Simulation of matrix-structured manufacturing systems

G Model JMSY-421; No. of Pages 9

ARTICLE IN PRESS Journal of Manufacturing Systems xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Journal of Manufacturing Systems journal homepage: www.elsevier.com/locate/jmansys

Technical Paper

Simulation of matrix-structured manufacturing systems Malte Schönemann ∗ , Christoph Herrmann, Peter Greschke, Sebastian Thiede Technische Universität Braunschweig, Institute of Machine Tools and Production Technology, Chair of Sustainable Manufacturing and Life Cycle Engineering, Langer Kamp 19b, 38106 Braunschweig, Germany

a r t i c l e

i n f o

Article history: Received 16 April 2015 Received in revised form 18 August 2015 Accepted 12 September 2015 Available online xxx Keywords: Intelligent manufacturing systems Agile manufacturing Matrix structures Simulation

a b s t r a c t Increasing product variety, shorter product life cycles, and unknown future demands for each product type are key challenges of manufacturing companies. This paper describes the concept of matrixstructured manufacturing systems (MMS) which aims at providing high operational flexibility and scalability. The main goal of MMS is to eliminate a constant cycle time by providing redundant work stations for same operations as well as a flexible product routing. This enables to avoid starving and blocking and to achieve high system utilization while producing multiple product types with unknown demands and high volumes. The paper explains the main principles, elements, and control strategies of MMS and presents a simulation approach for the evaluation of MMS configurations. A case study shows the application of the simulation approach and how it can be used in the planning of MMS. The results reveal that a MMS configuration can lead to better utilizations of the exemplary manufacturing system in comparison to a sequential assembly line configuration. © 2015 The Society of Manufacturing Engineers. Published by Elsevier Ltd. All rights reserved.

1. Introduction Manufacturing companies in general are opposed to various trends such as growing global competition, more individualized customer demands, new technologies and rapid technological progress, as well as strict environmental regulations. These trends lead to an increase in product variety, shorter product life cycles, uncertain and fluctuating demands, as well as higher cost pressure. The general manufacturing paradigm shifted from mass production to mass customization and more individualized products [1] (see Fig. 1). However, traditional manufacturing systems (MS) configurations such as sequential assembly lines or cellular job shop layouts struggle in handling either a high product variety and fluctuating demands or high production volumes. On the one hand, mixed-model assembly lines are in particular effected by different processing times of different products. The variations in processing times can cause starving or blocking of work stations and lead to an unbalanced utilization of resources [2]. A solution to this balancing problem becomes increasingly difficult if more product variants with significant differences in required work steps and processing times are introduced to the assembly line. On the other hand, cellular job shop layouts are able to handle a high product variety but are not capable of efficiently producing high volumes [3]. The

∗ Corresponding author. Tel.: +49 5313917693. E-mail address: [email protected] (M. Schönemann).

resulting challenge for manufacturing companies is to establish a MS which is able to produce high volumes of products with a high variety and uncertain demands. This requires MS which provide the necessary degree of flexibility and scalability to offer specialized products according to a changing demands. This paper presents an agile MS configuration concept for discrete manufacturing which allows using a sequential material flow structure for high volumes of multiple product types with sequential process steps. The proposed so-called matrix-structured MS (MMS) are able to react to changing market demands, produce a wide range of different products, allow the introduction of new products to the current set of products, be scaled to different output volumes as well as to handle layout and process reconfigurations. This can by achieved by eliminating a constant cycle time and allowing a flexible product routing. As a result, the MMS concept combines the advantages of a sequential flow line (handling high volumes) and a cellular job shop layout (handling high product variety) [4]. In a MMS, work stations are capable of performing different tasks and work packages for different product types. Work packages can be processes on different work stations which enables to avoid starving or blocking. Product instances are transported autonomously through the MMS making decentralized decisions such as selecting the next suitable work station. Consequently the product routing is not predetermined. The selection of a work station can be based on predefined and individual optimization objectives such as short transportation time/distance, short lead

http://dx.doi.org/10.1016/j.jmsy.2015.09.002 0278-6125/© 2015 The Society of Manufacturing Engineers. Published by Elsevier Ltd. All rights reserved.

