1 s2.0 S0007850614001139 Main

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

CIRP Annals - Manufacturing Technology 63 (2014) 457–460

Contents lists available at ScienceDirect

CIRP Annals - Manufacturing Technology


jou rnal homep age : ht t p: // ees .e lse vi er . com /ci r p/ def a ult . asp

Capacity management for assembly systems with dedicated and


reconfigurable resources
Dávid Gyulai a,b, Botond Kádár (2)a,*, András Kovács a, László Monostori (1)a,b
a
Fraunhofer Project Center for Production Management and Informatics, Institute for Computer Science and Control, Hungarian Academy of Sciences, Budapest,
Hungary
b
Department of Manufacturing Science and Technology, Budapest University of Technology and Economics, Budapest, Hungary

A R T I C L E I N F O A B S T R A C T

Keywords: Managing changes and disturbances resulted by fluctuating order streams and diverse product portfolios
Production planning
requires efficient capacity management decisions and production planning strategies. High volume
Reconfiguration
products can be produced cost efficiently on dedicated assembly lines, while the assembly of low runners
Optimization
is more efficient on reconfigurable lines. In the paper a hierarchical planning decision workflow is
introduced to assign the products to dedicated and reconfigurable lines, and to optimize the system
configuration and the production plan of the reconfigurable system in an integrated way. The proposed
solution is demonstrated through the results of an industrial case study.
ß 2014 CIRP.

1. Introduction Though both reconfigurable systems and capacity planning


methods have broad literatures, relatively few papers can be found
Strategic production planning and long-term resource alloca- on capacity planning for reconfigurable systems. In [7], Ceryan and
tion are complex tasks in a sense that they have to deal not only Koren introduce a capacity planning method with two products to
with the available production capacities and order volumes, but determine the cost-optimal investment strategies considering
also need to consider future stages of the product’s life cycle. In dedicated and reconfigurable resources. Based on the dynamically
case a company has a diverse product portfolio, conventional changing nature of the order stream, the capacity and system
production systems are not always flexible enough to adapt to the configuration planning process is often formulated as a Markovian
changes that occur and not all product life cycle phases can be Decision Problem that can be solved by dynamic programming or
handled cost-efficiently [1]. learning algorithms [8–12]. Hon and Xu propose a simulation-
The efficient management of variety in production is one of the based method to optimize the system structure of a reconfigurable
most crucial points in today’s industry [2]. Depending on the order system based on the different stages of the products’ life cycle [13].
volumes and the diversity of the product portfolio, different Arafa and ElMaraghy introduce a method for investigating the
solutions exist to ensure cost-efficient production. Dedicated manufacturing strategy of a firm that faces dynamic market
systems are designed around a certain product/product family, to environment, by calculating the volume flexibility of the produc-
produce large volumes with low product variety [3]. In contrast, tion [14]. In [15], a performance analysis method is formulated that
reconfigurable systems are capable of producing a set of different is based on the universal generating function technique, and can be
products with high variety by their changeable functionality, applied for modular reconfigurable systems. None of the above
scalable capacity, and modularity [2,4–6]. The main drawbacks of papers considers an integrated method that is aimed at solving the
the reconfigurable systems are the lower throughput and the line assignment problem as well as production planning of the
complex planning and control methods required. reconfigurable assembly lines.
In the paper the authors propose a new capacity planning In a preceding publication the authors of the paper proposed a
method that integrates hierarchically the long-term resource method for replacing a part of the dedicated assembly lines with a
investments with short-term modular configurations and optimal modular, reconfigurable assembly system on the base of standard-
production planning. The system applying the proposed approach ized assembly processes [16]. The reconfigurable lines are
supports the decisions on the allocation of products to one of the composed of standard, mobile workstations, which are configured
above types of production systems, as well as, simultaneously sequentially according to the successive assembly operations. The
performs the optimization of the configuration and resource scalability of the system is guaranteed by adding and removing
assignment of the modular reconfigurable system. standard, adjustable resources which allow human operators to
perform certain assembly processes (e.g., a screwing machine with
adjustable torque ranges). The changeable functionality of a
* Corresponding author. system like this provides cost-efficient production for low-volume

