- Email: [email protected]
Risk Efficient Migration Strategies for Global Production Networks
Available online at www.sciencedirect.com
ScienceDirect Procedia CIRP 57 (2016) 104 – 109
49th CIRP Conference on Manufacturing Systems (CIRP-CMS 2016)
Risk efficient migration strategies for global production networks E. Mosera*,Nicole Strickera, G. Lanzaa a
wbk Institute of Production Science, Karlsruhe Institute of Technology (KIT), Germany
* Corresponding author. Tel.: +49 -721-608-46939; fax: +49 -721-608-45005. E-mail address: [email protected]
Abstract The global competitive environment forces companies to increase their production efficiency. In recent years, companies attempted to increase efficiency by operating in globally distributed production networks. But these networks often grew historically neglecting a future-oriented strategy. Hence, nowadays investments in the adaptation of the network configurations are a promising lever to increase the efficiency. Since multiple influencing factors affect the efficiency of a network configuration and interdependencies of the factors have to be considered as well, the decision-making process has become very complex. Furthermore, due to the short-term cyclical nature of these factors, the adjustment time of the network configuration is even more important within this complex decision-making problem. Combining a stochastic, dynamic optimization model with a portfolio theory approach, a method for risk-efficient migration strategies for global production networks is presented in this article. The method considers both, financial and time expenses for the adaptation of the network configuration. Using the proposed method strategies for proactive adaptations of the network configuration considering the risk aversion of the decision maker can be recommended. © Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license ©2016 2015The The Authors. Published by Elsevier B.V. (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of Scientific committee of the 49th CIRP Conference on Manufacturing Systems (CIRP-CMS 2016). Peer-review under responsibility of the scientific committee of the 49th CIRP Conference on Manufacturing Systems Keywords: Global production networks, migration planning, optimization, changeability
1. Introduction Companies of all sizes established globally distributed production networks during the last decades [1],[27],[31]. The main reasons for the global expansion of the manufacturing sector were the development of new sales markets and the efforts to increase efficiency by using regional advantages of locations and concentrating on core competencies [1],[16],[27],[31]. However, these networks often grew historically or opportunistically neglecting a future-oriented strategy [23],[24]. As a result of these relocations, acquisitions and disposals, complex network structures with heterogeneous locations are widespread [6],[24]. In addition to globally distributed production networks, the competitive and turbulent business environment characterizes modern globalization. Nowadays, companies are faced with global competitors, shortened product life cycles, an increasing product variety and a volatile demand [31]. Hence, multiple influencing factors as well as their interdependencies challenge globally distributed production networks [12],[20],[27],[31]. The use of modern technologies (e.g. ICT) also reveals the
short-term cyclical nature of these changing factors [14]. Not least because of this, experts claim that the business environment is discontinuously and hardly predictable [29],[30]. In the context described above, further increases in efficiency and reduction of complexity require adaptations of grown network structures [24]. Among others, the integration of new supplier from developing economies (e.g. BRIC) but also relocations of value-added activities and investment in production resources (e.g. replacement of outdated production technology) are promising levers [24]. Simultaneously, today’s discontinuous business environment forces companies to adapt their production network more and more frequently [14],[26],[29]. Hence, it is obvious that the so-called migration planning of production networks is considered vital for operating efficient production networks. Beside migration costs, the time required for an adaptation of the network configuration must also be considered when generating efficient migration strategies.
2212-8271 © 2016 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 the 49th CIRP Conference on Manufacturing Systems doi:10.1016/j.procir.2016.11.019
105
E. Moser et al. / Procedia CIRP 57 (2016) 104 – 109
2. Objective The objective of this paper is to present a methodology for migration planning for globally distributed production networks. By using a stochastic, dynamic optimization model robust migration paths can be identified as reaction to changes of crucial influencing factors and lever to increase the efficiency of the network. Simultaneously, considering multiple future developments of these factors, it is possible to avoid potential misinvestments by generating migration strategies for a multiperiodic planning horizon. In addition, a portfolio theory approach helps finding risk-efficient bundles of change enabler for proactive implementation. Using this information, concrete migration processes for each decision point including an implementation plan for these enablers can be derived. The methodology provides migration strategies for a globally distributed production network on both, single and multiperiodic considerations. It allows to find the right compromise between a reactive and proactive behavior towards changes of crucial influencing factors. 3. Literature review Due to the complexity of decision making processes regarding production networks and the number of potential solutions, optimization models have been acknowledged to be a powerful method to support decision making. Recent research approaches provided significant contribution to the strategic planning of production networks. However, a variety of them neglects any multistage adaptation of the production network (cf. [4],[5],[9],[10]). Other approaches integrate efforts for the continuous adaptation (cf. [16],[21],[25]). Although uncertain future developments of multiple influencing factors are included (cf. [16],[25]), stochastic effects cannot be considered adequately as the approaches base on scenario technique [1]. A few approaches take stochastic effects into account, but they only concentrate on customer demand as crucial influencing factor (cf. [15],[18],[26]) Furthermore, approaches particularly for migration planning for production networks barely exist. And if so (cf. [23],[25]) they are still subject to the restrictions mentioned above. In general, migration planning and especially identifying change enablers as well as the generation of change processes are also a subject of discussion in research approaches in the field of changeability. Existing approaches mainly focus on the design and evaluation of changeable production systems (cf. [7],[8]). Several approaches address the selection of change enablers and the generation of migration processes (cf. [11],[28]). But a risk evaluation of change enablers for proactive implementation is not included in these approaches. 4. Methodology The underlying approach of this work consists of a combination by linking a stochastic, dynamic optimization to a portfolio selection approach for migration planning for global production networks. The approach enables the identification of risk-efficient migration strategies for global production
networks. Essentially, the methodology consists of three modules (Fig. 1). Module I:
2SWLPL]DWLRQ Migration Paths Module II:
(QDEOHURI&KDQJH Risk-efficient Portfolios Module III:
0LJUDWLRQ3URFHVV Implementation Plan Fig. 1. Structure of the presented approach.
