Available online at www.sciencedirect.

com

ScienceDirect
Procedia CIRP 41 (2016) 585 – 590

48th CIRP Conference on MANUFACTURING SYSTEMS - CIRP CMS 2015

The benefit of integrating production and transport scheduling

Jens Ehma,*, Michael Freitagb
a
BIBA - Bremer Institut für Produktion und Logistik GmbH at the University of Bremen, Hochschulring 20, 28359 Bremen, Germany
b
University of Bremen, Hochschulring 20, 28359 Bremen, Germany
* Corresponding author. Tel.: +49-421-218 50112; fax: +49-421-218-50003. E-mail address: [email protected]

Abstract

The production of many goods comprises several production steps at different places along a supply chain. Therefore, transportation of
intermediate products can be necessary between consecutive production steps. The efficient coordination of resources requires a detailed
scheduling. In current industry processes, the scheduling of production and transport is usually performed separately. As a consequence, the
optimization of each schedule can only be based on local criteria, which do not necessarily result in optimal solutions from the overall supply
chain perspective. Consequently, the integrated scheduling of production and transport processes is regarded as a promising approach for the
improvement of scheduling systems. Due to the combinatorial nature of both subproblems, the integrated problem is also very hard to solve.
Therefore, the research focuses on the development of powerful methods that are able to compute solutions in reasonable time. However, to
show the importance of an integrated view on production and transport scheduling, it is also necessary to quantify its benefit. This paper aims
to close this research gap to accelerate the consideration of integrated scheduling within advanced planning systems. First, it is shown how the
production and transport scheduling problems can be formalized in terms of mixed-integer programs (MIP). Then, it is highlighted how these
two problems can be integrated and which impact the integration has on the constraints of the optimization problem. All MIP-formulations are
implemented and used for a numerical study comprising test scenarios of different sizes. The benefit of an integration of the problems is
determined by investigating the impact on several performance indicators such as lead-times, storage times or delays.
©
© 2015 The Authors.
2015 The Authors. Published
Published by
by Elsevier
Elsevier B.V.
B.V.This is an open access article under the CC BY-NC-ND license
Peer-review under responsibility of the Scientific Committee of 48th CIRP Conference on MANUFACTURING SYSTEMS - CIRP CMS
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
2015.
Peer-review under responsibility of the scientific committee of 48th CIRP Conference on MANUFACTURING SYSTEMS - CIRP CMS 2015
Keywords: Integrated scheduling; production and transport scheduling; advanced planning systems; supply chains

1. Introduction avoid these costs, production and logistics processes should be

synchronized first in the single companies and then along the
The demands and requirements for industrial production whole supply chain [3]. The following sections show the
and logistics systems changed during the last years. challenges from production as well as logistics perspective
Companies face global competitive environments as well as and explain the idea of an integrated scheduling.
customer, supplier and distribution networks on a global scale
[1]. Therefore, whole supply chains as well as the single 1.1. Scheduling of production processes
partners within the chain are forced to a consequent reduction
of costs and lead-times. A widely used strategy is Just-In- The organization of production processes requires a
Time production in order to guarantee the allocation of the detailed scheduling as an important part of the short term
right materials at the right time in the right quality and planning [4], which means the exact allocation of available
quantity at the right place [2]. This strategy requires a distinct resources (i. e. machines, tools and working staff) over a
adherence to due dates, since a late delivery to one partner in certain planning horizon to perform a certain amount of jobs
the supply chain might cause delays in the subsequent [5]. A solution of the scheduling problem is a detailed
production steps and can potentially affect the whole supply schedule that determines for each job at what time it is
chain. This induces unnecessary costs due to idle times, processed with which resources. The scheduling problem
penalties for late delivery or high stock levels. In order to depends on the type of production. A typical setup is the so-

