling problem into a sequencing problem.
These heuristics are aimed to determine, for each permutation of jobs, entry dates that meet
the no-wait and minimize the makespan. Then, an adaption of the simulated armealing
algorithm is implemented to solve the sequencing problem.
This approach is tested and compared with Rajendran's heuristic in the case of no-wait
flowshop and yields quite good results. Copyright © 1998 IFAC

Keywords: Heuristics, Iterative methods, Optimization problems, Scheduling algorithms

1. IN1RODUCTION schedule is said to be feasible if it is free of the two

following conflicts :
Certain jobshops such as electroplating systems or
chemical systems are submitted to the following • A station conflict may occur when a job
particular constraints. Once the processing of a job finishes a process in a station and is ready to
begins, subsequent processing must be carried out move to its nex't station, but the next station
with no delay in the passage of this job from station is busy. Since the capacity of stations is one,
to station. If necessary, the job is delayed on the first the schedule is unfeasible.
station, so that the job needs not to wait for
processing on subsequent stations (Rajendran 1994). • A hoist conflict may occur when a job finishes
These particular jobshops are called no-wait in a station but the hoist is still transporting
jobshops . Moreover, in many cases, no-wait another job. The job has to wait for the hoist
jobshops are served by one or more hoists which to finish its current assignment. The schedule
ensure the transportation of jobs from a station to is then unfeasible.
another.
No-wait jobshop scheduling (this problem is also This problem has been proven to be NP-complete,
known as hoist scheduling problem or HSP if even if there is only one hoist and one product type
transportation times between stations are considered) in the system (Lei and Wang 1989).
is a difficult problem which consists in determining No-wait flowshops have already been studied.
schedules that minimize the processing duration and Sriskandarajah (1993) gives some heuristics with
that are feasible. Because there is no in-process performance guarantees for the no-wait flowshop with
storage between any two stations, a schedule can be parallel stations. Chan (1993) proposes an approach
not necessary feasible . According to Yih (1994), a based on the precedence constraint travelling

315
salesman problem. Rajendran (1994) gives a fast There are n type of jobs. A type is defined by a
heuristic for flowshop which we will use in sequence of stations to visit and by the associated
comparison to our approach. processing times. A same station can appear several
The cyclical case of the HSP has also been studied in times in a sequence. We assume that a sequence
the case where there is one hoist and one product always starts from the input buffer and always ends in
type. Phillips and Unger (1976) and Shapiro and the output buffer. There is one hoist which is
NuttIe (1988) propose mathematical programming responsible for transporting jobs through the system.
solutions for the HSP. Chen, Chu and Proth propose This hoist has a capacity of one job and its speed is
a model and an algorithm based on graph theory to constant.
fmd the optimal cyclic schedule of hoist moves. The We consider a given set of jobs the problem is to
case with two hoists has been studied by Lei & determine, for each job, an entry date such as:
Wang (1991). Reactive approaches (on-line
scheduling) are founded in the literature. For • the schedule is feasible : the job processing can
example, Thesen and Lei (1990) have developed an be performed entirely without any waiting
expert system and Yih (1994) proposed a method for time or deadlocks. This means that the hoist
dynamically scheduling the incoming jobs. is available as soon as the process of a job is
completed and that a station is available when
Unfortunately, the no-wait jobshop scheduling (with the hoist drops off ajob.
transportation times) seems not to have been studied
in the literature. Moreover, Manier and Baptiste • the makes pan is as low as possible.
(1994) mention in the conclusion of their HSP state
of the art that all the reviewed studies are limited to We assume that, when the process of a job on a
the one product type case except the on-line approach station is completed, the job can wait in this station
proposed by Thesen and Lei (1990). during the hoist moves from its current position to
this station. This moving time is supposed to be
We propose in this paper a method for no-wait constant, whatever the current position of the hoist
jobshop scheduling which can be applied in the case is. This assumption is reasonable because most et
where there is one hoist and several product types. the time, the processing times are not constant but
Our approach consists in solving the no-wait jobshop bounded by a minimum and a ma'{imum value or are
scheduling problem as a classical sequencing given with a tolerance.
problem. This sequencing problem is solved by In our approach, we consider the hoist as a station a.
using an optimization method, such as the simulated the jobshop. The processing time on this station
annealing. This method proposes at each iteration, a depends on the distance between the initial station
sequence of jobs. Then, entry dates of each job of the and the destination station.
sequence are determined thanks to an heuristic we
propose. This heuristic ensures that the obtained
schedule is feasible and minimizes the makespan 2.2 Sequencing problem formulation