The optimization modules identify robust migration paths for a defined planning horizon and consequently provide the basis for the subsequent modules. For every decision point of the planning horizon, robust migration paths will be identified by applying a stochastic, dynamic optimization model. As a result, this module provides a system of contingency plans for the migration of the production network and consequently the migration strategy for the complete planning horizon. Within the subsequent module, each decision point in the planning horizon will be analyzed in detail. First, the demand for change will be identified followed by the derivation of change enablers for the relevant network objects. Eventually, these enablers are bundled. Applying the idea of risk diversification according to the portfolio theory of Markowitz [19], risk efficient portfolios can then be generated. Based on these analyzes, concrete migration processes will be derived considering the preferences and available resources of the decision maker. These migration processes include an implementation plan for the identified change enablers. The modules are described in more detail in the following sections. 4.1. Module I: Optimization The optimization module allows the identification of robust migration paths of the production network. They include both, adaptations of the structural and capacitive network configuration. For this reason, an integrated network model is developed by applying the graph theory. Within this modelling approach, the integrated (structural and capacitive) network configuration is modelled via rated nodes (n) and edges (e) and comprises the following network objects: customer (modelled as node n) who are demanding products, production stages (n) and technologies (n) which are localized at sites, suppliers (n) who supply materials or semi-finished products and outsource partners (n) who handle single production stages of the value adding process. The connections between sites, suppliers and customers are modelled as transportation routes, and represented as the object transport (e). The capacitive network
106
E. Moser et al. / Procedia CIRP 57 (2016) 104 – 109
configuration includes the production volumes of the technologies and the transport volumes of materials, intermediate and final goods between the sites, between site and supplier/outsource partner and between site and customer. The capacitive configuration parameters are determined by the structural network configuration and calculated implicitly. Integer variables are used for specifying the capacitive configuration whereas for structural configuration parameters only binary variables are allowed. Beside existing objects representing the status quo, further objects need to be added that will reflect potential changes of the network configurations. Feasible combinations of these potential network objects will be bundled into consistent actions in the context of optimization. Again, in the context of optimization, a migration path includes a consistent bundle of these already bundled actions. In order to consider the multi-dimensional uncertainty of business environment, key drivers of change will be modelled stochastically. For this purpose, all internal and external factors that affect the focal company will be collected. Then, the key drivers of change will be determined considering the interdependencies and relevance of all collected factors. Having identified them, the induced impact and the corresponding parameters in the optimization model are assigned to each driver. Yet, it is often not possible to reliably forecast a future development (cf. 1). Hence, for each key driver of change with uncertain (incl. unknown) future development, multiple developments will be modelled. Subsequently, consistent developments for each discrete decision point of the planning horizon will be clustered to environmental states. Since this is a stochastic, dynamic optimization approach, all clustered parameters as well as the probability of occurrence of the clustered state will be quantified for each state. In the context of the optimization module, the overall objective is the minimization of the expected costs over the planning horizon by finding an optimal policy (cf. 5.3). Again, an optimal policy is a sequence of migration paths over the planning horizon. Since the optimization module is a flexible planning approach, an optimal migration path will be identified for each decision point of the planning horizon and state by applying the optimization model (cf. 5.3). Since costs incurred for the adaptation and operation of a network, both cost elements are considered in this process. Again, both cost elements include cost incurred in production and logistic and they allow the calculation of the total landed cost for a defined planning horizon. A holistic evaluation of the period costs is not limited to production costs but rather considers the total landed cost of operating a production network. Here, the period costs are the sum of the production costs logistic costs. The production costs include overhead costs, technology costs and procurement costs. The logistic costs consist of costs for the physical transportation of material and components from supplier to sites, of semi-finished goods between sites and outsource partners and of finished goods to customers. Furthermore, inventory costs including capital commitment costs and costs for safety stocks as well as customs costs are among the logistics costs. Customs costs include penalty costs of unfulfilled local content requirements. Migration costs occur
for structural and capacitive adaptations of the network configuration. In production, the localization of production stages at sites and the allocation of suitable technologies causes costs. Herein, also the replacement of existing technologies must be considered. Migration costs in logistics include costs for the integration of suppliers and outsource partners. To identify robust migration paths, the stochastic dynamic optimization is implemented in MATLAB by MathWorks® and, the MDP Toolbox will be used to solve the decision problem. In total, a system of contingency plans results including the migration paths for each decision point and environmental state. Hence, the optimal policy for migrating the network configuration is defined by applying the optimization module. 4.2. Module II: Enabler of Change The module concentrates on detailed analyzes for each decision point of the planning horizon. Hence, each discrete decision point is analyzed individually to generate risk-efficient portfolios of change enablers. Having identified robust migration paths for each decision point in the previous module, the demand for change of the network configuration will be determined, initially. This will be executed by comparing the configuration parameters of two consecutive decision points of the planning horizon. Thus, the network objects to be changed can be determined precisely for each decision point and environmental state of the planning horizon. For each network object to be changed, the demand for change is identified and appropriate change enablers (organizational, spatial and technology-based) are collected. However, the implementation of these enablers takes time (adjustment time to the desired effect of the change) and causes investment costs (e.g. initial, facility, replacement or additional investments), direct costs (e.g. costs for changeover, disestablishment and establishment of process capability) and opportunity costs (e.g. production downtimes, extra work or inventory costs) [7],[8],[30]. In case of a proactive implementation, meaning that the enablers of change are implemented before the identified point in time for change, the costs can be reduced. It becomes obvious that the inefficiency and, simultaneously, the costs in case of a proactive course of action can be significantly reduced [14]. But a proactive implementation of an enabler of change for a single object is not free of risk. This is due to the stochastic properties of the key driver of change and the accordingly modelled environmental states (cf. 4.1). Since the demand for change may differ for each state in the planning horizon (also for states of the same decision point) a proactive implementation of change enablers is associated with risk. It may happen, for example, that a network object will only be activated for one state but not for the other states of the same decision point. To consider these risks in the decision about a proactive implementation of change enablers the ߤ െ ߪ principle [17] is applied calculating the return on investment of a proactive implementation. The return on investment is determined by dividing the reduction on migration costs by the sum of the investment cost and direct costs of the change enablers. The
107
E. Moser et al. / Procedia CIRP 57 (2016) 104 – 109
mean ߤ and standard deviation ߪ of the return on investment depends on whether the change enabler is needed in the states or not. Again, this depends on the migration paths of the states. Since the migration paths and the probability of occurrence of the states are already known, the mean ߤ and standard deviation ߪ can be easily calculated. Subsequently and by applying the basic idea of the portfolio theory of Markowitz [19], several enablers of change for different network objects will be bundled to portfolios. The motive for this approach is that the risk of the portfolio is lower than the sum of the individual risks. By implementing several enablers of change for different network objects, the risk can be diversified. Hence, risk efficient portfolios will be generated. For this reason, consistent change enablers are bundled and then assessed by applying the described approach. To ensure the comparability of different portfolios, the difference in costs of each portfolio compared to the most expensive one is then invested risk-neutral into the capital market. Finally, pareto dominated portfolios are eliminated.
the migration process and a concrete implementation plan for the change enablers (proactive and reactive) will be derived. WŽŝŶƚŝŶƚŝŵĞĨŽƌĐŚĂŶŐĞ WƌŽĂĐƚŝǀĞ ZĞĂĐƚŝǀĞ ŶĂďůĞƌϮ ŶĂďůĞƌϭ
ŶĂďůĞƌϯ
ŶĂďůĞƌϰ
ŶĂďůĞƌϱ
Step 1 Step 2 Step 3 Step 4 Step 5 Step 6 Step 7 Step 8
ƚŝŵĞŽĨŵŝŐƌĂƚŝŽŶƉƌŽĐĞƐƐ Fig. 3. Exemplary migration process using a Gantt chart with an implementation plan for change enablers (proactive and reactive).
Finally, concrete migration processes including an implementation plan for the change enablers result from it. The recommended starting point of the migration processes can be derived too by considering the required adjustment time of the change and the available resources of the company. 5. Module formulation
Fig. 2. Risk efficient portfolios of change enablers.
Similar to the optimization module, parts of this module are implemented in MATLAB by MathWorks®. As a result, riskefficient portfolios of change enablers are identified for each decision point. 4.3. Module III: Migration process The final module aims at generating migration processes and takes into account specific requirements of the decision maker. Depending on the preferences of the decision maker (riskseeking, risk-averse or risk-neutral), the portfolios will be recommended in a user-specifical way. Thus, the decision maker can choose the favored portfolio of change enablers for a proactive implementation. Not only the preferences of the decision maker can be respected, but also the available resources (e.g. capital, human resources) of the company. Having chosen the portfolio of proactive change enablers, the migration process will be derived. Considering logical and causal interdependencies of the single process steps, the migration process can be structured and modelled, for example, by applying the metra-potential method (mpm). Among other things, the required time for the migration process can be calculated. Then, based on these analyzes, the starting point of
As part of the optimization module, a stochastic and dynamic optimization model is formulated, whose mathematical formulation and solution method are presented in the following sections. 5.1. Mathematical formulation In general, the decision problem of identifying robust migration paths can be described as a multi-stage decision process and is defined by the planning horizon, the discrete decision (finite) point in time, possible adaptations adjustments and possible states of key drivers of change. Hence, the decision problem can be formulated as a (finite horizon) Markovian Decision Process (MDP) under uncertainty [13]. Below, the MDP is formally described using the standard mathematical formulation. The formulation is partially similar to [15]: Planning horizon ൌ ͳǡ ǥ ǡ ܶ : The planning horizon ܶ represents the planning horizon of migration planning and is finite. Dividing the planning horizon into adequate time slices, a finite number of equidistant decision points is specified. For each point in time, decisions about the adaptation of the network configuration must be made. State Space ܵ௧ : For each point in time ݐ, a finite, non-empty set of states ܵ௧ is defined. A state ݏ௧ ൌ ሺ݇௧ ǡ ݊ሬԦ௧ ሻ ܵ א௧ defines the initial network configuration at the beginning of ݇ ݐ௧ ܭ א௧ and the demand situation ݊ሬԦ௧ ܰ א௧ . The demand for individual
108
E. Moser et al. / Procedia CIRP 57 (2016) 104 – 109
products is summarized in a vector ݊ሬԦ௧ . Thus, it follows:ܵ௧ ൌ ܭ௧ ൈܰ௧ . Since the decision model focusses on brownfield scenarios, the model is initialized with the current network configuration.
