International Journal of Production Research

ISSN: 0020-7543 (Print) 1366-588X (Online) Journal homepage: http://www.tandfonline.com/loi/tprs20

Production planning in automotive powertrain

plants: a case study

Idris Lalami, Yannick Frein & Jean-Philippe Gayon

To cite this article: Idris Lalami, Yannick Frein & Jean-Philippe Gayon (2017): Production planning
in automotive powertrain plants: a case study, International Journal of Production Research, DOI:
10.1080/00207543.2017.1315192

To link to this article: http://dx.doi.org/10.1080/00207543.2017.1315192

Published online: 12 Apr 2017.

Submit your article to this journal

Article views: 20

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=tprs20

Download by: [University of Newcastle, Australia] Date: 17 April 2017, At: 06:43
International Journal of Production Research, 2017
http://dx.doi.org/10.1080/00207543.2017.1315192

Production planning in automotive powertrain plants: a case study

Idris Lalami, Yannick Frein and Jean-Philippe Gayon*

Univ. Grenoble Alpes, CNRS, Grenoble INP, G-SCOP, F-38000 Grenoble, France
(Received 14 June 2016; accepted 27 March 2017)

Based on a real case study from the automotive industry, this paper deals with production planning in powertrain plants.
We present an overview of the production planning process and propose a mixed integer linear programme to determine
the production quantities of each product over a planning horizon of several days. Then, using real data of an engine
assembly line, we simulate the performance obtained through the proposed model within a rolling horizon planning pro-
cess. We perform multiple tests in order to evaluate the impact of two parameters involved in this process: planning fre-
quency and frozen horizon length. Furthermore, in order to illustrate the value of improving coordination between
engine plants and their customers, we evaluate the impact of the quality of demand information (orders and forecasts).
We analyse the simulation results and provide insights and recommendations in order to achieve a good trade-off
between service level, inventory, and planning stability.
Keywords: production planning; mixed integer linear programme; rolling horizon; planning frequency; frozen horizon;
automotive industry

1. Introduction
As product variety and demand volatility increase, a major challenge for automotive companies is to coordinate effec-
tively the flow of material in their supply chain. In recent years, several studies have been carried out in this field by
automotive companies in order to improve their supply chain processes (see for example Souilah 2008; Garcia-Sabater,
Maheut, and Garcia-Sabater 2011; Volling et al. 2013; Sali and Giard 2015). Our research falls within this scope. It was
conducted together with the automotive company PSA Peugeot Citroën and aims at improving the production planning
performance of powertrain plants. A powertrain plant can be either an engine plant, a gearbox plant, or a chassis part
plant. These plants supply mainly the car assembly plants but also other customers such as spare parts centres and plants
belonging to other automakers. In this paper, we particularly focus on engine plants.
An engine plant is usually composed of multiple workshops. Each workshop contains several production lines dedi-
cated to produce components and products of the same product family (a set of products having common technological
characteristics). The production process inside each workshop is independent from the other workshops. Figure 1 gives
a typical example of an engine production workshop that contains an assembly line and multiple machining lines pro-
ducing the main engine components (engine block, cylinder head, crankshaft, and connecting rods).
In the context of this study, the powertrain plants are implementing a hierarchical planning system that includes
three main processes: the Sales and Operations Planning (S&OP), the Master Production Scheduling (MPS), and the
Production Sequencing. Each hierarchical level has its own characteristics, including the execution frequency, the level
of product aggregation, the length of the planning horizon, and the size of the time buckets in which the planning hori-
zon is divided. The S&OP process is performed each month. It provides an aggregate planning (at product family or
subfamily level) over 18 months divided into months. The MPS process is performed every week or every day and pro-
vides a detailed production (at product reference level) over 16 weeks divided into days and weeks. The sequencing
process is performed every day and provides the detailed production sequence over 10 days.
The remainder of this article is organised as follows. Section 2 presents a literature review on production planning
models and rolling horizon planning systems. Section 3 describes the production planning model that we designed
according to the requirements stated by powertrain plants. A numerical example is provided in order to illustrate how
this planning model works. Section 4 provides a simulation of rolling horizon planning, which is based on a real case
study of an engine assembly line. The impact of planning frequency, frozen horizon length, and quality of demand
information will be analysed. The article ends with the conclusion and further research.

*Corresponding author. Email: [email protected]

© 2017 Informa UK Limited, trading as Taylor & Francis Group

2 I. Lalami et al.

Figure 1. Engine production process.

2. Literature review
This literature review focuses on production planning problems and rolling horizon planning systems.

2.1 Production planning models

The aim of production planning is to determine the production quantity of each product, the time at which such quanti-
ties have to be produced, and often the production facility on which the production must take place. According to
Pochet and Wolsey (2006, 3), it is usually an operational to tactical problem (short to medium-term) where the usual
objective is to meet forecast demand at minimum cost. The authors explain that ‘the goal of production planning is thus
to make planning decisions optimising the trade-off between economic objectives such as cost minimisation or maximi-
sation of contribution to profit and the less tangible objective of customer satisfaction.’
Comelli, Gourgand, and Lemoine (2008) give an overview of mathematical models used to solve tactical planning
problems. This review classifies the models according to five parameters: mono or multilevel, mono or multiproduct,
with or without capacity constraints, constant or variable demand, and small or big time buckets. The authors distin-
guish three types of planning problems. The first problem is sales and operations planning (S&OP). It is about balancing
sales and production level on a midterm horizon. Production capacity is then usually considered as a decision variable.
The second problem is master production scheduling (MPS). It is a more detailed production planning process that pro-
vides dates and production quantities of each item. The third problem is material requirements planning (MRP). It is the
simultaneous planning of final products and their components using bills of materials. According to the authors, the
MPS and MRP are considered as the main problems of tactical planning and are usually modelled as extensions from
two well-known models: the capacitated lot-sizing problem (CLSP), and the multilevel capacitated lot-sizing problem
(MLCLSP). See Drexl and Kimms (1997) for a literature review on lot-sizing models. These problems are usually
solved using mathematical programming (branch and bound, and branch and cut) and approximated methods (heuristics
and metaheuristics).
Many commercial software tools called Advanced Planning Systems (APS) have been developed to solve planning
problems. These software tools use methods of operational research to support the decision-making on a large scope of
supply chain processes such as procurement, production, transport, and demand planning (see Meyr, Wagner, and Rohde
2015). Nevertheless, according to Stadtler (2005, 583), ‘it does not seem wise to find an overall tool adequate for any
possible type of production.’ The author indicates that ‘production planning and detailed scheduling have to be adapted
to the specific needs and conditions.’
In the automotive industry, several papers have studied capacity and production planning. Meyr (2009) discusses
supply chain planning in the German automotive industry and reviews methods of operational research (linear program-
ming, and mixed integer programming) able to support the planning tasks in car assembly plants. Garcia-Sabater,
Maheut, and Garcia-Sabater (2011) present a case study from automotive industry and provide a description of the plan-
ning process of an engine plant. They present mixed integer linear programming (MILP) models that integrate produc-
tion with transport planning and take into account objectives and constraints related to lean manufacturing practices
(such as production levelling). Volling et al. (2013) provide an overview of operational research models used for the
planning of capacities and orders in build-to-order automobile production. Dörmer, Günther, and Gujjula (2013, 2015)
study MPS and sequencing in car assembly lines. They propose a mathematical model and heuristic solution procedures
for the MPS that attempt to minimise the workload variability and anticipate decisions on the sequencing problem.
Wochner, Grunow, Staeblein, and Stolletz (2016) develop a MILP to coordinate sales and ramp-up operations in the
automotive industry.
 International Journal of Production Research 3

Regarding the automotive industry, there are still few papers in the scientific literature giving a comprehensive over-
view of the production planning process and mathematical models. Moreover, automotive powertrain plant case studies
are rare with the exception of Garcia-Sabater, Maheut, and Garcia-Sabater (2011). Most of the papers focus on car
assembly plants.