http://dx.doi.org/10.1016/j.cirp.2014.03.110
0007-8506/ß 2014 CIRP.
458 D. Gyulai et al. / CIRP Annals - Manufacturing Technology 63 (2014) 457–460

products. The number and the types of the reconfigurable lines are The following assumptions are made. Order volumes and
calculated based on the order volumes, and the throughput is forecasts are available for the given time horizon. All products can
estimated by using automatically generated discrete-event simu- be assembled in either a reconfigurable or a dedicated line. It is
lation models. The industrial environment considered in this paper assumed that the capacity of a single line is sufficient to assemble
is similar, but the focus is on the cost-efficient capacity- and the product in the desired volume, and therefore, the option of
production planning of the reconfigurable lines. dividing the order volume between different production modes
In the following section a detailed formulation of the problem in can be ignored. Although certain products exist in different
question is given. The proposed solution approach is presented in variants, these variants are always produced together, and
Section 3, followed by the experimental results based on the therefore, they can be considered as single products on this level
previously stated real industrial dataset. of the production planning hierarchy. Machine prices and the costs
of human operators are constant over time.
2. Problem formulation
2.2. The underlying cost model
2.1. The capacity management problem
The line assignment problem can be seen as subdividing the set
In an assembly system with dedicated and reconfigurable lines, of products, P, into products assembled on the dedicated lines, D,
the key decision within capacity management is allocating each on the reconfigurable lines, R, and products outsourced, OS.
product to a dedicated or a reconfigurable line or, alternatively, For products p 2 D or p 2 OS, the production costs can be
outsourcing it to a supplier, while minimizing the total production assigned directly to individual products, and denoted by a
cost (Fig. 1). parameter Cp. In case p is assembled on a product-specific
dedicated line, the production cost Cp can be computed as the
sum of the investment cost (zero if a dedicated line for p already
exists), a high fixed cost, and a volume dependent operation cost.
Analogously, for an outsourced product p, Cp is composed of a small
fixed cost and a relatively high unit cost.
In contrast, the cost related to the reconfigurable lines depends
on the actual product mix and the production plan adopted, and
cannot be directly divided among individual products. With a
given optimization model for production planning, this cost can be
described as a general, non-linear function of the production
volumes, resource requirements, and further parameters of the
products assembled on the reconfigurable lines. Therefore, the
overall production cost incurred in the reconfigurable system is
captured by a function CR, and it incorporates the investment costs
and the operation costs related to those lines. A key challenge in
the line assignment problem is computing, as well as predicting
this cost for an arbitrary set of products R.
Fig. 1. Illustration of the capacity management problem.

2.3. The decision workflow

Since in the reconfigurable system the production cost depends The proposed hierarchical workflow for integrated line assign-
on the product mix in question and the production plan adopted, line ment and production planning for the reconfigurable system is
assignment and production planning of the reconfigurable system depicted in Fig. 2. Integration is established via feedback from
are strongly related. Therefore, the proposed method focuses on production planning to line assignment, in the form of multivariate
solving the line assignment and capacity planning problems (Fig. 2). linear regression for estimating the cost function CR.
Both line assignment and production planning are iterated over
time in a rolling horizon framework, which results in a potential
time-to-time relocation of the products among lines as order and
forecast volumes vary. In each step of periodic re-planning,
investment costs are calculated to reflect the necessary changes in
the resource pool with respect to the current capacities. The time
horizon of planning is a few months on both levels, with monthly
re-planning on a rolling horizon basis. While line assignment is a
continuous-time decision that can be revised only during periodic
re-planning, production planning is performed on a discrete time
scale with time units corresponding to one shift.