Terminal reward function ்ݒሺ݇ ் ሻ : In ݐൌ ܶ the allocated technologies of each site are sold. The revenues which are independent from usage time determine the terminal reward ்ݒሺ݇ ் ሻ.
State Space ܹ௧ : For each point in time ݐ, a finite, non-empty set of states ܹ௧ is defined. A state ݓ ሬሬԦ௧ ൌ ሺݓ௧ଵ ǡ ǥ ǡ ݓ௧ ሻ ܹ אఛ determines key drivers of change for a point in time. For each decision point, one to several different states may exist. These states are generated within a consistency check (cf. 4.1).
5.2. Transformation
Action space ܣ௧ : For each single state ݏ௧ ܵ א௧ , a finite, admissible set of actions ܣ௧ ሺݏ௧ ሻ exists. An action ܽ௧ ሺݏ௧ ሻ א ܣ௧ ሺݏ௧ ሻ can include both, structural and capacitive adaptions of network configuration. By choosing an action ܽ௧ ሺݏ௧ ሻ א ܣ௧ ሺݏ௧ ሻthe initial configuration migrates to configuration ݇௧ାଵ . It is assumed that the time for reconfiguration can be neglected. Hence, the demand in ݐis produced with configuration ݇௧ାଵ . The admissible action space ܣ௧ ሺݏ௧ ሻ is defined implicitly by equality and inequality constraints. Exemplary constraints are: •
Capacity restrictions of sites, suppliers and outsource partners
•
Optional strategic inputs (here: volume flexibility)
Transition law : The transition law defines the probability to reach ݏ௧ାଵ being in ݏ௧ and choosing action ܽ௧ in ݐ. Only the demand ݊ሬԦ௧ in period t is stochastic while the action ܽ௧ is deterministic. Furthermore, it is assumed that demand holds a so-called Markov-property, meaning that the demand in ݐ ͳ only depends on the current demand in ݐ. Thus, it ܲ൫ܵ߬ͳ ൌ ݐݏͳ ȁܵ߬ ൌ ݐܣ ٿ ݐݏൌ ܽ ݐ൯ ൌ ሺܰ߬ͳ ൌ follows:
ሬ݊Ԧݐͳ ȁܰ߬ ൌ ݊ሬԦ ݐሻ
The following section describes how the MDP under uncertainty, which was described previously, is transformed in a (finite horizon) Markovian Decision Process (MDP). This enables the application of backward induction to solve the stochastic, dynamic optimization problem recursively. Referring to [13], the transformation is performed according to the following instructions: It is: ܵመ ൌ ܵൈܹ , Ƹ ǣ ܣ௧ ൈܵመ ՜ ሾͲǡͳሿ and Ƹ ሺݏƸ௧ ǡ ܽ௧ ǡ ݏƸ௧ାଵ ሻ ൌ ሺݏ௧ ǡ ݓ ሬሬԦ௧ ǡ ܽ௧ ǡ ݓ ሬሬԦ௧ାଵ ሻݍሺݓ ሬሬԦ௧ ǡ ݓ ሬሬԦ௧ାଵ ሻ . Thus, it is: ݏƸ௧ ሺݏ௧ ǡ ݓ ሬሬԦ௧ ሻ ܵ אመ and ݏƸ௧ାଵ ሺݏ௧ାଵ ǡ ݓ ሬሬԦ௧ାଵ ሻ ܵ אመ. 5.3. Solution method The optimization module aims at identifying robust migration paths. Hence, for each state ݏƸ௧ an action ܽ௧ must be selected. This is based on a decision function ݂ǣݏƸ௧ ՜ ܽ௧ with ݂௧ ሺݏƸ௧ ሻ א ܣ௧ ሺݏƸ௧ ሻ. Again, an optimal migration strategy is a sequence of decision functions, specifying action ܽ௧ ൌ ݂௧ ሺݏƸ௧ ሻ to be taken in state ݏƸ௧ . The sequence is also named policy and formally described with ߨ ൌ ൫݂ ሺݏƸ ሻǡ ݂ଵ ሺݏƸଵ ሻǡ ǥ ǡ ்݂ିଵ ሺݏƸ ்ିଵ ሻ൯. The overall objective is the minimization of the expected costs over the planning horizon by finding the optimal policy. The optimal actions for all points in time and admissible states are identified by applying Bellmann’s Principle of Optimality [2].
Transition law ݍ: The transition law ݍdefines the probability to reach ܹ௧ାଵ being in ܹ௧ in ݐ. To hold Markov-property future developments of the clustered key drivers of change are modelled as a homogenous markov chain. Thus it follows that ݍሺݐሻ ൌ ሺܹఛାଵ ൌ ݓ ሬሬԦ௧ାଵ ȁܹఛ ൌ ݓ ሬሬԦ௧ ሻ ൌ ሺܹఛାଵ ൌ ݓ ሬሬԦ௧ାଵ ȁܹఛ ൌ ݓ ሬሬԦ௧ ǡ ܹ௧ିଵ ൌ ݓ ሬሬԦ௧ିଵ ǡ ǥ ǡ ܹ ൌ ݓ ሬሬԦ ሻ . Again, since the decision model focusses on brownfield scenarios, the model is initialized with the current condition of the key drivers of change ݓ ሬሬԦ .
It follows for ͳ ݐ൏ ܶ:
One-stage reward function ࢚࢘ : The one-stage reward function rሺ݇௧ ǡ ݊ሬԦ௧ ǡ ݓ ሬሬԦ௧ ǡ ܽ௧ ሻ includes the period costs ݈ܿݐሺ݇௧ ǡ ݊ሬԦ௧ ǡ ݓ ሬሬԦ௧ ǡ ܽ௧ ሻ and the cost for adapting the network configuration ݉ܿሺ݇௧ ǡ ܽ௧ ሻ. Since the reconfiguration time is neglected, it follows:
and finally for ݐൌ ܶ:
ሬሬԦ௧ ǡ ܽ௧ ሻ ൌ ݈ܿݐሺ݇௧ାଵ ǡ ݊ሬԦ௧ ǡ ݓ ሬሬԦ௧ ሻ ݈ܿݐሺ݇௧ ǡ ݊ሬԦ௧ ǡ ݓ
(1)
Migration costs are identified by comparing the configuration parameters of ݇௧ ݏ א௧ and ݇௧ାଵ ݏ א௧ାଵ : ݉ܿሺ݇௧ ǡ ܽ௧ ሻ ൌ ݉ܿሺ݇௧ ǡ ݇௧ାଵ ሻ
(2)
A detailed description of the metrics to calculate the period and migration costs will be omitted.