To this end, we fIrst formalize the problem to solve In a permutation jobshop scheduling problem, n
as a sequencing problem. The second section details jobs available at time 0 have to be performed. Each
the proposed approach and presents two heuristics job i consists in ni operations which are executed on
which permit to solve the problem to be solved as a specific machines. The jth operation of the job i is
sequencing problem. The following section ell."}}lains denoted (i,j). The operations of a job i must respect
the implementation of the simulated annealing to the sequence (i. I). (i.2), ... , (i.nJ . The operation (i,j)
solve the sequencing problem. Finally, uses the machine miJ during a time PiJ The starting
computational results are given and our approach is time of the operation (i,j) is denoted tiJ and the
restricted to no-wait flowshop so that the comparison ending time (or completion time) is denoted C;J.
with Rajendran's heuristic is possible. Solving a permutation jobshop scheduling problem
consists in finding a permutation of jobs that
minimize a criterion.
2. PROBLEM STATEMENT In the case of a no-wait jobshop, a solution must, in
addition, verify the following constraint:
2. J Problem description
tiJ = CiJ-1, I 5i 5n and 2 5j 5ni (Cl)
In this study, we consider a no-waitjobshop made et
M stations, including an input buffer and an output Unfortunately, a permutation only indicates the job
buffer as depicted by figure 1. Each station has a order while in a no-wait jobshop scheduling
capacity of one job. problem, the entry date of each job must be
------rtl....-----------
hois"-t
determined to meet the constraint Cl . Classical
methods for the permutation jobshop problem are no
longer suited to the no-wait jobshop problem.
, ···'UUl!J l..!JU'··
o 2 3 M-I M
We propose in the following to solve the no-wait
input buffer stations output buffer
Fig. l. No-wait jobshop jobshop problem as a classical permutation problem
by using an heuristic to compute entry dates that