2212-8271 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the scientific committee of 48th CIRP Conference on MANUFACTURING SYSTEMS - CIRP CMS 2015
doi:10.1016/j.procir.2015.12.143
586 Jens Ehm and Michael Freitag / Procedia CIRP 41 (2016) 585 – 590

called job shop, where machines for certain operations (e. g. the specific task as it is displayed in Figure 1 (based on [12]).
grinding, cutting or drilling) are assembled in groups that are Long-term decisions concerning the structure of the whole
decoupled by intermediate storages and where some machines supply chain, such as the choice of locations for production
might replace each other (parallel machines) [6]. Most sites and transport modes can be comprised in one central
scheduling problems, especially the ones arising from planning module that determines the framework for all other
practical applications, belong to the class of NP-hard modules. The shorter the planning horizon the more detailed
problems [7], which means that they are typically hard to decisions are necessary. In the mid-term, for example
solve. Therefore, heuristic solution methods are often used, forecasted demands are coordinated with available capacities
which do not necessarily compute exact solutions, but which of the single production sites. The detailed scheduling of
are often able to compute near-optimal solutions in relatively production and transport belongs to the short-term tasks and is
short computation time. However, this paper suggests a carried out individually by each site, where the information
mixed-integer programming model for computing exact about current system status is available (e. g. machine
solutions in order to compare solutions without uncertainty breakdowns, missing staff due to illness or delayed supply of
due to the solution method (Section 2). needed materials). This structure is called hierarchical
planning and represents a model that leads to solvable
1.2. Scheduling of transport processes planning tasks for all parts of the supply chain.

The efficient organization of transport processes also Procurement Production Distribution Sales
requires a detailed scheduling. In this case, a schedule
allocates jobs to available transport vehicles and determines long-
Strategic Network Design
transport routes so that given due dates are fulfilled. In many term

applications, the transport vehicles (e. g. trucks) process

mid-
several jobs at the same time and deliver goods to several term
Master Planning
Demand
customers during one tour. The scheduling of these processes Planning
comprises, in analogy to production, decisions about the Production Distribution
allocation of jobs to resources (vehicles instead of machines) Purchasing & Planning Planning
Material
as well as sequences of processing (delivery instead of Requirements Demand
production steps). Similar to production, there are also short- Planning Transport Fulfillment &
Scheduling
term Planning ATP
different common problem formulations depending on the
specific transport scenario. The most famous is probably the
Fig. 1. Integration highlighted in the supply chain planning matrix
Travelling Salesman Problem (TSP), which describes the task
of finding the shortest tour through a given set of cities where
In addition to the hierarchy in vertical direction, Figure 1
the tour comprises each city exactly once [8, 9]. If it is in
shows a division in horizontal direction along the tasks of a
addition necessary to split the amount of jobs into tours for
supply chain. The highlighted boxes represent the detailed
several vehicles it is called Vehicle Routing Problem (VRP).
scheduling of production as well as transport, which are two
Several other modifications are possible according to the
different modules in the APS structure. However, both
specific application, such as the VRP with time windows or
scheduling problems have a similar time horizon, are both on
under capacity restrictions [10]. The TSP is known to be an
the production site level and are in interdependency.
NP-hard problem. Since the VRP comprises a TSP for each
Deficiencies of transport can have direct impact on the
truck it also belongs to this complexity class and the same
production, e. g. in terms of lateness in the material supply.
holds for its variations with additional restrictions. Therefore,
On the other hand, production can handicap the efficiency of
the solution techniques for this kind of scheduling problem
transport, e. g. if a consolidation of deliveries is not possible
also range from exact methods (such as the mixed-integer
due to unfavorable production sequences. Thus, a good
programming approach from Section 2) to simple construction
coordination of production and transport scheduling prevents
heuristics or more advanced heuristic methods like genetic
waste of resources and permits a reliable forecasting and
algorithms [11]. Transport processes related to the production
adherence to due dates, which is essential for an efficient
of goods are not restricted to the delivery to customers or
management of supply chains.
subsequent supply chain partners but can also be necessary at
As described in the previous sections, the optimization
the production site, e. g. between two consecutive production
problems of production and transport scheduling are quite
steps or between machines and material stocks.
related and can be solved with similar methods. If both
scheduling problems are not regarded separately but in one
1.3. Integrated scheduling
single optimization problem it is called an integrated
scheduling problem and enables a perfect coordination of
The planning of production and transport processes is done
processes by the computation of globally optimal solutions
in many companies by so-called Advanced Planning Systems
[13]. The literature on integrated scheduling focuses on the
(APS), which are available as standard software tools from
development of heuristic, which enable the solution of large
different providers. Their strengths but also their limitations
problem instances under the assumption that integrated
can be seen in [12]. APS consist of several software modules
scheduling gives better solutions than sequential scheduling.
for different planning tasks dependent on the time horizon of
Even if this can be well justified considering Figure 1, the
 Jens Ehm and Michael Freitag / Procedia CIRP 41 (2016) 585 – 590 587