ܸ௧ ሺݏƸ௧ ሻ ൌ
א ሺ௦Ƹ ሻ
ቐݎሺݏƸ௧ ǡ ܽ௧ ሻ
௦Ƹశభ אௌመశభ
ܲ௧ ሺݏƸ௧ ǡ ܽ௧ ǡ ݏƸ௧ାଵ ሻ (3)
ܸ כ௧ାଵ ሺݏƸ௧ାଵ ሻቑ
்ܸ ሺݏƸ ் ሻ ൌ ்ݒሺ݇ ் ሻ ݏƸ௧ ܵ אመ௧
(4)
By using backward induction, the formulated MDP can be solved. Currently, MATLAB by MathWorks® is applied to identify the migration paths. 6. Conclusion This paper presents a methodology for migration planning in global production networks. The methodology consists of three interlinked modules. First, robust migration paths of the network configuration are identified by applying the presented optimization module. Subsequently, portfolios including several enablers of change are generated and assessed. Doing
E. Moser et al. / Procedia CIRP 57 (2016) 104 – 109
this, risk-efficient portfolios can be identified by taking the risk aversion of the decision maker into account. Finally, concrete migration processes can be derived considering the estimated adjustment time and the available resources. This paper is based on current research performed at wbk Institute of Production Science. Future work aims at more detailed elaboration of the methodology. The focus is laid on the detailed formulation of a multi-objective optimization model to identify optimal portfolios taking the risk aversion of a decision maker into consideration. Furthermore, the application of the methodology in a comprehensive industrial case study is an integral part of further works. Acknowledgements This research has been partially supported by the European Union’s 7th Framework Program Project No: NMP 2013609087, “Shock-robust Design of Plants and their Supply Chain Networks (RobustPlaNet)” and the research project LA2351/31-1 “Bestimmung von Wandlungsbedarf und – zeitpunkt globaler Produktionsnetzwerke: Methodische Unterstützung strategischer Mehrzielentscheidungen” of the German Research Foundation. This is gratefully acknowledged. References [1] Abele, E. (2008). Global production. A handbook for strategy and implementation. Springer, Berlin. ISBN: 978-3-540-71652-5. [2] Bellman, R. (1954). The theory of dynamic programming. Bull. Amer. Math. Soc., Vol. 60, Nr. 6, p. 503–515. [3] CIRP International Academy for Production Engineering Research (2013): Encyclopedia of Production Engineering, Keyword: Optimization in Manufacturing, Springer Reference (www.springerreference.com), Springer, Berlin Heidelberg. [4] Fleischmann, B., Ferber, S. & Henrich, P. (2006). Strategic Planning of BMW’s Global Production Network. Interfaces, Vol. 36, Nr. 3, p. 194–208. [5] Friese, M. (2008): Planung von Flexibilitäts- und Kapazitätsstrategien für Produktionsnetzwerke der Automobilindustrie, Dissertation, Leibniz Universität Hannover. [6] Friedl, T.; Thomas, S.; Mundt, A. (2013). Management globaler Produktionsnetzwerke – Strategie, Konfiguration, Koordination. München: Carl Hanser Verlag 2013 [7] Heger, C.L. (2007), Bewertung der Wandlungsfähigkeit von Fabrikobjekten, Dissertation, Leibniz Universität Hannover [8] Hernandez Morales, R. (2003), Systematik der Wandlungsfähigkeit in der Fabrikplanung, Dissertation, Leibniz Universität Hannover. [9] Jacob, F.(2006): Quantitative Optimierung dynamischer Produktionsnetzwerke, Dissertation, Technische Universität Darmstadt. [10] Kauder, S. (2008). Strategische Planung internationaler Produktionsnetzwerke in der Automobilindustrie, Dissertation, Wirtschaftsuniversität Wien. [11] Klemke, T. (2013), Planung der systematischen Wandlungsfähigkeit von Fabriken, Dissertation, Leibniz Universität Hannover.
[12] Koren, Y. (2010). Globalization and Manufacturing Paradigms. In: The Global Manufacturing Revolution: Product-Process-Business Integration and Reconfigurable Systems. p. 1–40. [13] Lange, V. (2013). Dynamische Kapazitätssteuerung in medizinischen Einrichtungen, Dissertation, Karlsruher Institut für Technologie. [14] Lanza, G., Moser, R. (2012). Strategic Planning of Global Changeable Production Networks. In. Procedia CIRP, Vol. 3, 2012, Pages 257–262 [15] Lanza, G., Peters, S. (2012). Integrated capacity planning over highly volatile horizons. In: CIRP Annals - Manufacturing Technology, Vol. 61, Issue 1, Pages 395–398. [16] Lanza, G., Moser, R. (2014). Multi-objective optimization of global manufacturing networks taking into account multi-dimensional uncertainty. CIRP Annals - Manufacturing Technology, Vol. 63, Issue 1, 2014, p. 397–400. [17] Laux, H. (2007), Entscheidungstheorie, Springer Verlag, Berlin [18] Lin, James T., Chen, Tzu-Li, Chu, Hsiao-Ching (2014). A stochastic dynamic programming approach for multi-site capacity planning in TFTLCD manufacturing under demand uncertainty. International Journal of Production Economics, Vol. 148, p. 21–36. [19] Markowitz, H. M.(1991). Portfolio Selection: Efficient Diversification of Investments. Yale University Press, New Haven & London. [20] McKinsey & Company Inc. (2009). Made in Germany Zukunftsperspektiven für die Produktion in Deutschland, Studie, Düsseldorf. [21] Melo, M.T., Nickel, S., Saldanha-Da-Gama, F. (2009). Facility location and supply chain management – a review. European Journal of Operational Research, Vol. 2, p. 401–412. [22] Nyhuis, P., Reinhart, G., Abele, E., 2008, Wandlungsfähige Produktionssysteme, PZH Produktionstechnisches Zentrum, Garbsen. [23] Reuter, C., Prote, J.P., Schmitz, T., Cremer, S. Ermittlung eines strategischen Migrationsplans für globale Produktionsnetzwerke. In ZWF – Zeitschrift für wirtschaftlichen Fabrikbetrieb, 110/15: 7-8. [24] Schmidt, B. (2011). Gestaltung Globaler Produktionsstrategien. In Wertschöpfung und Beschäftigung in Deutschland, p. 71–84. Springer Verlag, Berlin. [25] Schuh, G., Potente, T., Varandani, R., Schmitz, T. (2014). Global Footprint design based on genetic algorithms – An „Industry 4.0“ perspective. CIRP Annals – Manufacturing Technology, Vol. 63 p. 433436. [26] Stephan, H. A., Gschwind, T., Minner, S. (2010). Manufacturing capacity planning and the value of multi-stage stochastic programming under Markovian demand. Flexible services and manufacturing journal, Vol 22, Issue 3, p. 143–162. [27] Váncza, J.; Monostori, L.; Lutters, D.; Kumara, S.; Tseng, M.; Valckenaers, P.; Van Brussel, H. (2011). Cooperative and responsive manufacturing enterprises. In: CIRP Annals-Manufacturing Technology Vol. 60, Issue 2, p. 797–820. [28] Wagner, C. (2012), Kontinuierliche Gestaltung skalierbarer Produktionsstufen, Dissertation, Leibniz Universität Hannover. [29] Westkämper, E., Zahn, E., 2009, Wandlungsfähige Produktionsunternehmen. Das Stuttgarter Unternehmensmodell, Springer, Heidelberg. [30] Wiendahl, H.P.; Reichardt, J.; Nyhuis, P. (2009), Handbuch Fabrikplanung: Konzept, Gestaltung und Umsetzung wandlungfähiger Produktionsstätten, Carl Hanser Verlag, München. [31] Wiendahl H.-P., Lutz, S. (2002). Production in Networks. CIRP Annals – Manufacturing Technology 51(2):573–586.Van der Geer J, Hanraads JAJ, Lupton RA. The art of writing a scientific article. J Sci Commun 2000;163:51-9.