Please cite this article in press as: Schönemann M, et al. Simulation of matrix-structured manufacturing systems. J Manuf Syst (2015), http://dx.doi.org/10.1016/j.jmsy.2015.09.002

G Model

ARTICLE IN PRESS

JMSY-421; No. of Pages 9

M. Schönemann et al. / Journal of Manufacturing Systems xxx (2015) xxx–xxx

2

Fig. 1. Trend to mass customization according to Koren [1].

time, or high total system utilization. This leads to a dynamic and stochastic system behavior increasing the complexity of planning tasks. As a consequence, production planners need supporting tools for the prediction of the behavior of a MMS. Simulation can help to answer the question whether a MMS configuration is feasible and suitable for the manufacturing of a given set of products. This is important since a MMS configuration differs significantly from traditional assembly lines making it almost impossible to test MMS within an actual production environment without taking the high risks of corrupting the current production. However, it was found that commercial simulation software for manufacturing systems does not offer the required functionality for modeling MMS. This is why the development of a suitable simulation model was pursued. The remainder of this paper provides theoretical background about flexibility in MS, describes the concept of MMS in detail, presents the developed simulation tool for the evaluation of MMS, and demonstrates the application and advantages of a MMS in a case study. 2. Background 2.1. Flexibility in manufacturing systems The challenges caused by high variety and uncertainty has been addressed in industry and research by the development of reconfigurable MS (RMS) and flexible MS (FMS) [5–7]. RMS are scalable in capacity or functionality through reconfiguration of hardware and software in order to be able to respond to a change in market demands or regulations [8,9]. FMS usually provide various forms of flexibility and can react to changes of products or the introduction of new product types. However, FMS are not able to efficiently process multiple product types with altering demands at the same time. Furthermore, RMS and FMS are not able to quickly adapt to disturbances and the dynamic environment of a MS. They have to be stopped in order to react to changes or disturbances [10]. Thus, the responsiveness is limited by the capabilities of the centralized and hierarchical control strategy. At the beginning of the century the term agile manufacturing (AM) was introduced to describe MS which are able to quickly respond to unpredictable changes in the environment [11–13]. In the same context holonic MS (HMS) have been introduced as a new paradigm based on a concept for describing biological and social systems [14]. In a HMS the system elements such as products, work stations, machines, jobs or material handling devices (so-called holons) act as autonomous cooperative agents. HMS are multi-agent systems (MAS) allowing agents to make decentralized decisions [15,16]. This leads to an agile, self-regulating system behavior of HMS which is resistant to disturbances and enables an efficient use of resources [10,14,15,17]. However, there are still many unanswered questions regarding the design and implementation of HMS including the organization of

decentralized control strategies, mechanisms of self-organization, automation applications, and definition of the dynamic MS behavior [14]. The goal of the German Industry 4.0 project initiative – referring to the fourth industrial revolution – is the development of intelligent factories and manufacturing systems by utilizing cyber physical systems (CPS, [18]) and digital technologies (e.g. internet of things) [19]. A similar initiative with the goal of developing the next-generation of collaborating manufacturing systems is the Intelligent Manufacturing Systems Program (IMS) [20]. Enabler toward achieving intelligent factories is the combination of advances in production engineering disciplines with innovative information and communication technologies into CPS. These new technologies will allow connecting machines, products, and supporting devices in order to achieve a self-regulating, self-organizing and robust MS.