benefit of integrated scheduling should be investigated characteristics like the number of cities and the distances
quantitatively. Therefore, Section 2 presents a mixed-integer between them, the number of available transport vehicles and
formulation for the integrated scheduling problem and shows their capacities, fix and variable costs of tours as well as the
specific constraints regarding production as well as transport. required transport capacity of each job can be chosen freely.
In Section 3, this model is used for a numerical study on the
impact of integrated scheduling for several performance 2.2. Mixed-integer formulation for the integrated scheduling
indicators such as lead-times, storage times or delays.
Nomenclature
2. Mixed-integer scheduling
Binary variables
A solution of a scheduling problem is an “optimal” ‫ݔ‬௝,௠ Production of job j on machine m
allocation of resources to fulfill a given set of tasks. In this ‫ݕ‬௝,௝ ᇲ ,௠ Job j before j’ on machine m
section, it is shown how the integrated scheduling problem ‫ݏ‬௩,௥ Execution of tour r by vehicle v
can be expressed as a mixed-integer problem (MIP) and how ‫ݑ‬௝,௥ Job j on tour r
the term of optimality can be formalized. In order to compare ‫ݓ‬௜,௜ ᇲ ,,௥ City i’ subsequent to i on tour r
different key performance indicators (KPIs) like lead times, ‫ݍ‬௥ Execution of tour r
the amount of storage times or delays, all KPIs that shall be Continuous variables ( • 0 )
included in the model are expressed in terms of costs. For ௦௧,௣
‫ݐ‬௝,௟ Start of production of job j on level l
example each job can have an individual cost factor for each ௘௡ௗ,௣
machine, which expresses all costs that occur for each time ‫ݐ‬௝ Time when job j is ready for transport
unit that the job occupies the machine. The delay of jobs also ‫ݐ‬௥௦௧ Start time of tour r
௔௥௥
has a cost factor that penalizes each time unit of delayed ‫ݐ‬௥,௜ Arrival time of tour r at city i
delivery to a customer. In some applications the adherence to ‫ݐ‬௩,௥ Execution of tour r by vehicle v
due dates might be crucial. In that case, the related cost factor ‫ݐ‬௥ௗ௨௥ Duration of tour r
ௗ௟௬
has to be weighted high in relation to the others. The setting ‫ݐ‬௝ Delay of job j
of cost factors is part of modelling a specific scenario. ௘௟௬
‫ݐ‬௝ Earliness of job j
‫ݐ‬௝௦௧௢ Storage time of j between production and transport
2.1. Flow-shop scenario with subsequent transport
Parameters
M, L, J Numbers of machines, levels and Jobs
The integrated scheduling problems studied in this paper
I, V, R Numbers of cities, vehicles and max. no. of tours
are based on quite general models for production and ௣
‫ܥ‬௝,௠ Cost for production of job j on machine m
transport that apply to a wide range of practical applications. ௘௟௬ ௗ௟௬
The production is modelled as a flow-shop with a variable ‫ܥ‬௝ , ‫ܥ‬௝ Cost factors for earliness and delay of job j
number of subsequent production levels, where the number of ‫ ܥ‬௟௧ , ‫ ܥ‬௦௧௢ Cost factors for lead time and storage
parallel machines on each level is also variable. Each job can ‫ ܥ‬௙௜௫ , ‫ ܥ‬௩௔௥ Fix and variable costs of transportation
potentially be processed by each machine, the only restriction
is that each job has to be processed by exactly one machine of
each level. An example setting with three production levels is In the following model, an optimal schedule should
displayed in Figure 2. minimize the costs for the production and transport of all jobs,
Hamburg for the lead time in production, or storage between production
Bremen and transport and for unpunctuality (earliness as well as
Berlin
Production Production Production
Münster
Hannover delay). With the above nomenclature, this can be expressed as
level 1 level 2 level 3 Bielefeld
Machine 1 Machine 1 Essen Dortmund Leipzig the minimization problem (1).
Düsseldorf Göttingen
Machine 1
Köln Dresden
Machine 2 Machine 2
Bonn ௃ ெ ௃
Machine 2 Frankfurt
௣ ௘௡ௗ,௣ ௦௧,௣
Machine 3 Machine 3
Nürnberg
min ෍ ෍ ‫ܥ‬௝,௠ ‫ݔ‬௝,௠ + ‫ ܥ‬෍ ‫ݐ‬௝
௟௧
െ ‫ݐ‬௝,ଵ
Mannheim
Karlsruhe ௝ୀଵ ௠ୀଵ ௝ୀଵ
Stuttgart ௃ ோ
München
+ ෍ ‫ ܥ‬௦௧௢ ‫ݐ‬௝௦௧௢ + ෍(‫ ܥ‬௙௜௫ + ‫ ܥ‬௩௔௥ ‫ݐ‬௥ௗ௨௥ )‫ݍ‬௥
Fig. 2. Example scenario for integrated production and transport scheduling (1)
௝ୀଵ ௥ୀଵ
௃
The number of jobs to be scheduled, their supply and due ௗ௟௬ ௗ௟௬ ௘௟௬ ௘௟௬
+ ෍ ‫ܥ‬௝ ‫ݐ‬௝ + ‫ܥ‬௝ ‫ݐ‬௝
dates can be freely chosen as well as all parameters regarding
௝ୀଵ
production costs and times, which are individual for each job.
The machine setting represents a production facility located at under constraints (2) - (27).
a city that is embedded in a transport network. The example in In some of the following equations, a constant number ‫ ܭ‬is
Figure 2 shows an approximation of the German highway used to activate or deactivate constraints under certain
network as available transport connections between cities, conditions. This constant is considered to be “large”, which
with the production facility located quite centrally in the city means that if ‫ ܭ‬is added to only one side of an inequality, the
of Göttingen. For modelling a specific scenario, the transport absolute value of that side becomes larger than the other side.
588 Jens Ehm and Michael Freitag / Procedia CIRP 41 (2016) 585 – 590