2.2 Rolling horizon planning

Rolling horizon planning is a common practice in the industry and has received a lot of attention in the literature during
the last decades. Several authors have studied the impact of different planning parameters and rules on the performance
of the supply chain in terms of production cost, service level, and stability. A literature review done by Sahin, Narayanan,
and Robinson (2013) shows that research on rolling horizon planning problems is still incomplete. Despite its importance
and its extensive application in the industry, there is a lack of recent studies, and in particular studies that consider
complex supply chain configurations (e.g. a supply chain involving multiple production levels, multiple products, under
capacity constraints and demand uncertainty).
In a rolling horizon planning context, the production plan is updated periodically by using the most reliable informa-
tion available. According to Tang and Grubbström (2002), there are two basic reasons to update the production plan.
The first one is the shift of planning horizon. The planning horizon is usually composed of a finite number of periods.
As time moves, some periods of the production plan are executed and become past periods, other periods become closer
in time, and new periods must be placed to maintain the planning horizon length. The second reason is uncertainty.
When production planning is performed using demand forecast, it should be regularly reviewed in order to react to fore-
cast errors and take advantage of the latest demand information available. This reactivity contributes to reduce costs and
improves service level.
Dolgui and Prodhon (2007) examine the appropriate rules to deal with uncertainties in both demand (forecast errors)
and production system (machine breakdown, supply delays, etc.). The authors explain that under uncertainties, the pro-
duction plan needs to be updated quite frequently in order to optimise production decisions. However, too frequent
changes provoke disruptions in the supply chain and generate costs related to scheduled orders adjustment (personnel
scheduling, machine loading, etc.). The negative effect due to production planning instability is well known in the litera-
ture. It is usually referred to as ‘nervousness’ (Carlson, Jucker, and Kropp 1979; Blackburn, Kropp, and Millen 1986;
Ho 1989).
Studies dealing with rolling planning horizon often raise two fundamental questions: how often should the produc-
tion plan be updated (refers to planning frequency)? Should all the production plan be updated (refers to freezing part
of the production plan)? The performance indicators usually considered in the literature are: service level (measure of
customer demand satisfaction), production cost (including setup, variable production and inventory costs), and stability
(reflects the impact of the modifications in the production plan).
To ensure a relative planning stability, a common practice is to establish a frozen horizon. Thus, the planned periods
that belong to the frozen horizon are not changed when the production plan is updated (Sridharan, Berry, and Udayab-
hanu 1987; Sridharan and Berry 1990; Zhao and Lee 1993). However, other techniques to reduce instability have been
proposed in the literature such as using a mathematical model that minimises instability when the production plan is cal-
culated (Blackburn, Kropp, and Millen 1986, 1987; Kadipasaoglu and Sridharan 1995; Ho and Carter 1996; Herrera
2011).
Barrett and Laforge (1991) study the impact of planning frequency on the performance of a production system with
multiple products and multiple levels in MRP context. Using a simulation model, the authors evaluate the performance
of different planning frequencies (monthly, twice monthly, weekly, twice weekly, daily, and twice daily). The perfor-
mance is measured through service level, stock level, and the number of changes made in the production plans. The
results show that the rule of thumb that suggests a weekly planning does not provide the best performance trade-off. In
a similar context, Zhao and Lee (1993) examine different parameters involved in a rolling horizon planning system. The
authors evaluate the impact of forecast errors. Under deterministic demand conditions, they show that freezing the entire
planning horizon improves performance in terms of stability, cost, and service level. However, under demand uncer-
tainty, the determination of frozen horizon length is a trade-off between stability, cost, and service level. Regarding plan-
ning horizon length, the authors state that its prolongation improves the performance when demand is deterministic
while it degrades it when demand is uncertain.
Ho and Ireland (1998) study the impact of forecast errors on planning instability. The authors show that the exis-
tence of forecast errors increases instability. To mitigate this effect, they suggest using appropriate criteria when calculat-
ing the production plan. In others words, a proper lot-sizing rule can help to reduce the instability generated by forecast
errors. The same conclusion is stated by Venkataraman and D’Itri (2001) who conducted a similar study.
4 I. Lalami et al.

Xie, Zhao, and Lee (2003) study the case of a production system under capacity constraint and demand uncertainty.
The production system involves a single stage with multiple products. Regarding planning frequency, the authors find
that less frequent planning provides a better performance, both in terms of stability, cost, and quality service. They
explain that frequent changes in the production plan in successive planning cycles do not really help to improve service
level. This conclusion that recommends less frequent planning is also stated by other authors like Sridharan and Berry
(1990), however this does not receive a wide consensus in the literature.
When looking at the findings of numerous articles dealing with rolling horizon planning, there are different recom-
mendations, sometimes contradictory, especially regarding the choice of planning frequency. According to Hozak and
Hill (2009, 4955), ‘empirical research shows that companies that frequently reschedule perform better, despite some the-
oretical research that has discouraged high frequencies.’ The authors explain that providing systems with the timeliest
information possible allows a better planning optimisation and a better responsiveness to unexpected events.
As a conclusion, rolling horizon planning is a classic topic that received a large attention in the literature. Various
articles study the impact of different parameters (length of the planning horizon, length of the frozen horizon planning
frequency, forecast errors, and lot-sizing rules) on performance of the production system. Various production contexts
were considered: single level or multiple levels, single product or multiple products, deterministic or uncertain demand,
with or without resource constraints. However, there is no consensus regarding some conclusions (especially regarding
the choice of planning frequency). Furthermore, there is a lack of recent studies on this subject, and in particular case
studies coming from the industry.

2.3 Contributions
Our first contribution herein is to provide an overview of the production planning problem of automotive powertrain
plants. These plants have specific requirements such as maximising the usage of capacity, satisfying some lean objec-
tives such as production levelling while maintaining low inventories, production by batch, and satisfying minimum and
maximum production quantity constraint. We consider in this paper more specifically the case of engine plants. Based
on the needs stated by the plant planning team, we propose a mathematical model (a mixed integer linear programme)
that aims at optimising four objectives. The first objective is to satisfy the forecast demand. The second one is to reach
safety stock levels. The third is to balance the stock level between all products. The fourth is to level the production of
each product. Two of these four objectives, namely stock balancing and production levelling, are important in many
industrial applications. However, they are rarely studied in the literature within the context of a production planning
model. This model was tested in the case of an engine assembly line and also gearbox assembly and other production
workshops. We think that this model can be extended to other manufacturing contexts with similar requirements (e.g.
units producing mechanical parts having similar constraints and/or having objectives like production levelling and stock
balancing).
Our second contribution is to study how that planning model should be implemented in a rolling horizon planning
context. We analyse in particular the impact of two parameters: planning frequency and length of the frozen horizon. In
addition, we analyse the impact of the quality of demand information (orders and forecasts) on the performance of the
production system. Our simulation is based on the real data coming from an engine assembly line. We analyse the
results and provide insights and recommendations in order to achieve a good tradeoff between service level, inventory,
and planning stability.