316
meet the constraint Cl from a given permutation. fictitious job. At the end of the first iteration, a
The sequencing problem is then solved by an sequence of nl2 fictitious jobs is obtained. The
iterative method such as simulated annealing. At heuristic is then recursively called until only
each iteration, this method proposes a sequence et fictitious job is obtained.
jobs. Then, for each job of the sequence, the heuristic
computes the entry dates so that the schedule is Step I Pair the jobs of the sequence (jo.N.
feasible and returns the makespan corresponding to (jl.h) ... · (jn-l.j,J
the sequence of jobs. Step 2 For each pair (ji.ji+l)
Step 3 Set ti = 0 and ti + l = 0
Step 4 Delay j;+l until the schedule (ji.ji+l) is
3. TIlE PROPOSED APPROACH feasible
Step 5 Aggregate the jobsji andji+l
Given a permutation, the problem is to determine Step 6 Goto to step I until only one job is obtained
good entry dates tiJ of each operation j of a job i.
Because of the constraint Cl, only the entry date of a The makespan is then the completion time of the last
job on the first station is to be determined. These aggregated job retwned by the heuristic. The
entry dates must meet the constraint Cl and must complexity of this heuristic i~ O(n2) .
respect the order of the operations specified by the
permutation. We propose in the two following
subsections two heuristics aimed at converting a 4. SEQUENCING PROBLEM
permutation of jobs in a list of entry dates.
4. 1 General prinCiples
3.1 The earliest start heuristic The simulated annealing (Kirkpatrick 1983) is an
optimization method based on ideas from statistical
The earliest start heuristic computes the entry date et mechanics and motivated by an analogy to the
each job of a sequence (without modifying their behaviour of physical systems in the presence of a
order). In order to understand the principle of this heat bath. The simulated annealing method can
heuristic, let us consider a sequence of two jobs j 1 obtain an optimum or near optimum solution by
andj2 and a particular operation on a given station k. using stochastic principles to avoid to be trapped in
The processing time ofj1 (resp.j2) on this station is local minima. Indeed, general deterministic iterative
denoted PJl.k (resp. P)1,k) and starts at the date !.il.k improvement methods accept those solutions that
(resp. tp,k) . If the time windows [ tjU • tjl,k+ pjJ.k } decrease cost value, so that the search fails in certain
and [tp,k. tp,k+P)1,k J do not overlap, the schedule local optima. To avoid this, simulated annealing is
may be feasible. Conversely, if the time windows based on the following principles. The state
overlap, the operation starting at tp,k must be delayed transition by which the value of the cost function is
by (tjl ,k + Pjl,Jr) - tp,t to avoid the resource conflict. increased is also accepted with a certain probability
By extending this principle to all the operations et which is given by exp(-iljlT) where T is a parameter
two consecutive jobs, we progressively build a called temperature and ilj denotes the differen:e in
feasible schedule for a sequence of n jobs as follows: cost between the candidate solution and the current
solution. Decreasing the temperature T at a low rate
Step I Set tjl,l = 0 (the entry date of the first job on permits the control of the acceptance probability. A
the first machine is set to 0) classical simulated annealing method is the
Step 2 For each job j i in (j1. ....j n ) following:
Step 3 Set ti,l = t i-l,l (the entry date of a job is
set to the entry date of the previous job Step I generate an initial solution S
in the sequence) Step 2 generate a neighbour S' of S
Step 4 While conflicts are detected in the Step 3 il = cost(S') - cost(S)
schedule (jl .... .jJ. delay the job ji Step 4 if L1 5" 0 then S = S'
Step 5 if L1 > 0 then S = S' with probability
The makespan is then the maximal value of the exp(-flIT)
completion times of the jobs. The complexity of this Step 6 go to step 2 until a stopping criterion is
heuristic is O(n 2). reach:d

The temperature T is given by C / In(n+ 1) where C

3.2 The aggregation heuristic is a parameter called initial temperature and n is the
iteration number.
The aggregation heuristic shares common principles
of the earliest start heuristic. The jobs of a sequence
are paired (the first with the second, the third with
4.2 Adapting the simulated annealing to the
the fourth, etc). Then, for each job of a pair, we
permutation jobshop problem
assume the entry date of the first job is 0 and the
second job of the pair is delayed, as described in the
The use of simulated annealing for a particular
previous section, until the constraints are met. Then,
problem requires the definition of a coding of the
for each pair, the two jobs are aggregated into one

317
 Lahle I No-wait iohshoil. - makesil.ans

Case Processing times No schedule SA SA improvement

+ earliest start + ag~egation
1 from 10 to 30 7943 5890 5897 26%
2 from 30 to 50 14503 6997 6838 53%
3 from 50 to 100 26883 10142 10298 62%
4 from 200 to 400 101882 29326 34205 71 %