2.4. Notation

The notation used throughout the paper is presented below:

Parameters:
J set of machine types
Fig. 2. Workflow of the capacity management method.
P set of products
When searching for the optimal allocation, the current customer T set of shifts
orders as well as the forecast volumes are considered on a predefined ej purchase price of machine type j
time horizon. The total production cost is composed of the
oj operation cost of a machine of type j per shift
investment, the operation, and the personnel costs. The changeover
cost in the reconfigurable system is assumed to be an order of h cost of an operator per shift
magnitude smaller than the above cost components. qp total order and forecast volume of product p
D. Gyulai et al. / CIRP Annals - Manufacturing Technology 63 (2014) 457–460 459

pp processing time of product p The objective is to minimize the total production cost (3),
rjp required number of machines of type j by product p composed of the purchase price of the machines that are not readily
available in the current resource pool, the personnel costs, and the
n0j number of machines of type j in the resource pool
operation costs. Constraint (4) specifies the required number of
s duration of a shift machines from each type in each shift, while equality (5) states that
Decision variables (line assignment): every customer order must be fulfilled. Constraints (6) define the
R set of products in the reconfigurable system variable domains, and require that at most one reconfigurable line
should produce the same product in the same shift.
Decision variables (production planning):
The resulting production plan specifies the set of machines to be
xt p number of lines producing product p in shift t purchased and the system configuration in each shift. The above
nj number of required additional machines of type j model, however, ignores the cost and time of reconfiguration, and
leads to a plan in which the sequence of the shifts can be changed
3. Solution method arbitrarily. In order to minimize the number of reconfigurations, a
sequencing problem is solved that re-orders the shifts, but leaves
3.1. Line assignment the system configuration unchanged within each shift. This can be
represented as a symmetric Travelling Salesman Problem (TSP), in
The goal of line assignment is to decide whether a certain which the vertices are the shifts, while the cost of an edge is the
product should be assembled on a dedicated or on a reconfigurable number of different lines between the configurations in the two
line, or it should be outsourced (Fig. 2). While the production costs corresponding shifts.
in the dedicated system and by outsourcing can be computed in
closed form from the input parameters, the costs in the 4. Experimental results
reconfigurable system are dependent on the current product
mix. Therefore, we propose to predict this cost using multivariate The proposed capacity management and production planning
linear regression based on the production costs in randomly method was tested on an industry-related dataset, considering
generated scenarios. For the regression, the following calculation historical order and forecast volumes, as well as real production
model is applied: lines. The product portfolio consists of 67 products with diverse
0 1 order volumes and assembly processes. The training dataset for the
R
X X regression contained 80 production planning problem instances,
C ¼ b0 þ b1 jRj þ @ b2 p p q p þ b jþ2 r j p A þ e; (1) generated by assigning a random production volume to each
p2R j2J
product. The multivariate regression was computed using the R
where the bs are unknown parameters that are estimated, b0 is the environment, by applying its general linear regression function,
intercept and e is the error term [17]. By neglecting e, the formula which took ca. 2 s [18]. This provided an appropriately precise
above can be rearranged as follows: prediction of the production cost in the reconfigurable system,
0 1 with a coefficient of determination of R2 = 0.987, as shown in Fig. 3.
R
X X X
C  b0 þ @b þ b p p q p þ b jþ2 r j p A ¼ b0 þ ap (2) 10000
1 2
p2R j2J p2R 9000
8000
Thus, it is enough to estimate only the ap values subsequently. The 7000
Cost [Units]

regression was computed on randomly generated production 6000


scenarios in the reconfigurable assembly system, solved by the 5000
production planning model presented in Section 3.2. The scenarios 4000
were divided into independent training and test sets. As regression 3000
assigns a separate production cost ap to each product p, line 2000
1000
assignment can be performed for individual products, by
0
comparing the production costs associated to the three candidate
1 51 101 151
production modes. Products p where ap is the lowest among the Scenarios (#)
costs will be produced on reconfigurable lines, and hence, Calculated cost Predicted cost
constitute the set R for production planning.
Fig. 3. Comparison of the production costs predicted by multivariate linear
regression and calculated by production planning (scenarios are ordered by
3.2. Production planning for the reconfigurable system increasing total order volume).