3. Planning model
The planning model was developed according to the requirements (objectives and constraints) stated by different power-
train plants. In this model, the production capacity over the planning horizon is considered as an input set by the plan-
ning team.

3.1 Objectives
A work conducted together with the powertrain plant teams has led to a list of four objectives (by decreasing priority):
• The first and main objective is to satisfy the forecast demand.
• The second objective is to reach the safety stock levels.
• The third objective is to balance the stock levels between all the products. This objective aims to balance the risk
of being out of stock between all products. The stock levels are considered as perfectly balanced when the ratio
between the stock level and the safety stock is identical for all products.
 International Journal of Production Research 5

• The last objective is to level the production. The production percentage of each product has to be as stable as pos-
sible over the planning horizon. The main goal of this objective is to achieve a much more stable schedule and
reduce demand variability on upstream supply chain processes.

3.2 Constraints
We have also listed the production constraints that need to be considered when the production planning is made:
• Constraint 1: the production line capacity constraint. For economic reasons (minimising idle time), the total pro-
duction quantity on each production line must equal the capacity set by the planning team.
• Constraint 2: the packing unit constraint. The production quantity of each product has to be a multiple of its pack-
ing unit. For example, the packing unit is 6 for the engines and 8 for the gearboxes.
• Constraint 3: the minimum production quantity constraint for one product.
• Constraint 4: the minimum production quantity constraint for a set of products.
• Constraint 5: the maximum production quantity constraint for a set of products.
• Constraint 6: the production day constraint for a set of products. Some specific products (e.g. products dedicated
to spare part centres) need to be produced on specific days of the planning horizon.
Among all these six constraints, the first two constraints are always considered for all production lines. The other
constraints can be considered or not, depending on the production line characteristics. For example: an engine assembly
line has only the constraints 1 and 2. A gearbox assembly line usually has the constraints 1, 2, and 3. To limit the
length of the paper, only constraint 1 and 2 are considered in the rest of the paper.

3.3 Mathematical formulation

A mixed integer linear programme was developed according to the objectives and constraints stated above.

3.3.1 Notations
Let us consider an engine assembly line producing N different products (references). The goal is to calculate the produc-
tion plan over a planning horizon of H periods (in days). The index i denotes the different products (i = 1 … N) and h
denotes the planning horizon periods (h = 1 … H).
The model parameters (model inputs) are described below:
KðhÞ Production capacity of the engine assembly line in period h
U ðiÞ Packing unit size of product i
Dði; hÞ Customer order or demand forecast of product i in period h
xði; 0Þ Initial stock of product i
SSði; hÞ Safety stock of product i in the period h
TSði; hÞ Target stock of product i in the period h
p Number of consecutive periods considered for production levelling (1 < p <= H)
WBði; hÞ Cost of backorder for product i in period h
WSði; hÞ Cost of not reaching safety stock for product i in period h
WT ði; hÞ Cost of not balancing the stock level for product i in period h
WLði; hÞ Cost of not levelling the production of product i over the periods from h to h + p – 1

The decision variables (model outputs) are the following:

q(i, h) Production quantity of product i in period h
qu(i, h) Number of packing units of product i to produce in period h
x(i, h) Stock level of product i at the end of period h. In case of backorder, this variable is negative
b(i, h) Backorder of product i at the end of the period h
ess(i, h) Difference between the stock level and the safety stock, regarding product i in period h, when the stock level is
under the safety stock
ets(i, h) Difference between the stock level and the target stock regarding product i in period h
m(i, h) Production percentage of product i in period h
mmax(i, h) Maximum value of the production percentage of product i over the periods from h to h + p – 1
mmin(i, h) Minimum value of the production percentage of product i over the periods from h to h + p – 1
6 I. Lalami et al.

3.3.2 Objective function

The objective function is a weighted sum that can be decomposed in four terms:
X
H X
N
WBði; hÞ  bði; hÞ (1)
h¼1 i¼1

X
H X
N
þ WS ði; hÞ  ess ði; hÞ (2)
h¼1 i¼1

X
H X
N
þ WT ði; hÞ  ets ði; hÞ (3)
h¼1 i¼1

X X
Hpþ1 N
þ WLði; hÞ  ðmmax ði; hÞ  mmin ði; hÞÞ (4)
h¼1 i¼1

(1) is related to the objective of satisfying the forecast demand. (2) measures the difference between the stock level
and the corresponding safety stock. This difference is set to zero if the stock level is greater than the safety stock. (3) is
related to the objective of balancing the stock levels between products. Stock levels are perfectly balanced when each
stock level equals the corresponding target stock. (4) penalises the difference between the maximum and the minimum
value of the production percentage of product i over each p consecutive periods. The parameter p is usually set equal to
5 working days (one week). The target stock levels are calculated before the model is launched, so as to allocate the
overall stock level between the different products in proportion to the corresponding safety stocks as shown in Equation
(5). As the overall production equals the capacity, the overall stock level for each period is calculated with (6).

SSði; hÞ XN
8h; 8i; TSði; hÞ ¼ PN  xði; hÞ (5)
i¼1 SSði; hÞ i¼1

X
N X
N X
h h X
X N
8h; xði; hÞ ¼ xði; 0Þ þ Kðh0 Þ  Dði; h0 Þ (6)
i¼1 i¼1 h0 ¼1 h’¼1 i¼1

3.3.3 Constraints
The main constraints of the model are described by the following equations:
8h; 8i; qði; hÞ ¼ qu ði; hÞ  U ðiÞ (7)

8h; 8i; qu ði; hÞ is a non  negative integer (8)

X
N
8h; KðhÞ  e  qði; hÞ  KðhÞ (9)
i¼1

8h; 8i; xði; hÞ ¼ xði; h  1Þ þ qði; hÞ  Dði; hÞ (10)

8h; 8i; bði; hÞ   xði; hÞ (11)

8h; 8i; bði; hÞ  0 (12)

8h; 8i; ess ði; hÞ  SS ði; hÞ  xði; hÞ (13)

8h; 8i; ess ði; hÞ  0 (14)

 International Journal of Production Research 7

8h; 8i; ets ði; hÞ  TS ði; hÞ  xði; hÞ (15)

8h; 8i; ets ði; hÞ  xði; hÞ  TS ði; hÞ (16)

8h; 8i; mði; hÞ  K ðhÞ ¼ 100  qði; hÞ (17)

8h  H  p þ 1; 8h0 ¼ 0. . .p  1; 8i; mmax ði; hÞ  mði; h þ h0 Þ (18)

8h  H  p þ 1; 8h0 ¼ 0. . .p  1; 8i; mmin ði; hÞ  mði; h þ h0 Þ (19)

The production quantity of each product must be a non-negative integer and be a multiple of the packing unit size.
This is expressed with (7) and (8). The total production quantity must equal the capacity of the assembly line. This is
expressed through (9). To avoid infeasibility due to the packing units, a small tolerance ε is considered (the value of ε
is based on the size of the smallest packing unit among products). Equation (10) describes the inventory dynamic.
(11) and (12), together with term (1) of the objective function which is minimised, allow the evaluation of backorder
quantities. The (13) and (14), together with (2) of the objective function, allow the evaluation of the difference between
the net stock and the safety stock. This difference is set to zero if the net stock level is greater than the safety stock.
(15) and (16), together with (3) of the objective function, allow the evaluation of the positive difference between the net
stock level and the target stock.
(17) allows the evaluation of the production percentage of product i in period h. The total production quantity for a
given period is approximated by the production capacity. (18) and (19) together with (4) allow to penalise the produc-
tion unevenness of product i over the periods h to h + p – 1.