Table 2 No-wait (lowshop - makespans

Case Optimal value Rajendran Simulated annealing

heuristic + earliest start
1 474 486 474
2 2084 2092 2099

solutions and the definition of a method to obtain a 5.2 Comparison with Rajendran's heuristic for no-
neighbour from a solution. In the case of sequencing wait jlowshop
problems, a solution is a sequence of jobs. A
neighbour of a solution can be obtained by It is very difficult to evaluate the quality of the above
pennutation. To this end, two different jobs are results because the optimal value is unknown as well
randomly chosen in the sequence and they are as lower bounds. In this section, we compare the
pennuted. Permutation ensure that the search space is results given by our method with the results of a
connected and that all solutions are reachable. heuristic proposed by Rajendran (1994). This
heuristic minimizes the makespan for a no-wait
flowshop . We consider two cases. The first one
5. COMPUTATIONAL RESULTS contains 10 jobs and 10 stations and the processing
times range from 15 to 30. The second case contains
5. 1 No-wait jobshop results 10 jobs and 10 stations and the processing times
range from 4 to 364. Transportation times are not
The method we propose has been considered since Rajendran's heuristic do not take
implemented on a SUN SP ARC 10 workstation. them into account.
The programs are implemented in C language. In
order to test our method, we study a no-wait jobshop The table 2 gives the optimal value (computed by
composed of 18 stations, (16 workstations, an input enumeration) and heuristic results. Only one of our
buffer and an output buffer). The number of jobs to heuristic is mentioned because in the case ci
schedule is 36. There are 10 types of job. A type ci flowshops, it is easy to show that the two heuristics
job consists in a sequence of 8 to 10 stations we proposed yield the same result. Our general
randomly chosen among the 16 stations. The approach gives good results compared with an
transportation time between stations i and j is I i-j I approach dedicated to flowshops . The number ci
and the time during which the idle hoist moves from iterations peIforrned during the simulated annealing
its current position to a station is constant and set to is 200 000 and the computational time is about 8
18 (the number of stations). We consider four cases minutes.
for the processing times. These cases represents
different values of the ratio processing times /
transportation times. 6. CONCLUSION
Results are given in table 1. The column "no
schedule" corresponds to the case where jobs are In this paper we have proposed a method for
processed one by one. Because there is only one job minimizing the makespan of a set of jobs in a 00-
in the jobshop at a given time, no conflicts are wait jobshop with transport. Although this problem
possible. The number of iterations peIforrned during is not a pennutation problem, our approach permits
the simulated annealing is 200000. The to use classical methods for permutation problem
computation speed is about 400 iterations per minute such as simulated annealing. This is possible thanks
for the annealing + aggregation method and about to a heuristic which determines the entry date of each
800 iterations per minute for the annealing + earliest job in order to meet the no-wait constraints. We have
start method. two heuristics: the earliest start heuristic and the
The table 1 shows that the best results are obtained aggregation heuristic. This approach can take into
in the case where the processing times are long. account several job types and seems promising as
Conversely, in the case where processing times are shown by the results obtained. Further work will
short, the improvement is limited because the hoist consist in improving the heuristics used to determine
is overburdened and becomes a very critical resource the entry date of jobs.
in the no-wait jobshop.

318
 REFERENCES

Chen, H., C. Chu and 1.-M. Proth (1995). Cyclic

hoist scheduling based on graph theory.
INRJAlIEEE Symposium on Emerging
Technologies and Factory Automation, 451-
459.
Chan, D . (1993). Precedence constrained TSP
applied to circuit board assembly and no wait
flowshop. 1nl. J Prod. Res., 31(9), 2171-2177.
Kirkpatrick, S., C.D. Gelatt lr and M.P. Vecchi
(1983). Optimization by simulated annealing.
Science, 220, 671-680.
Lei, L., T.l. Wang (1989). A proof: the cyclic hoist
scheduling problem is NP-complete. Working
paper #89-0016, Rutgers University.
Lei, L., T.l. Wang (1991). The minimum common-
cycle algorithm cyclic scheduling of two
material handling hoists with time window
constraints. Management &ience, 37(2), 1629-
1639.
Manier, M .A., P. Baptiste (1994). Etat de rart :
ordonnancement de robots de manutention en
galvanop1astie. Revue Automatique Productique
Informatique Industrielle, 28(1), 7-35.
Phillips, L.W., P.S. Unger (1976). Mathematical
programming solution for a hoist scheduling
program. AIIE transactions, 8(2), 219-225 .
Rajendan, C. (1994). A no-wait flowshop scheduling
heuristic to minimize makespan. J Op/. Res.
Soc., 45(4), 472-478.
Shapiro, G.W, H.L.W. Nuttle (1988). Hoist
scheduling for a PCB electroplating facility. lIE
transactions, 20(2), 157-167.
Sriskandarajah, C. (1993). Performance of scheduling
algorithms for no-wait flowshops with parallel
machines. European Journal of Operational
Research, 70, 365-378.
Thesen, A. , L. Lei (1990). An ex-pert scheduling
system for material handling hoists. Journal of
Manufacturing Systems, 9(3), 247-252.
Yih, Y. (1994). An algorithm for hoist scheduling
problems. Int. J Prod. Res. 32(3), 501-516 .

319