Medium term production planning in the reconfigurable


system addresses the integrated configuration optimization and The line assignment problem was solved iteratively using a
resource assignment of the system. Planning is solved on a discrete rolling horizon scheme, with a fixed horizon of 3 months in each
time horizon with time units corresponding to individual shifts. iteration. In order to evaluate the efficiency of the method, a
The planning problem is formulated as a mixed integer linear reference solution is considered in which all products are
programme as follows: assembled on dedicated lines. The results show that applying
minimize reconfigurable lines and the proposed planning method results in
X XX XX X significant cost savings even in case of fluctuating order streams. It
e jn j þ h xt p þ o j r j p xt p (3) is typical that savings are higher (up to 30%) in periods with lower
j2J t2T p2R t2T j2J p2R
production volumes, whereas they are lower (10–15%) around
subject to peak production volumes (Fig. 4.).
X Since most products in the industrial data set were in the
n j þ n0j  r j p xt p 8 j; t (4) maturity stage of their life cycle, their production volumes typically
p2R
decreased over the four-year horizon. Accordingly, the number of
X products assembled in the dedicated system slightly decreased,
qp pp
d e ¼ xt p 8p (5) whereas the number of products in the reconfigurable system and
s t2T products outsourced increased over time (Fig. 5).
The production planning model was run on the products
n j  0; xt p 2 f0; 1g 8 j; t; p (6) assigned to the reconfigurable system. The proposed MIP model
460 D. Gyulai et al. / CIRP Annals - Manufacturing Technology 63 (2014) 457–460

1400000 35% portfolio consisting of high- and low-volume products. The


Total producon volumes [pcs.]

1200000 30% approach offers an integrated way for the assignment of products
to dedicated or reconfigurable resources and for the production
1000000 25%
planning of the reconfigurable ones. An essential element of the

Cost saving [%]


800000 20% system developed within an industrial project is that cost
600000 15%
predictions computed by multivariate linear regression on virtual
production scenarios support the solution of the line assignment
400000 10% problem. The production planning level also incorporates a
200000 5% sequencing module for minimizing the number of reconfigura-
tions. The applicability of the proposed method was demonstrated
0 0%
by results of experiments, using industrial data. Further research
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
and development activities will include the refinement of the
Planning periods [3 months]
underlying models in the course of further industrial applications.
Producon volume Cost saving

Fig. 4. Results of line assignment: cost savings and production volume over a four- Acknowledgements
year horizon.
Research has been partially supported by National Develop-
ment Agency, Hungary Grant No. ED_13-2-2013-0002 and by the
70
European Union 7th Framework Programme Project No. NMP
60 2013-609087, Shock-robust Design of Plants and their Supply
Number of products