If it is not denoted differently, the following constraints

௃ ூ ூ
represent all possible combinations of the indices that are
෍ ‫ݑ‬௝,௥ + ෍ ෍ ‫ݓ‬௜,௜ ᇲ ,௥ ൑ ‫ ܭ‬ή ‫ݍ‬௥ (11)
contained and not specified by a summation index, e. g. ௝ୀଵ ௜ୀଵ ௜ ᇲ ୀଵ
equation (2) represents ‫ ܬ‬constraints ‫ ׊‬1 ൑ ݆ ൑ ‫ ܬ‬and
inequality (3) represents (‫ ܮ‬െ 1) ή ‫ ܬ‬constraints (‫ ׊‬1 ൑ ݆ ൑ ‫ܬ‬ The sum of required capacities ‫݌ܽܥ‬௝ of jobs on a tour
௥௘௤

and ‫ ׊‬2 ൑ ݈ ൑ ‫)ܮ‬. must not excess the available capacity of the vehicle ‫݌ܽܥ‬௩௔௩ .
Equation (2) makes sure that each job is allocated on
exactly one machine of each level. Here, ݅௟ଵ denotes the index ௃
௥௘௤
௏
of the first machine and ܵ௟ the number of machines on level ݈. ෍ ‫݌ܽܥ‬௝ ‫ݑ‬௝,௥ െ ෍ ‫݌ܽܥ‬௩௔௩ ‫ݏ‬௩,௥ ൑ 0 (12)
௝ୀଵ ௩ୀଵ

௜೗భ ାௌ೗
෍ ‫ݔ‬௝,௠ = 1 (2) Tours have to be closed, which means that the number of
௠ୀ௜೗భ arrivals at a city has to be equal to the number of leavings.