3.4 Numerical example

In order to show how this planning model works, we give a numerical example based on theoretical data. We consider
an engine assembly line producing five products (N = 5). We suppose that the planning horizon is equal to two weeks.
As we exclude weekends, the planning horizon is equivalent to 10 working days (H = 10).
We suppose that the production planning is performed on Monday, at the beginning of the day, in order to plan the
quantities to produce over the two coming weeks.
The production capacity over the planning horizon is constant and equals 540 engines per day. For each product, the
packing unit is equal to 6. The demand data are created randomly using a normal distribution with different parameters
in order to have some products with high demand and others with lower demand. These data are given in Table 1.
Initial stocks and safety stocks are also provided in Table 1.
As the first priority is given to satisfy the demand forecast (minimise backorders), the highest weight parameter is
given to this objective. The second objective is to reach safety stock levels. A lower weight is then given to this objec-
tive. The third objective is to reach safety stocks and the last one is production levelling. The chosen weight values are
given in (20).

Table 1. Demand forecast, initial stocks, and safety stocks.

Parameters Product i = 1 Product i = 2 Product i = 3 Product i = 4 Product i = 5

Demand forecasts D (i, 1) 156 168 102 30 48

D (i, 2) 168 114 132 72 54
D (i, 3) 162 102 150 48 48
D (i, 4) 156 114 156 60 54
D (i, 5) 198 108 132 60 60
D (i, 6) 120 156 120 66 54
D (i, 7) 162 144 72 42 84
D (i, 8) 168 198 126 84 48
D (i, 9) 126 162 120 54 60
D (i, 10) 192 162 102 48 66
Initial stocks x (i, 0) 306 282 258 132 102
Safety stocks SS (i, h) 234 204 180 102 78
8 I. Lalami et al.

216
192
168
144
120
96
72
48
24
0
h=1 h=2 h=3 h=4 h=5 h=6 h=7 h=8 h=9 h=10
Product 1 Product 2 Product 3 Product 4 Product 5

Figure 2. Production.

1.65
1.60
1.55
1.50
1.45
1.40
1.35
1.30
1.25
1.20
1.15
h=1 h=2 h=3 h=4 h=5 h=6 h=7 h=8 h=9 h=10
Product 1 Product 2 Product 3 Product 4 Product 5

Figure 3. Stock ratio (Stock level/Safety stock).

8h; 8i; WBði; hÞ ¼ 105 ; WS ði; hÞ ¼ 103 ; WT ði; hÞ ¼ 102 ; WLði; hÞ ¼ 1 (20)
The calculation of the production plan was performed using CPLEX Solver. The obtained solution is illustrated by
two figures. Figure 2 illustrates the production quantities of each product over the planning horizon. Since levelling the
production is considered as having less priority, we can see that the proposed quantities are not stable over the planning
horizon. Note that as the production capacity is constant, it is correct to look at the production quantity instead of the
percentage that it represents.
Figure 3 presents the ratio between the stock level and safety stock, for each product and each period. In this figure,
we can see that all stock levels are positive and are higher than the safety stocks (stock ratio ≥ 1). This shows that the
two most important objectives (satisfying demand forecast and reaching safety stocks) are achieved. Furthermore, for
each period, the ratio between stock and safety stock is almost identical for the all products. This means that the stock
levels are well balanced.

216
192
168
144
120
96
72
48
24
0
h=1 h=2 h=3 h=4 h=5 h=6 h=7 h=8 h=9 h=10

Figure 4. Production.
 International Journal of Production Research 9

1.65
1.60
1.55
1.50
1.45
1.40
1.35
1.30
1.25
1.20
1.15
h=1 h=2 h=3 h=4 h=5 h=6 h=7 h=8 h=9 h=10
Product 1 Product 2 Product 3 Product 4 Product 5

Figure 5. Stock ratio (Stock level/Safety stock).

If the weight parameters of the objective function are modified, a different production plan can be calculated. For
example, we modified the weight parameters in order to give a higher priority to production levelling. The obtained
results are illustrated in Figures 4 and 5. Figure 4 shows that the production quantity becomes more stable than in
Figure 2. In the other hand, Figure 5 shows that the stocks become less balanced than in Figure 3.
Through this numerical example, we showed how the planning model works. This model was tested within the com-
pany and was able to properly fulfil the requirements stated by the engine plant, both in terms of solution relevance and
computation times.

3.5 Computation times

In order to evaluate the computation times, several tests have been performed with different values of N and H. Based
on the previous example, we have designed 49 instances (see Table 2). Initial stocks, safety stocks, and demand infor-
mation are defined by duplicating the settings of the numerical example we presented previously. The production
capacity is then defined in proportion to the number of products. The weights involved in the objective function are
time-dependent. For each objective and each period h, the weight values are defined by multiplying a constant coeffi-
cient by the function (0.95)h–1. This is done in order to give higher priority to earlier periods of the planning horizon.
The 49 test instances were solved using IBM ILOG CPLEX Optimisation Studio (Version: 12.6.0.0) and using a
personal computer (DELL VOSTRO V131 Intel® CoreTM i3 2.1GHz, 2GB RAM). The direct formulation of the model
in CPLEX did not provide good computation time results for large-sized instances (i.e. when N × H > 2000). However,
we obtained better results thanks to a reformulation we made on the model. As the demand information for each product
is always a multiple of the product packing unit, we add a constraint stating that the variation of stock level, for each
product, has to be a multiple of the corresponding packing unit. This constraint is provided by (21) and (22).
8h; 8i; xði; hÞ  xði; h  1Þ ¼ Dði; hÞ  U ðiÞ (21)

Table 2. Computation time.

Number of periods (H)

H=5 H = 10 H = 20 H = 40 H = 60 H = 80 H = 100

Number of products (N) N = 5 0.10a 0.10a 0.20a 0.30a 0.50a 0.60a 0.60a
N = 10 0.10a 0.10a 0.10a 0.40a 0.90a 1.90a 1.90a
N = 20 0.00a 0.10a 0.40a 4.70a 8.70a 13.00a 17.20a
N = 40 0.10a 0.30a 5.50a 95.90a 0.05%b 0.17%b 0.17%b
N = 60 2.20a 5.40a 20.90a 0.37%b 0.34%b 0.35%b 0.35%b
N = 80 0.40a 3.40a 21.40a 0.24%b 0.26%b 0.31%b 0.36%b
N = 100 0.60a 4.60a 89.20a 0.43%b 0.43%b 0.39%b 0.81%b
a
Computation time needed to obtain the optimal solution (in seconds).
b
Relative gap to the lower bound when the optimal solution is not obtained within 120 s.
10 I. Lalami et al.