109
ScienceDirect Procedia CIRP 57 (2016) 104 – 109
49th CIRP Conference on Manufacturing Systems (CIRP-CMS 2016)
Risk efficient migration strategies for global production networks E. Mosera*,Nicole Strickera, G. Lanzaa a
wbk Institute of Production Science, Karlsruhe Institute of Technology (KIT), Germany
* Corresponding author. Tel.: +49 -721-608-46939; fax: +49 -721-608-45005. E-mail address: [email protected]
Abstract The global competitive environment forces companies to increase their production efficiency. In recent years, companies attempted to increase efficiency by operating in globally distributed production networks. But these networks often grew historically neglecting a future-oriented strategy. Hence, nowadays investments in the adaptation of the network configurations are a promising lever to increase the efficiency. Since multiple influencing factors affect the efficiency of a network configuration and interdependencies of the factors have to be considered as well, the decision-making process has become very complex. Furthermore, due to the short-term cyclical nature of these factors, the adjustment time of the network configuration is even more important within this complex decision-making problem. Combining a stochastic, dynamic optimization model with a portfolio theory approach, a method for risk-efficient migration strategies for global production networks is presented in this article. The method considers both, financial and time expenses for the adaptation of the network configuration. Using the proposed method strategies for proactive adaptations of the network configuration considering the risk aversion of the decision maker can be recommended. © Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license ©2016 2015The The Authors. Published by Elsevier B.V. (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of Scientific committee of the 49th CIRP Conference on Manufacturing Systems (CIRP-CMS 2016). Peer-review under responsibility of the scientific committee of the 49th CIRP Conference on Manufacturing Systems Keywords: Global production networks, migration planning, optimization, changeability
1. Introduction Companies of all sizes established globally distributed production networks during the last decades [1],[27],[31]. The main reasons for the global expansion of the manufacturing sector were the development of new sales markets and the efforts to increase efficiency by using regional advantages of locations and concentrating on core competencies [1],[16],[27],[31]. However, these networks often grew historically or opportunistically neglecting a future-oriented strategy [23],[24]. As a result of these relocations, acquisitions and disposals, complex network structures with heterogeneous locations are widespread [6],[24]. In addition to globally distributed production networks, the competitive and turbulent business environment characterizes modern globalization. Nowadays, companies are faced with global competitors, shortened product life cycles, an increasing product variety and a volatile demand [31]. Hence, multiple influencing factors as well as their interdependencies challenge globally distributed production networks [12],[20],[27],[31]. The use of modern technologies (e.g. ICT) also reveals the
short-term cyclical nature of these changing factors [14]. Not least because of this, experts claim that the business environment is discontinuously and hardly predictable [29],[30]. In the context described above, further increases in efficiency and reduction of complexity require adaptations of grown network structures [24]. Among others, the integration of new supplier from developing economies (e.g. BRIC) but also relocations of value-added activities and investment in production resources (e.g. replacement of outdated production technology) are promising levers [24]. Simultaneously, today’s discontinuous business environment forces companies to adapt their production network more and more frequently [14],[26],[29]. Hence, it is obvious that the so-called migration planning of production networks is considered vital for operating efficient production networks. Beside migration costs, the time required for an adaptation of the network configuration must also be considered when generating efficient migration strategies.
2212-8271 © 2016 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 the 49th CIRP Conference on Manufacturing Systems doi:10.1016/j.procir.2016.11.019
105
E. Moser et al. / Procedia CIRP 57 (2016) 104 – 109
2. Objective The objective of this paper is to present a methodology for migration planning for globally distributed production networks. By using a stochastic, dynamic optimization model robust migration paths can be identified as reaction to changes of crucial influencing factors and lever to increase the efficiency of the network. Simultaneously, considering multiple future developments of these factors, it is possible to avoid potential misinvestments by generating migration strategies for a multiperiodic planning horizon. In addition, a portfolio theory approach helps finding risk-efficient bundles of change enabler for proactive implementation. Using this information, concrete migration processes for each decision point including an implementation plan for these enablers can be derived. The methodology provides migration strategies for a globally distributed production network on both, single and multiperiodic considerations. It allows to find the right compromise between a reactive and proactive behavior towards changes of crucial influencing factors. 3. Literature review Due to the complexity of decision making processes regarding production networks and the number of potential solutions, optimization models have been acknowledged to be a powerful method to support decision making. Recent research approaches provided significant contribution to the strategic planning of production networks. However, a variety of them neglects any multistage adaptation of the production network (cf. [4],[5],[9],[10]). Other approaches integrate efforts for the continuous adaptation (cf. [16],[21],[25]). Although uncertain future developments of multiple influencing factors are included (cf. [16],[25]), stochastic effects cannot be considered adequately as the approaches base on scenario technique [1]. A few approaches take stochastic effects into account, but they only concentrate on customer demand as crucial influencing factor (cf. [15],[18],[26]) Furthermore, approaches particularly for migration planning for production networks barely exist. And if so (cf. [23],[25]) they are still subject to the restrictions mentioned above. In general, migration planning and especially identifying change enablers as well as the generation of change processes are also a subject of discussion in research approaches in the field of changeability. Existing approaches mainly focus on the design and evaluation of changeable production systems (cf. [7],[8]). Several approaches address the selection of change enablers and the generation of migration processes (cf. [11],[28]). But a risk evaluation of change enablers for proactive implementation is not included in these approaches. 4. Methodology The underlying approach of this work consists of a combination by linking a stochastic, dynamic optimization to a portfolio selection approach for migration planning for global production networks. The approach enables the identification of risk-efficient migration strategies for global production
networks. Essentially, the methodology consists of three modules (Fig. 1). Module I:
2SWLPL]DWLRQ Migration Paths Module II:
(QDEOHURI&KDQJH Risk-efficient Portfolios Module III:
0LJUDWLRQ3URFHVV Implementation Plan Fig. 1. Structure of the presented approach.