ூ ூ
The production on a level can only start if the job was
෍ᇲ ‫ݓ‬௜,௜ ᇲ ,௥ െ ෍ ᇲᇲ ܹ௜ ᇲᇲ ,௜,௥ = 0 (13)
finished on the level before as defined by constraint (3), ௜ ୀଵ ௜ ୀଵ
௣ ௜ᇱஷ௜ ௜ ᇲᇲ ஷ௜
where ܶ௝,௠ determines the production time of ݆ on ݉.
భ Equation (14) makes sure that each tour contains the depot.
௜೗షభ ାௌ೗షభ
௣ ௦௧,௣ ௦௧,௣
෍ ܶ௝,௠ ‫ݔ‬௝,௠ + ‫ݐ‬௝,௟ିଵ െ ‫ݐ‬௝,௟ ൑0 with l ൒ 2 (3)
ூ
భ
௠ୀ௜೗షభ
െ෍ ‫ݓ‬ଵ,௜,௥ ൑ െ‫ݍ‬௥ (14)
௜ୀଵ
Whenever two jobs are allocated on the same machine, a
sequence has to be defined. Each city can have only exactly one successive city due to
constraint (15).
‫ݔ‬௝,௠ + ‫ݔ‬௝ ᇲ ,௠ െ ‫ݕ‬௝,௝ ᇲ ,௠ െ ‫ݕ‬௝ ᇲ ,௝,௠ ൑ 1 with ݆ < ݆Ԣ (4)
ூ
෍ᇲ ‫ݓ‬௜,௜ ᇲ ,௥ ൑ 1 (15)
This sequence of jobs has to be unique. ௜ ୀଵ
௜ᇱஷ௜

‫ݕ‬௝,௝ ᇲ ,௠ + ‫ݕ‬௝ ᇲ ,௝,௠ ൑ 1 with ݆ < ݆Ԣ (5) The destination cities of all jobs of a tour have to be part of
the tour, which is assured by constraint (16). The binary value
Constraint (6) makes sure that no sequence is determined ‫ݐݏ݁ܦ‬௝,௜ determines whether city ݅ is destination of job ݆.
for the production of jobs on different machines.
ூ
‫ݔ‬௝,௠ + ‫ݔ‬௝ ᇲ ,௠ െ 2 ή ‫ݕ‬௝,௝ ᇲ ,௠ െ 2 ή ‫ݕ‬௝ ᇲ ,௝,௠ ൒ 0 with ݆ < ݆Ԣ (6) ෍ᇲ ‫ݓ‬௜ ᇲ ,௜,௥ ൒ ‫ݐݏ݁ܦ‬௝,௜ ή ‫ݑ‬௝,௥ with i ൒ 2 (16)
௜ ୀଵ
௜ ᇲ ஷ௜

Each machine can only process one job at a time.

According to constraint (17), a tour can only start if all its
௦௧,௣ ௣ ௦௧,௣ jobs are ready after production.
െ‫ ܭ‬ή ൫1 െ ‫ݕ‬௝,௝ ᇲ ,௠ ൯ + ‫ݐ‬௝,௟೘ + ܶ௝,௠ ൑ ‫ݐ‬௝ ᇲ ,௟ with ݆ ് ݆Ԣ (7)
೘
௘௡ௗ,௣
‫ݐ‬௥௦௧ ൒ ‫ݐ‬௝ െ ‫ ܭ‬ή (1 െ ‫ݑ‬௝,௥ ) (17)
Constraint (8) determines the time when each job j is ready
for transport.
The start time for all tour numbers that are not performed
௜భಽ ାௌಽ
shall be equal to zero as well as the arrival times at cities that
௦௧,௣ ௣ ௘௡ௗ,௣ are not part of a tour. This is determined by constraints (18)
‫ݐ‬௝,௅ + ෍ ܶ௝,௠ ‫ݔ‬௝,௠ ൑ ‫ݐ‬௝ (8)
௠ୀ௜భಽ and (19), respectively.