8h; 8i; Dði; hÞ is a an integer (22)

The obtained results are presented in Table 2. Note that the computation time limit was set to 120 s.
Table 2 shows that for 34 instances, the optimal solution is obtained within 120 s. The average computation time
for these instances is 9 s. For 15 instances, the solution we obtain after 120 s is not guaranteed to be optimal. Neverthe-
less, the relative gap to optimum is lower than 0.81%.

4. Rolling horizon planning simulation

4.1 Simulation procedure
We consider an engine assembly line that produces N engine references. Every day, the customers provide demand fore-
casts that the engine assembly line uses to perform its production planning (in this case, more than 97% of the demand
comes from car assembly plants). The planning horizon length set by the engine assembly line is H days. We denote by
T the simulation time length and by t a specific day in the simulation time (t = 0 … T).
The first calculation of the production plan (initialisation) is performed at t = 0. The time between two successive
planning updates determines the planning periodicity and is denoted by δ. As time moves, a part of the production plan
is executed and new information becomes available (demand forecast, stock levels, and capacity). At t = δ, a new calcu-
lation of the production plan is made. This is the second iteration in the simulation. This procedure is then repeated at
t = 2δ, t = 3δ, t = 4δ, …, until the end of the simulation time.
In addition, the engine plant may establish a frozen horizon. If the frozen horizon length is X days, then the first X
days of the planning horizon cannot be modified when the production plan is updated. In other terms, when the produc-
tion plan is calculated, the first δ + X periods of the production plan cannot be modified in the future.
At each iteration, the production plan is recalculated using the planning model presented in Section 3. This recalcu-
lation is made at the end of day t in order to plan the production from day t + 1 to t + H. The main input parameters
involved in this recalculation are demand information dt(i, h), production capacity Kt(h), and stock levels xt(i). dt(i, h)
is the demand forecast at t regarding the day t + h and the product i, Kt(h) is the capacity of the engine assembly line
in day t + h, and xt(i) denotes the stock level of the product i at the end of day t. The calculation output is qt(i, h) which
is the production quantity planned at t regarding the day t + h and the product i.
As the engine plant may consider a frozen horizon of X days, we have added an additional constraint (23) within
the planning model which states that for each day that belongs to the frozen horizon, the quantity calculated at t must
equal the quantity previously calculated at t–δ.

8i; 8t [ d; if h  X then qt ði; hÞ ¼ qtd ði; h þ dÞ (23)

4.2 Input data

We gathered real data from the engine assembly line so as to simulate the planning process over T = 60 working days
(12 weeks) and N = 34 references. The average overall production is around 500 engines per day. The production plan-
ning horizon length is H = 10 working days (2 weeks). In Table 3, we provide indicators about demand variability for
each product i.
Every day, the customers send demand forecasts that indicate the quantity of each product that must be shipped by
the engine plant over the next 10 working days. The customers usually confirm the real demand by sending orders three
days before the shipping date. Thus, the demand information related to the three first days of the planning horizon is
almost certain (forecast errors exist but are very minor). However, starting from the fourth day of the horizon, the
demand is uncertain and forecast errors are more significant. In Figure 6, we indicate, for each period h of the planning
horizon, the forecast error expressed as a percentage of actual demand. Thus, we can clearly see that the forecast errors
are very low for h = 1, h = 2 and h = 3 (0%, 2%, and 8%, respectively) and become much higher starting from h = 4.
In our simulations, we assume that, for each week, the production capacity is equal to the average demand of this
week. In other words, within each week, we produce exactly the total demand of this week. Accordingly, the overall net
stock at the beginning of each week is always the same and is equal to the initial stock. This is why the overall initial
stock is a key parameter in our simulations. This assumption of a weekly demand equal to the weekly capacity is pes-
simistic, as it assumes no production flexibility. In practice, we have observed that each line has some flexibility on pro-
duction.
 International Journal of Production Research 11

Table 3. Demand variability.

Average Standard deviation Percentage of periods with Maximum demand Minimum demand
Product i demand of demand demand > 0 (%) quantity quantity

i=1 73.2 21.1 100 132 24

i=2 61.4 19.8 100 108 18
i=3 40.0 20.1 93 96 0
i=4 32.6 17.8 100 90 6
i=5 27.3 17.4 92 78 0
i=6 21.6 10.9 97 60 0
i=7 23.5 31.1 53 120 0
i=8 21.3 15.2 95 60 0
i=9 19.6 19.5 93 102 0
i = 10 17.8 7.7 97 36 0
i = 11 20.8 32.5 51 138 0
i = 12 13.8 10.9 83 36 0
i = 13 16.4 7.4 100 42 6
i = 14 16.7 27.7 37 114 0
i = 15 12.0 8.1 92 36 0
i = 16 12.1 7.6 92 30 0
i = 17 11.2 11.2 80 72 0
i = 18 12.0 9.3 83 36 0
i = 19 9.6 5.8 85 18 0
i = 20 7.6 8.0 63 30 0
i = 21 7.6 4.0 90 18 0
i = 22 6.7 10.0 42 42 0
i = 23 3.8 6.4 34 24 0
i = 24 2.4 3.7 36 18 0
i = 25 2.3 3.0 39 6 0
i = 26 1.3 3.0 19 12 0
i = 27 1.2 2.7 19 12 0
i = 28 0.8 2.6 10 12 0
i = 29 1.1 7.8 3 60 0
i = 30 0.7 2.0 12 6 0
i = 31 0.4 1.5 7 6 0
i = 32 0.4 1.5 7 6 0
i = 33 0.4 1.9 5 12 0
i = 34 0.1 0.8 2 6 0

50%
40% 40% 41% 41% 42% 42%
40%

30% 26%

20%
8%
10%
0% 2%
0%
h=1 h=2 h=3 h=4 h=5 h=6 h=7 h=8 h=9 h=10

Figure 6. Forecast accuracy over the planning horizon.

Multiple scenarios were simulated. A scenario is defined by setting a planning periodicity (δ) and a frozen horizon
length (X). In order to analyse the impact of δ and X, we simulated 8 different scenarios: scenario 1 (δ = 1, X = 0), sce-
nario 2 (δ = 1, X = 1), scenario 3 (δ = 1, X = 2), scenario 4 (δ = 1, X = 3), scenario 5 (δ = 1, X = 4), scenario 6 (δ = 1,
X = 5), scenario 7 (δ = 5, X = 0), and scenario 8 (δ = 5, X = 5). In these scenarios, we are testing two planning frequen-
cies that are the most likely to be used in practice: daily planning (scenarios from 1 to 6) and weekly planning (scenar-
ios 7 and 8). In case of a daily planning, we test different frozen horizon length (from X = 0 to X = 5 days). In case of
a weekly planning, we test X = 0 and X = 5 days.
12 I. Lalami et al.