The optimization modules identify robust migration paths for a defined planning horizon and consequently provide the basis for the subsequent modules. For every decision point of the planning horizon, robust migration paths will be identified by applying a stochastic, dynamic optimization model. As a result, this module provides a system of contingency plans for the migration of the production network and consequently the migration strategy for the complete planning horizon. Within the subsequent module, each decision point in the planning horizon will be analyzed in detail. First, the demand for change will be identified followed by the derivation of change enablers for the relevant network objects. Eventually, these enablers are bundled. Applying the idea of risk diversification according to the portfolio theory of Markowitz [19], risk efficient portfolios can then be generated. Based on these analyzes, concrete migration processes will be derived considering the preferences and available resources of the decision maker. These migration processes include an implementation plan for the identified change enablers. The modules are described in more detail in the following sections. 4.1. Module I: Optimization The optimization module allows the identification of robust migration paths of the production network. They include both, adaptations of the structural and capacitive network configuration. For this reason, an integrated network model is developed by applying the graph theory. Within this modelling approach, the integrated (structural and capacitive) network configuration is modelled via rated nodes (n) and edges (e) and comprises the following network objects: customer (modelled as node n) who are demanding products, production stages (n) and technologies (n) which are localized at sites, suppliers (n) who supply materials or semi-finished products and outsource partners (n) who handle single production stages of the value adding process. The connections between sites, suppliers and customers are modelled as transportation routes, and represented as the object transport (e). The capacitive network
106
E. Moser et al. / Procedia CIRP 57 (2016) 104 – 109
configuration includes the production volumes of the technologies and the transport volumes of materials, intermediate and final goods between the sites, between site and supplier/outsource partner and between site and customer. The capacitive configuration parameters are determined by the structural network configuration and calculated implicitly. Integer variables are used for specifying the capacitive configuration whereas for structural configuration parameters only binary variables are allowed. Beside existing objects representing the status quo, further objects need to be added that will reflect potential changes of the network configurations. Feasible combinations of these potential network objects will be bundled into consistent actions in the context of optimization. Again, in the context of optimization, a migration path includes a consistent bundle of these already bundled actions. In order to consider the multi-dimensional uncertainty of business environment, key drivers of change will be modelled stochastically. For this purpose, all internal and external factors that affect the focal company will be collected. Then, the key drivers of change will be determined considering the interdependencies and relevance of all collected factors. Having identified them, the induced impact and the corresponding parameters in the optimization model are assigned to each driver. Yet, it is often not possible to reliably forecast a future development (cf. 1). Hence, for each key driver of change with uncertain (incl. unknown) future development, multiple developments will be modelled. Subsequently, consistent developments for each discrete decision point of the planning horizon will be clustered to environmental states. Since this is a stochastic, dynamic optimization approach, all clustered parameters as well as the probability of occurrence of the clustered state will be quantified for each state. In the context of the optimization module, the overall objective is the minimization of the expected costs over the planning horizon by finding an optimal policy (cf. 5.3). Again, an optimal policy is a sequence of migration paths over the planning horizon. Since the optimization module is a flexible planning approach, an optimal migration path will be identified for each decision point of the planning horizon and state by applying the optimization model (cf. 5.3). Since costs incurred for the adaptation and operation of a network, both cost elements are considered in this process. Again, both cost elements include cost incurred in production and logistic and they allow the calculation of the total landed cost for a defined planning horizon. A holistic evaluation of the period costs is not limited to production costs but rather considers the total landed cost of operating a production network. Here, the period costs are the sum of the production costs logistic costs. The production costs include overhead costs, technology costs and procurement costs. The logistic costs consist of costs for the physical transportation of material and components from supplier to sites, of semi-finished goods between sites and outsource partners and of finished goods to customers. Furthermore, inventory costs including capital commitment costs and costs for safety stocks as well as customs costs are among the logistics costs. Customs costs include penalty costs of unfulfilled local content requirements. Migration costs occur
for structural and capacitive adaptations of the network configuration. In production, the localization of production stages at sites and the allocation of suitable technologies causes costs. Herein, also the replacement of existing technologies must be considered. Migration costs in logistics include costs for the integration of suppliers and outsource partners. To identify robust migration paths, the stochastic dynamic optimization is implemented in MATLAB by MathWorks® and, the MDP Toolbox will be used to solve the decision problem. In total, a system of contingency plans results including the migration paths for each decision point and environmental state. Hence, the optimal policy for migrating the network configuration is defined by applying the optimization module. 4.2. Module II: Enabler of Change The module concentrates on detailed analyzes for each decision point of the planning horizon. Hence, each discrete decision point is analyzed individually to generate risk-efficient portfolios of change enablers. Having identified robust migration paths for each decision point in the previous module, the demand for change of the network configuration will be determined, initially. This will be executed by comparing the configuration parameters of two consecutive decision points of the planning horizon. Thus, the network objects to be changed can be determined precisely for each decision point and environmental state of the planning horizon. For each network object to be changed, the demand for change is identified and appropriate change enablers (organizational, spatial and technology-based) are collected. However, the implementation of these enablers takes time (adjustment time to the desired effect of the change) and causes investment costs (e.g. initial, facility, replacement or additional investments), direct costs (e.g. costs for changeover, disestablishment and establishment of process capability) and opportunity costs (e.g. production downtimes, extra work or inventory costs) [7],[8],[30]. In case of a proactive implementation, meaning that the enablers of change are implemented before the identified point in time for change, the costs can be reduced. It becomes obvious that the inefficiency and, simultaneously, the costs in case of a proactive course of action can be significantly reduced [14]. But a proactive implementation of an enabler of change for a single object is not free of risk. This is due to the stochastic properties of the key driver of change and the accordingly modelled environmental states (cf. 4.1). Since the demand for change may differ for each state in the planning horizon (also for states of the same decision point) a proactive implementation of change enablers is associated with risk. It may happen, for example, that a network object will only be activated for one state but not for the other states of the same decision point. To consider these risks in the decision about a proactive implementation of change enablers the ߤ െ ߪ principle [17] is applied calculating the return on investment of a proactive implementation. The return on investment is determined by dividing the reduction on migration costs by the sum of the investment cost and direct costs of the change enablers. The
107
E. Moser et al. / Procedia CIRP 57 (2016) 104 – 109
mean ߤ and standard deviation ߪ of the return on investment depends on whether the change enabler is needed in the states or not. Again, this depends on the migration paths of the states. Since the migration paths and the probability of occurrence of the states are already known, the mean ߤ and standard deviation ߪ can be easily calculated. Subsequently and by applying the basic idea of the portfolio theory of Markowitz [19], several enablers of change for different network objects will be bundled to portfolios. The motive for this approach is that the risk of the portfolio is lower than the sum of the individual risks. By implementing several enablers of change for different network objects, the risk can be diversified. Hence, risk efficient portfolios will be generated. For this reason, consistent change enablers are bundled and then assessed by applying the described approach. To ensure the comparability of different portfolios, the difference in costs of each portfolio compared to the most expensive one is then invested risk-neutral into the capital market. Finally, pareto dominated portfolios are eliminated.
the migration process and a concrete implementation plan for the change enablers (proactive and reactive) will be derived. WŽŝŶƚŝŶƚŝŵĞĨŽƌĐŚĂŶŐĞ WƌŽĂĐƚŝǀĞ ZĞĂĐƚŝǀĞ ŶĂďůĞƌϮ ŶĂďůĞƌϭ
ŶĂďůĞƌϯ
ŶĂďůĞƌϰ
ŶĂďůĞƌϱ
Step 1 Step 2 Step 3 Step 4 Step 5 Step 6 Step 7 Step 8
ƚŝŵĞŽĨŵŝŐƌĂƚŝŽŶƉƌŽĐĞƐƐ Fig. 3. Exemplary migration process using a Gantt chart with an implementation plan for change enablers (proactive and reactive).
Finally, concrete migration processes including an implementation plan for the change enablers result from it. The recommended starting point of the migration processes can be derived too by considering the required adjustment time of the change and the available resources of the company. 5. Module formulation
Fig. 2. Risk efficient portfolios of change enablers.
Similar to the optimization module, parts of this module are implemented in MATLAB by MathWorks®. As a result, riskefficient portfolios of change enablers are identified for each decision point. 4.3. Module III: Migration process The final module aims at generating migration processes and takes into account specific requirements of the decision maker. Depending on the preferences of the decision maker (riskseeking, risk-averse or risk-neutral), the portfolios will be recommended in a user-specifical way. Thus, the decision maker can choose the favored portfolio of change enablers for a proactive implementation. Not only the preferences of the decision maker can be respected, but also the available resources (e.g. capital, human resources) of the company. Having chosen the portfolio of proactive change enablers, the migration process will be derived. Considering logical and causal interdependencies of the single process steps, the migration process can be structured and modelled, for example, by applying the metra-potential method (mpm). Among other things, the required time for the migration process can be calculated. Then, based on these analyzes, the starting point of
As part of the optimization module, a stochastic and dynamic optimization model is formulated, whose mathematical formulation and solution method are presented in the following sections. 5.1. Mathematical formulation In general, the decision problem of identifying robust migration paths can be described as a multi-stage decision process and is defined by the planning horizon, the discrete decision (finite) point in time, possible adaptations adjustments and possible states of key drivers of change. Hence, the decision problem can be formulated as a (finite horizon) Markovian Decision Process (MDP) under uncertainty [13]. Below, the MDP is formally described using the standard mathematical formulation. The formulation is partially similar to [15]: Planning horizon ൌ ͳǡ ǥ ǡ ܶ : The planning horizon ܶ represents the planning horizon of migration planning and is finite. Dividing the planning horizon into adequate time slices, a finite number of equidistant decision points is specified. For each point in time, decisions about the adaptation of the network configuration must be made. State Space ܵ௧ : For each point in time ݐ, a finite, non-empty set of states ܵ௧ is defined. A state ݏ௧ ൌ ሺ݇௧ ǡ ݊ሬԦ௧ ሻ ܵ א௧ defines the initial network configuration at the beginning of ݇ ݐ௧ ܭ א௧ and the demand situation ݊ሬԦ௧ ܰ א௧ . The demand for individual
108
E. Moser et al. / Procedia CIRP 57 (2016) 104 – 109
products is summarized in a vector ݊ሬԦ௧ . Thus, it follows:ܵ௧ ൌ ܭ௧ ൈܰ௧ . Since the decision model focusses on brownfield scenarios, the model is initialized with the current network configuration.