Each performed tour is allocated to exactly one vehicle by ‫ݐ‬௥௦௧ ൑ ‫ ܭ‬ή ‫ݍ‬௥ (18)
constraint (9). ூ
௔௥௥
‫ܭ‬ή෍ᇲ ‫ݓ‬௜ ᇲ ,௜,௥ ൒ ‫ݐ‬௥,௜ᇱ (19)
௜ ୀଵ
௏ ௜ ᇲ ஷ௜
෍ ‫ݏ‬௩,௥ = ‫ݍ‬௥ (9)
௩ୀଵ The earliest arrival of a tour at a city is the arrival at the
previous city plus the travel time between both cities
Each job goes on exactly one tour.
(constraint (20)). If the previous city is the depot, it is the
ோ starting time of the tour plus the travel time (constraint (21)).
෍ ‫ݑ‬௝,௥ = 1 (10)
௥ୀଵ with i ൒ 2
௔௥௥
‫ݐ‬௥,௜ + ‫ܦ‬௜ ᇲ ,௜ െ ‫ ܭ‬ή ൫2 െ ‫ݓ‬௜ ᇲ ,௜,௥ െ ‫ݍ‬௥ ൯ ൑ ‫ݐ‬௥,௜
௔௥௥
(20)
i ് ‹Ԣ
Constraint (11) makes sure that only tours that are with i ൒ 2
performed can have allocated jobs and sequences of jobs. ‫ݐ‬௥௦௧ + ‫ܦ‬ଵ,௜ െ ‫ ܭ‬ή ൫2 െ ‫ݓ‬ଵ,௜,௥ െ ‫ݍ‬௥ ൯ ൑ ‫ݐ‬௥,௜
௔௥௥
(21)
i ് ‹Ԣ
 Jens Ehm and Michael Freitag / Procedia CIRP 41 (2016) 585 – 590 589

Constraints (22) - (24) determine the duration of the tours x Storage: a time span between the end of production for a
and the delay respectively the earliness of jobs as the job and the start of its transport tour leads to costs for
difference of the actual arrival times to the due dates ܶ௝ௗௗ . storing the products. If the sum of storage costs over all
jobs is low, this indicates a smooth material flow and a
௔௥௥
‫ݐ‬௥,ଵ െ ‫ݐ‬௥௦௧ ൑ ‫ݐ‬௥ௗ௨௥ (22) good concurrence of production and transport.
ௗ௟௬ x Unpunctuality: the adherence to due dates is crucial in
‫ݐ‬௥,௜ െ ܶ௝ௗௗ െ ‫ ܭ‬ή (1 െ ‫ݐݏ݁ܦ‬௝,௜ ή ‫ݑ‬௝,௥ ) ൑ ‫ݐ‬௝
௔௥௥
(23)
௘௟௬ order to fulfill customer demands. Late as well as early
ܶ௝ௗௗ െ ‫ݐ‬௥,௜
௔௥௥
െ ‫ ܭ‬ή (1 െ ‫ݐݏ݁ܦ‬௝,௜ ή ‫ݑ‬௝,௥ ) ൑ ‫ݐ‬௝ (24)
deliveries can have negative consequences for the
customer, so the total unpunctuality should be minimized,
Each vehicle can only perform one tour at a time, which is which is defined as the sum of all deviations from the
assured by constraint (25). given due dates.
x Lead-time: during the time that a job spends in the
௔௥௥
‫ݐ‬௥,ଵ െ ‫ ܭ‬ή (2 െ ‫ݏ‬௩,௥ െ ‫ݏ‬௩,௥ ᇲ ) ൑ ‫ݐ‬௥௦௧ᇲ with r < ”Ԣ (25) production system, the resources used for performing the
job are tied capital, which should be as low as possible.
Constraint (26) computes the storage times of jobs. x Total cost: The fundamental performance indicator that
decides about the quality of a schedule is the sum of all
௘௡ௗ,௣
‫ݐ‬௥௦௧ െ ‫ݐ‬௝ െ ‫ ܭ‬ή (1 െ ‫ݑ‬௝,௥ ) ൑ ‫ݐ‬௝௦௧௢ (26) costs that this schedule induces. It is the objective value of
the minimization problem.
Finally, the processing of a job can only start after its
supply date ܶ௝௦ௗ . 3.2. Results