Thus, for each scenario, we test different initial stock levels so as to determine the amount of stock required to
achieve a given service level. Ten initial stock values are tested: 0, 198, 402, 600, 798, 1002, 1200, 1398, 1602, and
1800. These values were defined by considering an increase by step of 200 engines, and taking the closest multiple of
6, since the stock level must be multiple of 6 (size of the engine packing unit). In terms of days of coverage, 200 engi-
nes represent approximately 0.4 days. We first start with the zero initial stock value. Then, we gradually increase the
value of the initial stock (198, 402, 600, …). We stop when all the values are tested or when the obtained service level
is greater than or equal to 99%.
In Section 3, we explained that the planning model considers that the overall production must be almost equal to the
capacity set by the planning team. In this situation, the simulated performance of the engine assembly line is highly
influenced by the capacity level and the initial stocks. For example, if the initial stock and the capacity are too low,
many backorders will occur and the obtained service level will be very poor. Therefore, to improve that service level,
the initial stock level must be higher or the production capacity must be adjusted. In the simulation model presented
herein, we consider that the capacity is fixed. Then, we perform multiple simulations with different initial stock levels
so as to determine the stock level which is required to achieve a desired service level (this approach will be discussed
in the conclusion of this article). In other words, we consider that the overall initial stock available at the beginning of
the simulation (t = 0) is set by the user (e.g. 300 engines). As we also need to determine how this overall stock is allo-
cated to the different products (determine the initial stock x0(i) for each product i), we introduced a modification within
the planning model so as, at the beginning of the simulation (t = 0), the allocation of the overall initial stock is
optimised automatically according to the planning requirements (3.1).

4.3 Performance indicators

For a given scenario and initial stock value, we simulate the planning process over T = 60 days. Then, we measure the
performance indicators at the end of the simulation. As is usually done in the literature, we consider three performance
indicators: stock level indicator (as a cost indicator), service level (as a customer satisfaction indicator), and a planning
stability indicator.
As a stock level indicator, we consider the value of the overall initial stock set by the user. This stock value is
expressed in terms of days of coverage. We calculate the ratio between the initial stock level and the average daily
demand. This calculation is illustrated by (24) where dRt(i) denotes the actual demand of the product i in day t.
PN
x0 ð i Þ
PT i¼1PN R (24)
 i¼1 dt ðiÞ
1
T t¼1

The service level indicator reflects the percentage of demand satisfaction. The calculation of this indicator is illus-
trated by (25) where bt(i) denotes the backorder quantity of the product i in day t.
PT PN
bt ð i Þ
1  PTt¼1 PNi¼1 (25)
i¼1 dt ðiÞ
R
t¼1

In order to measure planning stability, we compare the quantity that was really produced by the assembly line to the
quantity that was planned h days before. Then, we calculate the percentage of the quantity really produced in compar-
ison to the quantity planned h days before. For a given value of h, this calculation is given by (26) where qRt(i) denotes
the real quantity produced of product i in day t. As an example, for a single product, if the quantity produced in day t
is qRt(i) = 90 and the quantity planned h days before was qt–h(i,h) = 100, then the planning stability indicator is equal to
90%. If the quantity produced in day t was qRt(i) = 100, then the indicator would be equal to 100%.
PT PN  R 
t¼1 i¼1 Min qt ðiÞ; qth ði; hÞ
PT PN R (26)
t¼1 i¼1 qt ðiÞ

This planning stability indicator can be calculated for different values of h. In our case study, the management was
particularly interested in the planning stability regarding the five first days of the horizon. Therefore, we measure this
indicator for h = 1 … 5, and then, we calculate the average in order to obtain a synthetic indicator. This calculation is
described in (27).
h¼5 PT PN
 R 
1 X i¼1 Min qt ðiÞ; qth ði; hÞ
 t¼1
PT PN R (27)
i¼1 qt ðiÞ
5 h¼1 t¼1
 International Journal of Production Research 13

As the different scenarios and performance indicators were introduced, we will now present the simulation results.
The simulation model has been developed in CPLEX (to calculate the production plans) and Excel (to manage input
and output data).

4.4 Impact of planning frequency and frozen horizon

For each scenario, we made multiple simulations by testing different initial stock levels in order to determine the mini-
mum stock level needed to obtain a service level greater than or equal to 99%. This target service level was determined
together with the engine plant management.
In Table 4, we present the simulation results. For each scenario, this table provides the stock level needed to obtain
the desired service level (expressed in days of coverage), the value of the service level, and the value of the planning
stability indicator that we previously described in (27).
In order to analyse the impact of the frozen horizon length, let us look at the scenarios with daily planning (δ = 1).
The results show that the increase in the frozen horizon length improves planning stability. However, this will require
higher stock levels if we desire to maintain a good service level (≥99%). For example, to switch from a frozen horizon
of 1 day (X = 1) to 5 days (X = 5), the stock level passes from 0.4 to 2.4 days of coverage.
To analyse the impact of planning periodicity (δ), we compare the scenarios 7 and 8 (weekly planning) to the
scenarios 1 and 6 (daily planning), respectively. The scenarios 7 and 1 do not consider a frozen horizon (X = 0), while
the scenarios 8 and 6 consider a frozen horizon of 5 days (X = 5). The simulation results show that increasing the plan-
ning periodicity from daily (δ = 1) to weekly planning (δ = 5) improves stability. This is observable when comparing
the scenario 1 to scenario 7. For the scenarios 6 and 8, the planning stability is identical and equals 100%. This is
because the frozen horizon length in these scenarios is 5 days (X = 5) and we are measuring the planning stability indi-
cator over the first 5 days of the planning horizon. This being said, the increase in planning periodicity leads to higher
stock levels. The needed stock to achieve the desired service level passes, in case of a X = 0, from 0.4 days (scenario 1)
to 1.2 days (scenario 7), and from 2.4 days (Scenario 6) to 3.2 days (scenario 8) in case of X = 5.
As a conclusion, the simulation results showed that the increase in frozen horizon length (X) and planning periodic-
ity (δ) improves the performance in terms of planning stability but worsens it in terms of stock needed to achieve the
target service level. Based on the real data of an engine assembly, the simulations we made helped to quantify the per-
formance obtained in different scenarios. The obtained results illustrate the trade-off between stability, stock, and service
levels and provide some insights to help the decision makers in choosing the most suitable scenario (planning periodic-
ity and frozen horizon length).
Having a good planning stability is particularly important for the procurement department because it reduces the
uncertainty regarding the quantities of components required by the engine assembly line, and thereby reduces upstream
safety stocks and associated costs. In our case study, the procurement department requested to freeze the production plan
of the engine assembly line over 5 days (X = 5), which makes possible to supply just in time the exact amount of the
required components. If we put this request as the first priority, the scenarios 6 and 8 may be considered. However, our
simulation results suggest that the scenario 6 (daily planning with a frozen horizon of 5 days) leads to better production
reactivity and improves the performance in terms of stock and service level. This being said, in this scenario, the stock
needed to reach the desired service level is 2.4 days of coverage, which is relatively high. To achieve a better stock per-
formance, a trade-off solution would be to consider a frozen horizon of 3 days (scenario 4) instead of 5 days (scenario

Table 4. Impact of planning frequency and frozen horizon.

