Professional Documents
Culture Documents
discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/255565008
CITATION
DOWNLOADS
VIEWS
56
33
2 AUTHORS:
Erwin W Hans
Alain Hat
University of Twente
SEE PROFILE
SEE PROFILE
Introduction
We study multi-resource multi-project capacity planning at the tactical level (mediumterm planning); following Hans (2001), we refer to our problem setting as rough-cut
capacity planning (RCCP). Our work is part of the development of a decision support
system for helicopter maintenance, within the framework of a French industrial project
called Hlimaintenance, a European project that aims to build a center for civil helicopter
maintenance. Aircraft maintenance consists of carrying out all the necessary actions to
guarantee the required level of reliability, safety and operational capacity of the aircraft.
Helicopters have a number of specicities compared to airplanes: compact body, centralized equipment, higher mechanical stress, . . . (Poncelin et al. 2006). Helicopter maintenance planning and scheduling has already been studied in the military domain (Hahn and
Newman 2008, Sgaslik 1994), but compared to military helicopters, a civil eet tends to
be more heterogeneous and the maintenance center is typically external to the exploitation, whereas in the military sector maintenance planning is integrated with planned and
expected missions. To the best of our knowledge, only little work has been published on
planning of civil helicopter maintenance (Poncelin et al. 2006), and none on the eective
timing and resource management of the maintenance tasks (see Masmoudi (2011) for more
background).
Our overall object of study is the capacity management of the specic overhaul type that
is called Heavy Maintenance Visit (HMV), which are the most severe check-ups, aecting
all parts of the machine (structure, avionics, mechanics) and which can last up to several
months. The HMV entails planned maintenance tasks and also corrective maintenance,
because problems may be discovered during the inspection of the helicopter. Precedence
constraints exist, due to technical and accessibility considerations. Consequently, an HMV
may be seen as a project involving various resources, equipment and spare parts; our
capacity management problem therefore classies as multi-project management.
Managing aeronautical maintenance is a complex task because it is aected by many uncertainties. The three problem parameters most susceptible to suering deviations are the
release date (arrival time), the inspection tasks durations, and the procurement throughput times. In this paper we study the integration of uncertainty into RCCP, aiming for the
development of robust plans, which are, up to a certain degree, protected from variability
in the parameters. We propose a simulated-annealing meta-heuristic to solve the capacity planning problem and validate its performance against exact and heuristic solution
approaches in a deterministic as well as a stochastic setting.
Problem statement
At the tactical level, a project is viewed as a set of tasks that are subjected to precedence
and resource constraints. RCCP allocates the available workforce to these tasks in order
to complete the tasks within their time windows at a minimum cost. Hans (2001) has
solved a deterministic variant of this RCCP by means of a branch-and-price algorithm. His
model is the starting point of the problem studied in this paper; the main modication is
the explicit incorporation of uncertainty regarding the workloads, in line with the multiproject context of helicopter HMVs. Wullink et al. (2004) have extended Hans model
with dierent scenarios based on a discretization of the stochastic work content. Their
objective is to minimize the expected costs over all scenarios. This scenario-based approach,
however, turns out to be computationally very demanding, even when only a small sample
of scenarios is considered. We will therefore not model uncertainty by discrete scenarios
but rather focus only on expectation and variance of the work content.
2.1
Deterministic RCCP
(1)
subject to
Ybjt
1
bj
Xj = 1
jN
(2)
abjt Xj 0
j N ; b Nj ; t = 1, . . . , T
(3)
j N ; b Nj
(4)
i I; t = 1, . . . , T
(5)
i I; t = 1, . . . , T
(6)
i I; t = 1, . . . , T
(7)
(8)
j N ; j
(9)
jN
t=1
bNj
Ybjt = 1
Xj {0, 1}
In this formulation, i1t is the regular capacity available of resource i in period t, i2t is
the sum of regular and overtime capacity, and i3t equals i2t augmented with the capacity
available by hiring non-regular temporary (interim) sta. Variable Oit counts the number
of overtime hours of resource i in period t, Hit represents the number of hours performed by
interim workers and Sit is the number of subcontracted hours. The objective function (1) is
a linear function that minimizes the cost of non-regular capacity; overtime is less expensive
than interim work, which in turn is cheaper than subcontracting. The concept of order
plans stems from Hans (2001); these incorporate the time windows and the precedence
constraints. Constraints (2) and (9) ensure that exactly one order plan is selected for
each project. Constraints (3) species a minimum duration bj for task (b, j) and impose
consistency of the project schedule (the Y -variables) with the order plan. Constraint (4)
guarantees that all work is done. Finally, equations (5)(7) are the capacity constraints.
The described RCCP is NP-hard. Since the number of order plans to be considered can be
very high, Hans (2001) uses branch-and-price to nd an optimal integer solution.