ܶ௝௦ௗ ൑ ‫ݐ‬௝,ଵ
௦௧
(27) The generated test scenarios were solved as integrated
scheduling problems as well as sequentially. Attention has to
3. Numerical study on the benefit of integrated scheduling be paid to the sequential case. First, two separate schedules
for production and transport are produced, which have to be
The integrated as well as the sequential scheduling models composed to an overall schedule that can be compared with
were implemented and solved for several test scenarios using the integrated one. Second, it is clear that the given due date
the Gurobi Optimizer Version 5.6.3. In the following, it is of a job should not serve as the due date for the separate
first explained how the test scenarios were generated and the production scheduling, since in that case there would be no
results of sequential and integrated scheduling are compared. time left for transportation. Therefore, an intermediate due
date is introduced for each job, which represents a guideline
3.1. Scenario generation and evaluation criteria for finishing the production process. For each job, this date
has to range between the supply date and the due date. The
For the numerical analysis a production scenario with three later it is, the more time is reserved for production at the
levels of three, two and three machines is considered as it is expense of the transport and vice versa for an early
shown in Figure 2. The transport scenario was reduced to intermediate due date. This means that the sequential solution
three cities. The results are generated by solving several test of a scheduling scenario depends on the chosen intermediate
instances of six jobs. In order to produce comparable test due dates, so a numerical evaluation requires several runs for
instances, the number of jobs and all parameters related to the each scenario. For the coupling of the sequential production
production and transport facilities are fixed, like the time and and transport scheduling, the actual production finishing times
costs needed for a production step on each machine, the cost are considered as supply dates for transport.
factors for unpunctuality, storage, transport and lead-time, the Figure 3 shows for a single scenario the four performance
number and capacities of available vehicles as well as the indicators of the sequential solution relative to the integrated
distances between the cities. The test scenarios differ in the solution (the storage costs were normed to the largest
parameters related to the set of jobs to scheduled, i. e. supply occurring value, since the integrated solution was zero). The
and due dates of the jobs as well as the required capacity of horizontal axis represents the choice of intermediate due
each job on a transport vehicle. The dates were computed dates, where e. g. the value 60 means that 60 % of the
randomly within a given planning horizon, which was chosen available time span for each job are reserved for production
1.5 times longer than the shortest possible time for a job to be and 40 % for transport. The storage costs shown in Figure 3
delivered (i. e. if the job can choose the fastest machine on follow an expected trend: early intermediate due dates cause
each level, has the nearest city as destination and is delivered that all jobs are ready for transport quickly and thus have to
directly without waiting times). The required transport be stored so that the delivery does not cause a massive
capacities of each job were computed randomly between 10 % earliness at the destination. In this range, the lead-times of the
and 50 % of the largest available truck capacity. production processes are even shorter than in the integrated
Four performance indicators were selected for the schedule, but the advantage is outweighed by higher storage
evaluation of the quality of a computed schedule. They can be and unpunctuality costs, as the total costs show. Late
derived from the cost factors of the objective functions: intermediate due dates provoke that too much time is reserved
for production and many jobs are delivered late, which means
the unpunctuality costs grow and storage need is low.
590 Jens Ehm and Michael Freitag / Procedia CIRP 41 (2016) 585 – 590

Fig. 5. Box plot of the total costs for sequential scheduling, where the costs of
the integrated schedule were normed to 1