Planning Frozen Obtained

periodicity horizon Initial stock level needed to service level Planning stability (over the 5 first
Scenario (δ) length (X) obtain a service level ≥ 99% (%) days of the planning horizon) (%)

Scenario 1 1 (every 0 0.4 99.9 85

Scenario 2 day) 1 0.4 99.7 88
Scenario 3 2 0.8 100.0 91
Scenario 4 3 1.2 99.0 96
Scenario 5 4 2.0 99.2 100
Scenario 6 5 2.4 99.3 100
Scenario 7 5 (every 0 1.2 99.2 96
Scenario 8 week) 5 3.2 99.4 100
14 I. Lalami et al.

6). Table 4 shows that the scenario 4 (δ = 1, X = 3) requires half less amount stock (1.2 instead of 2.4 days). Moreover,
this scenario provides a very good planning stability (96%). Therefore, we recommend the implementation of scenario 4
which does not meet strictly the request of X = 5 but leads to a very good performance trade-off between service level
(99.9%), stock (1.2 days), and planning stability over the 5 first days of the horizon (96%).

4.5 Impact of the quality of demand information

Now, we investigate the impact of the quality of demand information (forecast) that is used by the engine assembly line
to perform its production planning. The simulation results presented previously (Table 4) were made with the real
demand data (real forecast errors). Now, we reduce the amount of forecast errors and analyse the impact on the perfor-
mance. We will focus on three scenarios: scenario 4, scenario 6, and scenario 8.
In order to simulate the improvement of forecast accuracy, we multiply the real forecast errors by a coefficient (1–α)
where α represents the percentage of improvement of the forecast accuracy. Four different values of α are tested: 20, 40,
60, and 80%. Table 5 gives the simulation results (stock and service level) for each scenario and each value of α.
As we can expect, Table 5 shows that improving forecast accuracy leads to stock reduction and service level
improvement. For example, if forecast accuracy is improved by 20%, we can see that a significant stock reduction is
achieved in scenario 6 and 8, and service level is improved from 99.0 to 99.7% in scenario 4. These results indicate the
prerequisites to establish in terms of forecast accuracy if one desires to implement a specific scenario. For example, if
the target is to implement the scenario 6 with a stock level not exceeding 1.6 days, then the forecast accuracy needs to
be improved by at least 40%. Finally, let us remark that the scenario 4 requires lower stock levels in comparison to sce-
narios 6 and 8.
At the beginning of this section, we explained that the engine plant customers usually send their orders three days
before the shipping date. Thus, demand information over the three first days of the planning horizon is almost certain
since forecast errors are very minor. So, we also wanted to simulate a situation in which the customers would send their
orders four or five days (instead of three days) before the shipping date. For each scenario (4, 6, and 8), we performed
simulations by supposing that the real demand is perfectly known over the λ first days of the planning horizon. Three
values of λ were tested: λ = 3 (which is very close to the actual situation), λ = 4, and λ = 5. The obtained results are pre-
sented in Table 6.
The results show that passing from λ = 3 to λ = 4 improves the stock performance. In fact, the stock is reduced from
3.2 to 2.8 days in scenario 8, from 2.4 to 2.0 days in the scenario 6, and from 1.2 to 0.4 days in scenario 4. When pass-
ing from λ = 4 to λ = 5, the service level is improved by 0.2% for scenarios 4 and 8, and by 0.6% for the scenario 6.
Finally, note that scenario 4 is systematically better than the scenarios 6 and 8. In particular, an implementation of sce-
nario 4 (δ = 1, X = 3) with a λ = 4 days would lead to a service level of 99.6% with very low stock (0.4 days). Remem-
ber that this scenario also provides a very good stability (96%).

Table 5. Impact of forecast accuracy improvement.

Without forecast improvement With forecast improvement

Frozen Percentage of Initial stock level
Planning horizon Initial stock level Obtained forecast accuracy needed to obtain a Obtained
periodicity length needed to obtain a service improvement (α) service level ≥ service
Scenario (δ) (X) service level ≥ 99% level (%) (%) 99% level (%)

Scenario 4 1 (every 3 1.2 99.0 20 1.2 99.7

day) 40 1.2 99.9
60 0.8 99.8
80 0.4 99.4
Scenario 6 5 2.4 99.3 20 2.0 99.9
40 1.6 99.8
60 1.2 99.2
80 0.8 99.2
Scenario 8 5 (every 5 3.2 99.4 20 2.8 99.9
week) 40 2.0 99.4
60 1.2 99.0
80 0.8 99.3
 International Journal of Production Research 15

Table 6. Impact of sooner receiving of customer orders.

Planning Frozen Obtained

periodicity horizon length Number of days with real Initial stock level needed to service level
Scenario (δ) (X) demand known (λ) obtain a service level ≥ 99% (%)

Scenario 4 1 (every day) 3 3 days 1.2 99.4

4 days 0.4 99.6
5 days 0.4 99.8
Scenario 6 5 3 days 2.4 99.5
4 days 2.0 99.0
5 days 2.0 99.6
Scenario 8 5 (every 5 3 days 3.2 99.5
week) 4 days 2.8 99.0
5 days 2.8 99.2

5. Conclusion and further research

In this paper, we have developed a planning model, based on the requirements provided by the engine plant teams of
PSA Peugeot Citroën. This model was tested within the company and was able to properly fulfil the requirements stated
by the engine plant, both in terms of solution relevance and computation times.
Using the planning model, we developed a simulation model in order to evaluate the performance obtained in rolling
horizon planning context. Increasing planning periodicity and frozen horizon length improve the performance in terms
of stability but worsen it in terms of stock and service level. In this case study, considering the stability requested by
the engine procurement department, we recommended a daily planning with a frozen horizon of 3 (scenario 4). In this
scenario, we obtain a service level equal to 99% with a stock level of 1.2 days. Moreover, the planning stability over
the 5 first days of the horizon is 96%. When analysing the impact of demand information, we showed that performance
of this scenario could be significantly improved if the customer orders were known 1 day sooner. In fact, in this situa-
tion, we could achieve a service level equal to 99.6% with a stock level of 0.4 days. More generally, the analysis of the
impact of forecast errors indicates the prerequisites that must be established in order to achieve the desired performance.
The assumption to produce at maximum capacity is not respected in practice and it would be interesting to relax this
assumption. In this case, inventory costs should be included in the objective function. Another avenue for research
would be to consider production capacity as a decision variable and not as an input of our model. Consequently, a better
simulation approach would be to use the actual initial stock of the engine plant and to adjust the production capacity
over the simulation time.

Disclosure statement
No potential conflict of interest was reported by the authors.