Terminal reward function ்ݒሺ݇ ் ሻ : In ݐൌ ܶ the allocated technologies of each site are sold. The revenues which are independent from usage time determine the terminal reward ்ݒሺ݇ ் ሻ.
State Space ܹ௧ : For each point in time ݐ, a finite, non-empty set of states ܹ௧ is defined. A state ݓ ሬሬԦ௧ ൌ ሺݓ௧ଵ ǡ ǥ ǡ ݓ௧ ሻ ܹ אఛ determines key drivers of change for a point in time. For each decision point, one to several different states may exist. These states are generated within a consistency check (cf. 4.1).
5.2. Transformation
Action space ܣ௧ : For each single state ݏ௧ ܵ א௧ , a finite, admissible set of actions ܣ௧ ሺݏ௧ ሻ exists. An action ܽ௧ ሺݏ௧ ሻ א ܣ௧ ሺݏ௧ ሻ can include both, structural and capacitive adaptions of network configuration. By choosing an action ܽ௧ ሺݏ௧ ሻ א ܣ௧ ሺݏ௧ ሻthe initial configuration migrates to configuration ݇௧ାଵ . It is assumed that the time for reconfiguration can be neglected. Hence, the demand in ݐis produced with configuration ݇௧ାଵ . The admissible action space ܣ௧ ሺݏ௧ ሻ is defined implicitly by equality and inequality constraints. Exemplary constraints are: •
Capacity restrictions of sites, suppliers and outsource partners
•
Optional strategic inputs (here: volume flexibility)
Transition law : The transition law defines the probability to reach ݏ௧ାଵ being in ݏ௧ and choosing action ܽ௧ in ݐ. Only the demand ݊ሬԦ௧ in period t is stochastic while the action ܽ௧ is deterministic. Furthermore, it is assumed that demand holds a so-called Markov-property, meaning that the demand in ݐ ͳ only depends on the current demand in ݐ. Thus, it ܲ൫ܵ߬ͳ ൌ ݐݏͳ ȁܵ߬ ൌ ݐܣ ٿ ݐݏൌ ܽ ݐ൯ ൌ ሺܰ߬ͳ ൌ follows:
ሬ݊Ԧݐͳ ȁܰ߬ ൌ ݊ሬԦ ݐሻ
The following section describes how the MDP under uncertainty, which was described previously, is transformed in a (finite horizon) Markovian Decision Process (MDP). This enables the application of backward induction to solve the stochastic, dynamic optimization problem recursively. Referring to [13], the transformation is performed according to the following instructions: It is: ܵመ ൌ ܵൈܹ , Ƹ ǣ ܣ௧ ൈܵመ ՜ ሾͲǡͳሿ and Ƹ ሺݏƸ௧ ǡ ܽ௧ ǡ ݏƸ௧ାଵ ሻ ൌ ሺݏ௧ ǡ ݓ ሬሬԦ௧ ǡ ܽ௧ ǡ ݓ ሬሬԦ௧ାଵ ሻݍሺݓ ሬሬԦ௧ ǡ ݓ ሬሬԦ௧ାଵ ሻ . Thus, it is: ݏƸ௧ ሺݏ௧ ǡ ݓ ሬሬԦ௧ ሻ ܵ אመ and ݏƸ௧ାଵ ሺݏ௧ାଵ ǡ ݓ ሬሬԦ௧ାଵ ሻ ܵ אመ. 5.3. Solution method The optimization module aims at identifying robust migration paths. Hence, for each state ݏƸ௧ an action ܽ௧ must be selected. This is based on a decision function ݂ǣݏƸ௧ ՜ ܽ௧ with ݂௧ ሺݏƸ௧ ሻ א ܣ௧ ሺݏƸ௧ ሻ. Again, an optimal migration strategy is a sequence of decision functions, specifying action ܽ௧ ൌ ݂௧ ሺݏƸ௧ ሻ to be taken in state ݏƸ௧ . The sequence is also named policy and formally described with ߨ ൌ ൫݂ ሺݏƸ ሻǡ ݂ଵ ሺݏƸଵ ሻǡ ǥ ǡ ்݂ିଵ ሺݏƸ ்ିଵ ሻ൯. The overall objective is the minimization of the expected costs over the planning horizon by finding the optimal policy. The optimal actions for all points in time and admissible states are identified by applying Bellmann’s Principle of Optimality [2].
Transition law ݍ: The transition law ݍdefines the probability to reach ܹ௧ାଵ being in ܹ௧ in ݐ. To hold Markov-property future developments of the clustered key drivers of change are modelled as a homogenous markov chain. Thus it follows that ݍሺݐሻ ൌ ሺܹఛାଵ ൌ ݓ ሬሬԦ௧ାଵ ȁܹఛ ൌ ݓ ሬሬԦ௧ ሻ ൌ ሺܹఛାଵ ൌ ݓ ሬሬԦ௧ାଵ ȁܹఛ ൌ ݓ ሬሬԦ௧ ǡ ܹ௧ିଵ ൌ ݓ ሬሬԦ௧ିଵ ǡ ǥ ǡ ܹ ൌ ݓ ሬሬԦ ሻ . Again, since the decision model focusses on brownfield scenarios, the model is initialized with the current condition of the key drivers of change ݓ ሬሬԦ .
It follows for ͳ ݐ൏ ܶ:
One-stage reward function ࢚࢘ : The one-stage reward function rሺ݇௧ ǡ ݊ሬԦ௧ ǡ ݓ ሬሬԦ௧ ǡ ܽ௧ ሻ includes the period costs ݈ܿݐሺ݇௧ ǡ ݊ሬԦ௧ ǡ ݓ ሬሬԦ௧ ǡ ܽ௧ ሻ and the cost for adapting the network configuration ݉ܿሺ݇௧ ǡ ܽ௧ ሻ. Since the reconfiguration time is neglected, it follows:
and finally for ݐൌ ܶ:
ሬሬԦ௧ ǡ ܽ௧ ሻ ൌ ݈ܿݐሺ݇௧ାଵ ǡ ݊ሬԦ௧ ǡ ݓ ሬሬԦ௧ ሻ ݈ܿݐሺ݇௧ ǡ ݊ሬԦ௧ ǡ ݓ
(1)
Migration costs are identified by comparing the configuration parameters of ݇௧ ݏ א௧ and ݇௧ାଵ ݏ א௧ାଵ : ݉ܿሺ݇௧ ǡ ܽ௧ ሻ ൌ ݉ܿሺ݇௧ ǡ ݇௧ାଵ ሻ
(2)
A detailed description of the metrics to calculate the period and migration costs will be omitted.