The models were implemented and applied to six test

instances of a flow-shop scenario with subsequent road
transport. It could be seen that the sequential solutions depend
on the coupling between production and transport. For the
specific test scenario it turned out that if a sequential
Fig. 3. KPIs of sequential solutions relative to integrated solution for different
scheduling is desired, 80 % of the time span between supply
intermediate due dates (percentage of time-span from supply to due date) and due date of a job should be reserved for production.
However, it was shown that for the test scenario an integration
Figure 4 shows the storage and unpunctuality costs for all of scheduling always resulted in better solutions and saved
test scenarios in a box plot, which means that each box between 10 % and 30 % of the total costs in most cases.
represents information about the distribution of the specific
performance indicator for all test scenarios with a fixed Acknowledgements
strategy for choosing intermediate due dates. This figure The authors gratefully acknowledge the financial support
confirms the correlations that were described by means of by the German Research Foundation (DFG), grant SCHO
Figure 3. The optimization criterion that represents the quality 540/30-1.
of a schedule is the sum of all costs. Figure 5 shows a box plot
of the total costs for all six test scenarios relative to the total References
cost value of the respective integrated schedule, which are
normed to one. It can be seen that the solution quality depends [1] Busch, A.; Dangelmaier, W.: Integriertes Supply Chain Management:
Theorie und Praxis effektiver unternehmensübergreifender
on the choice of intermediate due dates and the best solutions Geschäftsprozesse. 2. Aufl. Wiesbaden : Gabler, 2004
were obtained when 80 % of the available time span of a job [2] Dickmann, P.: Schlanker Materialfluss: mit Lean Production, Kanban und
was reserved for production. However, regardless of the Innovationen. 2. Berlin, Heidelberg : Springer-Verlag (VDI-Buch), 2009
specific choice of intermediate due dates, the sequential [3] Ullrich, C. A.: Integrated machine scheduling and vehicle routing with
solution never represented the optimal solution. In fact, time windows. In: European Journal of Operational Research 227, 2012;
1: 152–165
sequential scheduling produced more than 10 % extra costs in [4] Quadt, D.; Kuhn, H.: A taxonomy of flexible flow line scheduling
most cases compared to integrated scheduling. procedures. In: European Journal of Operational Research 178, 2007; 3:
686–698
[5] Jungwattanakit, J.; Reodecha, M.; Chaovalitwongse, P.; Werner, F.:
Algorithms for flexible flow shop problems with unrelated parallel
machines, setup times, and dual criteria. In: The International Journal of
Advanced Manufacturing Technology 37, 2008; 3–4: 354–370
[6] Pfohl, H.-C.: Logistiksysteme: betriebswirtschaftliche Grundlagen. 8., neu
bearb. und aktualisierte Aufl. Berlin [u.a.]: Springer, 2010
[7] Gondek, V.: Hybrid flow shop scheduling mit verschiedenen
Restriktionen: Heuristische Lösung und LP-basierte untere Schranken,
Universität Duisburg-Essen, Diss., 2011
[8] Papadimitriou, C. H.: Computational complexity. Reprinted with corr.
Reading, Mass. [u.a.] : Addison-Wesley, 1995
Fig. 4. Box plot for storage and unpunctuality costs of all six test scenarios [9] Morton, T. E. ; Pentico, D. W.: Heuristic scheduling systems : with
applications to production systems and project management. New York,
NY [u.a.] : Wiley, 1993
4. Conclusions
[10] Golden, B.; Raghavan, S.; Wasil, E. A.: The vehicle routing problem:
latest advances and new challenges. New York, NY : Springer, 2008
This paper showed a mixed-integer approach for the [11] Clarke, G.; Wright, J. W.: Scheduling of Vehicles from a Central Depot
optimization of schedules for production and transport to a Number of Delivery Points. In: Operations Research 12, 1964; 4:
scenarios. Since in practice, production and transport are often 568–581
[12] Stadtler, H.: Supply Chain Management und Advanced Planning:
subsequent processes, it is desired to find schedules from an
Konzepte, Modelle und Software. Berlin [u.a.] : Springer, 2010
integrated perspective on both processes. It was shown in the [13] Scholz-Reiter, B.; Frazzon, E. M. ; Makuschewitz, T.: Integrating
paper, how optimal schedules can be computed sequentially Manufacturing and Logistic Systems along Global Supply Chains. In:
as well as by an integrated optimization approach. CIRP Journal of Manufacturing Science and Technology, 2010; 2: 216–
223