2.2
Uncertainty modeling
We now consider each tasks work content pbj to be a random variable (rv). The choice
of an appropriate distribution for these rvs is not trivial, and we propose a distributionindependent procedure that only uses expectation E[
pbj ] and variance var[
pbj ]. The work
it =
load W
p
Y
per
period
is
then
also
a
rv,
with
jN
bNj bj bji bjt
it ] =
E[W
jN bNj
Ybjt bji E[
pbj ]
and
it ] =
var[W
2 2
Ybjt
bji var[
pbj ],
jN bNj
so in particular, the expectation is a linear function of the Y -variables and the variance is
quadratic (and we assume the dierent pbj to be independent).
For operational scheduling problems, Sarin et al. (2010) propose to examine the multiobjective problem of minimizing both the expectation as well as the variance of the nominal (deterministic) performance measure of interest. In our case, we will opt for a slightly
altered bi-objective problem, namely the minimization of the expected cost and the minimization of the sum of the workload variances. The rst is obviously the most important
measure, while the second expresses the desire to obtain solutions that are, in some sense,
stable: they will change only little at implementation. This second measure is therefore our
robustness measure. Consequently, our objectives are:
{
,H
, S )]
min
z1 = E[cost(O
Tit it it
(10)
min
z2 = iI t=1 var[Wit ]
it , H
it and Sit to z1 are rvs that are the stochastic counterparts of the
The arguments O
dierent components of non-regular capacity with the same name but without tilde in
the deterministic model. When a schedules quality is evaluated on multiple performance
criteria, in most cases there will be no schedule that achieves the optimal value for all
criteria simultaneously. Consequently, sacrices will need to be made with regard to the
performance for at least one of the conicting criteria; see Hoogeveen (2005) and Tkindt
and Billaut (2006) for an overview of multicriteria scheduling problems. Our interest lies
in the generation of the set of Pareto-optimal solutions for objectives (10).
3
Results
while respecting the minimum-duration constraint (3). Subsequently, the project plan is
modied: we choose a task and modify its time window by increasing or decreasing either
its starting or completion time by one period, while respecting the precedence constraints.
We propose two variants of this procedure: in SA1, = 70 is constant and in SA2 this
number is updated at each iteration according to = + exp(1/) (initial value = 70).
The performance of the algorithm is compared for the deterministic case (no uncertainty)
against the LP-based heuristic of Gademann and Schutten (2005) (GS); note that in this
case only objective z1 is relevant. The results are summarized in Table 1 (all times are in
seconds). We see that both SA1 and SA2 consume more CPU-time but both deliver better
results than GS; SA2 is computationally more expensive than SA1 but also delivers the
best-quality schedules.
References
N. Gademann and M. Schutten (2005). Linear-programming-based heuristics for project capacity
planning. IIE Transactions, 37, 153165.
R.A. Hahn and A.M. Newman (2008). Scheduling United States Coast Guard helicopter deployment and maintenance at Clearwater Air Station, Florida. Computers & Operations Research,
35(6), 18291843.
E.W. Hans (2001). Resource loading by branch-and-price techniques. PhD Thesis, University of
Twente.
H. Hoogeveen (2005). Multicriteria scheduling. European Journal of Operational Research, 167,
592623.
M. Masmoudi (2011). Tactical and operational project planning under uncertainties: application
to helicopter maintenance. PhD Thesis, Institut Suprieur de lAronautique et de lEspace
(ISAE), University of Toulouse.
G. Poncelin, M. Glade, A. Cauvin, D. Dufrne and P. Lyonnet (2006). Design to maintenance cost
by the control of the products environment. Proceedings of Virtual Concept 2006. Available at
http://b-dig.iie.org.mx/BibDig/P07-0610/proceedings/Articles/PLEDM2006/79PLEDM.pdf.
S.H. Sarin, B. Nagarajan and L. Liao (2010). Stochastic Scheduling. Expectation-Variance Analysis
of a Schedule. Cambridge University Press.
A. Sgaslik (1994). Planning German army helicopter maintenance and mission assignment. Masters Thesis, Naval Postgraduate School Monterey.
V. Tkindt and J.C. Billaut (2006). Multicriteria Scheduling Theory, Models and Algorithms.
Springer.
G. Wullink, A.J.R.M. Gademann, E.W. Hans and A. van Harten (2004). A scenario based approach
for exible resource loading under uncertainty. International Journal of Production Research,
42(24), 50795098.
10
20
50
10
20
50
10
20
50
SA1
S12
z1
time
z1
time
z1
time
153,46
31,86
299,18
340,02
90,69
1,2
227,01
801,99
79,61
4,13
26,83
884,94
42,57
67,95
523,2
51,5
224,15
467,9
144,08
83
187,55
373,76
63,41
60,74
214,85
820,97
139,58
186
277,29
722,52
271,41
412,01
787,86
337,2
628,14
1212
101,08
28,21
4,63
86,61
22,16
0
48,28
596,62
0
1053
1631
3998,5
1531,6
2434,2
4562,1
1855,3
3289
6869,1