ܸ௧ ሺݏƸ௧ ሻ ൌ
א ሺ௦Ƹ ሻ
ቐݎሺݏƸ௧ ǡ ܽ௧ ሻ
௦Ƹశభ אௌመశభ
ܲ௧ ሺݏƸ௧ ǡ ܽ௧ ǡ ݏƸ௧ାଵ ሻ (3)
ܸ כ௧ାଵ ሺݏƸ௧ାଵ ሻቑ
்ܸ ሺݏƸ ் ሻ ൌ ்ݒሺ݇ ் ሻ ݏƸ௧ ܵ אመ௧
(4)
By using backward induction, the formulated MDP can be solved. Currently, MATLAB by MathWorks® is applied to identify the migration paths. 6. Conclusion This paper presents a methodology for migration planning in global production networks. The methodology consists of three interlinked modules. First, robust migration paths of the network configuration are identified by applying the presented optimization module. Subsequently, portfolios including several enablers of change are generated and assessed. Doing
E. Moser et al. / Procedia CIRP 57 (2016) 104 – 109
this, risk-efficient portfolios can be identified by taking the risk aversion of the decision maker into account. Finally, concrete migration processes can be derived considering the estimated adjustment time and the available resources. This paper is based on current research performed at wbk Institute of Production Science. Future work aims at more detailed elaboration of the methodology. The focus is laid on the detailed formulation of a multi-objective optimization model to identify optimal portfolios taking the risk aversion of a decision maker into consideration. Furthermore, the application of the methodology in a comprehensive industrial case study is an integral part of further works. Acknowledgements This research has been partially supported by the European Union’s 7th Framework Program Project No: NMP 2013609087, “Shock-robust Design of Plants and their Supply Chain Networks (RobustPlaNet)” and the research project LA2351/31-1 “Bestimmung von Wandlungsbedarf und – zeitpunkt globaler Produktionsnetzwerke: Methodische Unterstützung strategischer Mehrzielentscheidungen” of the German Research Foundation. This is gratefully acknowledged. References [1] Abele, E. (2008). Global production. A handbook for strategy and implementation. Springer, Berlin. ISBN: 978-3-540-71652-5. [2] Bellman, R. (1954). The theory of dynamic programming. Bull. Amer. Math. Soc., Vol. 60, Nr. 6, p. 503–515. [3] CIRP International Academy for Production Engineering Research (2013): Encyclopedia of Production Engineering, Keyword: Optimization in Manufacturing, Springer Reference (www.springerreference.com), Springer, Berlin Heidelberg. [4] Fleischmann, B., Ferber, S. & Henrich, P. (2006). Strategic Planning of BMW’s Global Production Network. Interfaces, Vol. 36, Nr. 3, p. 194–208. [5] Friese, M. (2008): Planung von Flexibilitäts- und Kapazitätsstrategien für Produktionsnetzwerke der Automobilindustrie, Dissertation, Leibniz Universität Hannover. [6] Friedl, T.; Thomas, S.; Mundt, A. (2013). Management globaler Produktionsnetzwerke – Strategie, Konfiguration, Koordination. München: Carl Hanser Verlag 2013 [7] Heger, C.L. (2007), Bewertung der Wandlungsfähigkeit von Fabrikobjekten, Dissertation, Leibniz Universität Hannover [8] Hernandez Morales, R. (2003), Systematik der Wandlungsfähigkeit in der Fabrikplanung, Dissertation, Leibniz Universität Hannover. [9] Jacob, F.(2006): Quantitative Optimierung dynamischer Produktionsnetzwerke, Dissertation, Technische Universität Darmstadt. [10] Kauder, S. (2008). Strategische Planung internationaler Produktionsnetzwerke in der Automobilindustrie, Dissertation, Wirtschaftsuniversität Wien. [11] Klemke, T. (2013), Planung der systematischen Wandlungsfähigkeit von Fabriken, Dissertation, Leibniz Universität Hannover.
[12] Koren, Y. (2010). Globalization and Manufacturing Paradigms. In: The Global Manufacturing Revolution: Product-Process-Business Integration and Reconfigurable Systems. p. 1–40. [13] Lange, V. (2013). Dynamische Kapazitätssteuerung in medizinischen Einrichtungen, Dissertation, Karlsruher Institut für Technologie. [14] Lanza, G., Moser, R. (2012). Strategic Planning of Global Changeable Production Networks. In. Procedia CIRP, Vol. 3, 2012, Pages 257–262 [15] Lanza, G., Peters, S. (2012). Integrated capacity planning over highly volatile horizons. In: CIRP Annals - Manufacturing Technology, Vol. 61, Issue 1, Pages 395–398. [16] Lanza, G., Moser, R. (2014). Multi-objective optimization of global manufacturing networks taking into account multi-dimensional uncertainty. CIRP Annals - Manufacturing Technology, Vol. 63, Issue 1, 2014, p. 397–400. [17] Laux, H. (2007), Entscheidungstheorie, Springer Verlag, Berlin [18] Lin, James T., Chen, Tzu-Li, Chu, Hsiao-Ching (2014). A stochastic dynamic programming approach for multi-site capacity planning in TFTLCD manufacturing under demand uncertainty. International Journal of Production Economics, Vol. 148, p. 21–36. [19] Markowitz, H. M.(1991). Portfolio Selection: Efficient Diversification of Investments. Yale University Press, New Haven & London. [20] McKinsey & Company Inc. (2009). Made in Germany Zukunftsperspektiven für die Produktion in Deutschland, Studie, Düsseldorf. [21] Melo, M.T., Nickel, S., Saldanha-Da-Gama, F. (2009). Facility location and supply chain management – a review. European Journal of Operational Research, Vol. 2, p. 401–412. [22] Nyhuis, P., Reinhart, G., Abele, E., 2008, Wandlungsfähige Produktionssysteme, PZH Produktionstechnisches Zentrum, Garbsen. [23] Reuter, C., Prote, J.P., Schmitz, T., Cremer, S. Ermittlung eines strategischen Migrationsplans für globale Produktionsnetzwerke. In ZWF – Zeitschrift für wirtschaftlichen Fabrikbetrieb, 110/15: 7-8. [24] Schmidt, B. (2011). Gestaltung Globaler Produktionsstrategien. In Wertschöpfung und Beschäftigung in Deutschland, p. 71–84. Springer Verlag, Berlin. [25] Schuh, G., Potente, T., Varandani, R., Schmitz, T. (2014). Global Footprint design based on genetic algorithms – An „Industry 4.0“ perspective. CIRP Annals – Manufacturing Technology, Vol. 63 p. 433436. [26] Stephan, H. A., Gschwind, T., Minner, S. (2010). Manufacturing capacity planning and the value of multi-stage stochastic programming under Markovian demand. Flexible services and manufacturing journal, Vol 22, Issue 3, p. 143–162. [27] Váncza, J.; Monostori, L.; Lutters, D.; Kumara, S.; Tseng, M.; Valckenaers, P.; Van Brussel, H. (2011). Cooperative and responsive manufacturing enterprises. In: CIRP Annals-Manufacturing Technology Vol. 60, Issue 2, p. 797–820. [28] Wagner, C. (2012), Kontinuierliche Gestaltung skalierbarer Produktionsstufen, Dissertation, Leibniz Universität Hannover. [29] Westkämper, E., Zahn, E., 2009, Wandlungsfähige Produktionsunternehmen. Das Stuttgarter Unternehmensmodell, Springer, Heidelberg. [30] Wiendahl, H.P.; Reichardt, J.; Nyhuis, P. (2009), Handbuch Fabrikplanung: Konzept, Gestaltung und Umsetzung wandlungfähiger Produktionsstätten, Carl Hanser Verlag, München. [31] Wiendahl H.-P., Lutz, S. (2002). Production in Networks. CIRP Annals – Manufacturing Technology 51(2):573–586.Van der Geer J, Hanraads JAJ, Lupton RA. The art of writing a scientific article. J Sci Commun 2000;163:51-9.
109