50 Chain Networks (RobustPlaNet) Grant No. NMP-2013-609087,


40 www.robustplanet.eu.
30
20
10
References
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 [1] Váncza J, Monostori L, Lutters D, Kumara S, Tseng M, Valckenaers P, Van Brussel
Dedicated Reconfigurable Outsource H (2011) Cooperative and Responsive Manufacturing Enterprises. CIRP Annals
– Manufacturing Technology 60(2):797–820.
Fig. 5. Results of line assignment: number of products assembled using the three [2] ElMaraghy HA, Schuh G, ElMaraghy W, Piller F, Schönsleben P, Tseng M,
production modes over a four-year horizon. Bernard A (2013) Product Variety Management. CIRP Annals – Manufacturing
Technology 62(2):629–652.
[3] Koren Y, Heisel U, Jovane F, Moriwaki T, Pritschow G, Ulsoy G, Van Brussel H
(1999) Reconfigurable Manufacturing Systems. CIRP Annals – Manufacturing
was solved using FICO Xpress and its default branch and bound Technology 48(2):527–540.
method. The search was run until an optimality gap of at most 4% [4] Koren Y (2013) The Rapid Responsiveness of RMS. International Journal of
Production Research 51(23–24):6817–6827.
was achieved, which required 116 s on average. The subsequent [5] Putnik G, Sluga A, ElMaraghy HA, Teti R, Koren Y, Tolio T, Hon B (2013)
sequencing problem was solved using the open-source solver LKH Scalability in Manufacturing Systems Design and Operation: State-of-the-
[19], which implements the Lin–Kernighan heuristic [20]. Solving art and Future Developments Roadmap. CIRP Annals – Manufacturing Technol-
ogy 62(2):751–774.
the problem using the default randomized restart strategy with 10 [6] Wang W, Koren Y (2012) Scalability Planning for Reconfigurable Manufactur-
runs required 59 s altogether. Sequencing reduced the number of ing Systems. Journal of Manufacturing Systems 31:83–91.
reconfigurations by 51%, resulting in a significantly smoother [7] Ceryan O, Koren Y (2009) Manufacturing Capacity Planning Strategies. CIRP
Annals – Manufacturing Technology 58(1):403–406.
production plan, as depicted in Fig. 6. [8] Lanza G, Peters S, Herrmann H-G (2012) Dynamic Optimization of Manufactur-
ing Systems in Automotive Industries. CIRP Journal of Manufacturing Science
and Technology 5(4):235–240.
[9] Lanza G, Peters S (2012) Integrated Capacity Planning over Highly Volatile
Horizons. CIRP Annals – Manufacturing Technology 61(1):395–398.
[10] Asl FM, Ulsoy AG (2003) Stochastic Optimal Capacity Management in Recon-
figurable Manufacturing Systems. CIRP Annals – Manufacturing Technology
52(1):371–374.
[11] Colledani U, Tolio T (2005) A Decomposition Method to Support the Configu-
ration/Reconfiguration of Production Systems. CIRP Annals – Manufacturing
Technology 54(1):441–444.
[12] Deif AM, ElMaraghy W (2006) Effect of Reconfiguration Costs on Planning for
Capacity Scalability in Reconfigurable Manufacturing Systems. International
Journal of Flexible Manufacturing Systems 18(3):225–238.
[13] Hon K, Xu S (2007) Impact of Product Life Cycle on Manufacturing Systems
Reconfiguration. CIRP Annals – Manufacturing Technology 56(1):455–458.
[14] Arafa A, ElMaraghy W (2011) Manufacturing Strategy and Enterprise Dynamic
Fig. 6. Part of the production plan before (above) and after (below) performing the Capability. CIRP Annals – Manufacturing Technology 60(1):507–510.
sequencing: each product is identified by a unique colour. [15] Youssef AM, ElMaraghy HA (2008) Performance Analysis of Manufacturing
Systems Composed of Modular Machines Using the Universal Generating
Function. Journal of Manufacturing Systems 27:55–69.
[16] Gyulai D, Vén Z, Pfeiffer A, Váncza J, Monostori L (2012) Matching Demand and
5. Conclusions System Structure in Reconfigurable Assembly Systems. Procedia CIRP 3:579–
584.
[17] Rawlings JO, Pantula SG, Dickey DA (2001) Applied Regression Analysis: A
Both dedicated and reconfigurable assembly lines can show up
Research Tool, 2nd ed. Springer Verlag, NY, USA.
benefits and drawbacks, consequently, their co-existence in [18] R Core Team (2013) R: A Language and Environment for Statistical Computing, R
industrial practice is natural and frequent. In the paper, a novel Foundation for Statistical Computing. http://www.R-project.org/.
approach for capacity management for assembly system with [19] Helsgaun K (2000) An Effective Implementation of the Lin–Kernighan Traveling
Salesman Heuristic. European Journal of Operational Research 126(1):106–130.
dedicated and reconfigurable assembly lines was presented that [20] Lin S, Kernighan BW (1973) An Effective Heuristic Algorithm for the Traveling-
facilitates the economical production of a diverse, varying product Salesman Problem. Operations Research 21(2):498–516.

You might also like