References

Barrett, Robert T., and R. Lawrence LaForge. 1991. “A Study of Replanning Frequencies in a Material Requirements Planning
System.” Computers & Operations Research 18 (6): 569–578. doi:10.1016/0305-0548(91)90062-V.
Blackburn, Joseph D., Dean H. Kropp, and Robert A. Millen. 1986. “A Comparison of Strategies to Dampen Nervousness in MRP
Systems.” Management Science 32 (4): 413–429. doi:10.1287/mnsc.32.4.413.
Blackburn, Joseph D., Dean H. Kropp, and Robert A. Millen. 1987. “Alternative Approaches to Schedule Instability: A Comparative
Analysis.” International Journal of Production Research 25 (12): 1739–1749.
Carlson, Robert C., James V. Jucker, and Dean H. Kropp. 1979. “Less Nervous MRP Systems: A Dynamic Economic Lot-sizing
Approach.” Management Science 25 (8): 754–761.
Comelli, Michael, Michel Gourgand, and David Lemoine. 2008. “A Review of Tactical Planning Models.” Journal of Systems Science
and Systems Engineering 17 (2): 204–229. doi:10.1007/s11518-008-5076-8.
Dolgui, Alexandre, and Caroline Prodhon. 2007. “Supply Planning under Uncertainties in MRP Environments: A State of the Art.”
Annual Reviews in Control 31 (2): 269–279.
Dörmer, Jan, Hans-Otto Günther, and Rico Gujjula. 2013. “Master Production Scheduling and Sequencing at Mixed-model Assembly
Lines in the Automotive Industry.” Flexible Services and Manufacturing Journal 27 (1): 1–29. doi:10.1007/s10696-013-9173-8.
16 I. Lalami et al.

Dörmer, J., H. O. Günther, and R. Gujjula. 2015. “Master Production Scheduling and Sequencing at Mixed-model Assembly Lines in
the Automotive Industry.” Flexible Services and Manufacturing Journal 27 (1): 1–29.
Drexl, A., and A. Kimms. 1997. “Lot Sizing and Scheduling – Survey and Extensions.” European Journal of Operational Research
99 (2): 221–235. doi:10.1016/S0377-2217(97)00030-1.
Garcia-Sabater, Jose P., Julien Maheut, and Julio J. Garcia-Sabater. 2011. “A Two-stage Sequential Planning Scheme for Integrated
Operations Planning and Scheduling System Using MILP: The Case of an Engine Assembler.” Flexible Services and Manufac-
turing Journal 24 (2): 171–209. doi:10.1007/s10696-011-9126-z.
Herrera, Carlos. 2011. “Cadre generique de planification logistique dans un contexte de decisions centralisees et distribuees.” Thèse
De Doctorat, Université Henri Poincaré – Nancy I. https://tel.archives-ouvertes.fr/tel-00639761/document.
Ho, Chrwan-Jyh. 1989. “Evaluating the Impact of Operating Environments on MRP System Nervousness.” International Journal of
Production Research 27 (7): 1115–1135. doi:10.1080/00207548908942611.
Ho, Chrwan-Jyh, and P. L. Carter. 1996. “An Investigation of Alternative Dampening Procedures to Cope with MRP System
Nervousness.” International Journal of Production Research 34 (1): 137–156. doi:10.1080/00207549608904895.
Ho, Chrwan-Jyh, and T. C. Ireland. 1998. “Correlating MRP System Nervousness with Forecast Errors.” International Journal of Pro-
duction Research 36 (8): 2285–2299. doi:10.1080/002075498192904.
Hozak, Kurt, and James A. Hill. 2009. “Issues and Opportunities regarding Replanning and Rescheduling Frequencies.” International
Journal of Production Research 47 (18): 4955–4970. doi:10.1080/00207540802047106.
Kadipasaoglu, Sukran N., and V. Sridharan. 1995. “Alternative Approaches for Reducing Schedule Instability in Multistage Manufac-
turing under Demand Uncertainty.” Journal of Operations Management 13 (3): 193–211. doi:10.1016/0272-6963(95)00023-L.
Meyr, Herbert. 2009. “Supply Chain Planning in the German Automotive Industry.” In Supply Chain Planning, 1–23. Springer Berlin
Heidelberg. http://link.springer.com.gaelnomade.ujf-grenoble.fr/chapter/10.1007/978-3-540-93775-3_13.
Meyr, Herbert, Michael Wagner, and Jens Rohde. 2015. “Structure of Advanced Planning Systems.” In Supply Chain Management
and Advanced Planning, edited by Hartmut Stadtler, Christoph Kilger, and Herbert Meyr, 99–106. Springer Texts in Business
and Economics. Springer Berlin Heidelberg. http://link.springer.com.gaelnomade.ujf-grenoble.fr/chapter/10.1007/978-3-642-
55309-7_5.
Pochet, Yves, and Laurence A. Wolsey. 2006. Production Planning by Mixed Integer Programming. New York: Springer Science &
Business Media.
Sahin, Funda, Arunachalam Narayanan, and E. Powell Robinson. 2013. “Rolling Horizon Planning in Supply Chains: Review, Impli-
cations and Directions for Future Research.” International Journal of Production Research 51 (18): 5413–5436. doi:10.1080/
00207543.2013.775523.
Sali, Mustapha, and Vincent Giard. 2015. “Monitoring the Production of a Supply Chain with a Revisited MRP Approach.” Produc-
tion Planning & Control 26 (10): 769–785. doi:10.1080/09537287.2014.983579.
Souilah, Sihem. 2008. “Reengineering du pilotage des flux dans une relation client/fournisseur. Application au cas de l’industrie auto-
mobile.” Thèse de doctorat, Ecole Centrale Paris. https://tel.archives-ouvertes.fr/tel-00375995/document.
Sridharan, V., and William L. Berry. 1990. “Freezing the Master Production Schedule under Demand Uncertainty.” Decision Sciences
21 (1): 97–120. doi:10.1111/j.1540-5915.1990.tb00319.x.
Sridharan, V., William L. Berry, and V. Udayabhanu. 1987. “Freezing the Master Production Schedule under Rolling Planning
Horizons.” Management Science 33 (9): 1137–1149. doi:10.1287/mnsc.33.9.1137.
Stadtler, Hartmut. 2005. “Supply Chain Management and Advanced Planning – Basics, Overview and Challenges.” European Journal
of Operational Research 163 (3): 575–588. doi:10.1016/j.ejor.2004.03.001.
Tang, Ou, and Robert W. Grubbström. 2002. “Planning and Replanning the Master Production Schedule under Demand Uncertainty.”
International Journal of Production Economics 78 (3): 323–334. doi:10.1016/S0925-5273(00)00100-6.
Venkataraman, Ray, and Michael P. D’Itri. 2001. “Rolling Horizon Master Production Schedule Performance: A Policy Analysis.”
Production Planning & Control 12 (7): 669–679. doi:10.1080/09537280010016774.
Volling, Thomas, Andreas Matzke, Martin Grunewald, and Thomas S. Spengler. 2013. “Planning of Capacities and Orders in Build-
to-order Automobile Production: A Review.” European Journal of Operational Research 224 (2): 240–260. doi:10.1016/
j.ejor.2012.07.034.
Wochner, S., M. Grunow, T. Staeblein, and R. Stolletz. 2016. “Planning for Ramp-ups and New Product Introductions in the Automo-
tive Industry: Extending Sales and Operations Planning.” International Journal of Production Economics 182: 372–383.
Xie, Jinxing, Xiande Zhao, and T. S. Lee. 2003. “Freezing the Master Production Schedule under Single Resource Constraint and
Demand Uncertainty.” International Journal of Production Economics 83 (1): 65–84. doi:10.1016/S0925-5273(02)00262-1.
Zhao, Xiande, and T. S. Lee. 1993. “Freezing the Master Production Schedule for Material Requirements Planning Systems under
Demand Uncertainty.” Journal of Operations Management 11 (2): 185–205. doi:10.1016/0272-6963(93)90022-H.