2.2. Simulation of manufacturing systems Simulation is an established method in industry and science which aims at representing or imitating a real world system over time [21]. For this purpose, an appropriate model is necessary to represent the system of interest. A comprehensive overview over recent simulation applications in manufacturing and business is provided by Jahangirian [22]. In manufacturing, simulation is widely used as a supporting method in designing, planning, analyzing, and optimizing of manufacturing systems. Negahban et al. provide an up-to-date review about simulation approaches particular for MS design and operation [23]. In this context, simulation enables testing of alternative system designs, control strategies and new system elements without acquiring them or disturbing the real system (supporting “what if” analyzes). Furthermore simulation helps to provide insight to a system with respect to its cause and effect relations and to make the system’s behavior visible to all stakeholders [21,23,24]. In the context of FMS, simulation is further used for scheduling tasks [25] and to generate data for real-time control systems [26]. Simulation approaches in general can be distinguished regarding different characteristics [24]. As an example, simulation models can be static, representing a certain state of a system, or dynamic, showing the development of system over time. Also models can be based on deterministic inputs or include random variables for modeling a stochastic behavior. Furthermore, a differentiation can be made between discrete and continuous simulation models. In discrete models, variables only change at discrete points of time whereas variables in continuous models change continuously. Also possible are combined approaches [21,27]. The common simulation approach for the analysis of MS operation is discrete event simulation (DES) [23]. Passive entities (which represent e.g. products) move through a MS and trigger certain actions at discrete points in time. This approach is typically used for scheduling tasks, capacity planning, and bottle neck identification. Another approach is agent-based simulation (ABS) which focuses on decentralized modeling of individual object behavior in a defined environment [27]. Each system element acts individually based on its inherent logic (usually modeled with state charts) and interacts dynamically with other elements. The principles of the agent-based approach can also be used for modeling, simulation, and control of MS [28–33]. This is especially the case for HMS, in which system elements are considered as autonomous entities interacting with each other [15,34–40]. However, none of these published approaches and tools for HMS was found to be directly suitable for the simulation of MMS regarding all relevant planning problems within the process of MMS development as well as for an easy-to use application in industry.

Please cite this article in press as: Schönemann M, et al. Simulation of matrix-structured manufacturing systems. J Manuf Syst (2015), http://dx.doi.org/10.1016/j.jmsy.2015.09.002

G Model JMSY-421; No. of Pages 9

ARTICLE IN PRESS M. Schönemann et al. / Journal of Manufacturing Systems xxx (2015) xxx–xxx

3. Matrix-structured manufacturing systems The proposed concept of MMS aims at combining the principles of agile HMS with cyber physical systems solutions (e.g. sensors, smart transportation devices, autonomous decision making algorithms, microcomputers, communication protocols) in order to enable the realization of a flexible and scalable MS configuration. This section briefly describes the principles of MMS, the system elements, the layout and operation of MMS, and different control strategies. A more detailed description is given in [4]. 3.1. Principles of MMS The principles underlying the concept of MMS differentiate MMS from traditional MS concepts. These principles relate to the optimization objectives, the system layout and the system’s operation [4]. 3.1.1. Individual cycle time The main goal of the MMS concept is avoiding a constant cycle time. This is motivated by the fact that a constant cycle time leads to an unbalanced utilization of work stations (WSh) in a mix model sequential flow shop (starving and blocking). In MMS, work stations are allowed to have an individual cycle time which might also be different for each product type. This means that work stations forward products as soon as the related process step is completed. In order to achieve a balanced product flow through the MS, the work station for the next work package has to be available. However, since this work station also has its individual cycle time it might still be processing another product. As a result, the problem of starving and blocking would not be eliminated by just allowing work stations to operate on an individual cycle time. 3.1.2. Redundancy of offered work packages In order to avoid blocking of work stations as well as large buffer capacities, MMS can provide more than just one work station for the processing of a specific work package (WP). This increases the likelihood of having a suitable work station available for the processing of a particular work package. Thus, products can choose from the available and suitable work stations. 3.1.3. Multiple work packages per work station In order to avoid starving of works stations, they can be set up to perform multiple work packages by providing the required functions and technologies. This increases the likelihood of having a product requesting one of the offered work packages at this work station. Furthermore it enables reducing the total amount of work stations in a MMS. Of course in the real world it has to be possible to combine the manufacturing equipment required for the work packages into one work station. 3.1.4. Flexible product routing The two previous principles require a flexible material flow through the MMS. This includes an intelligent control strategy for the routing of products. Since there are multiple work stations offering the same work packages, the routing of products is not predetermined. As a result of this variation in routing, the lead time of products is not predetermined but a result of decisions about the selection of work stations. Other challenges aside the control strategy are the physical realization of the material flow between work stations as well as the communication between products and work stations. 3.1.5. Utilization Without a constant cycle time the main objective is a high utilization of MMS. The utilization is an important performance

3

indicators since it strongly determines the economic performance of a manufacturing system. Under the assumption that a resource is only defined as utilized with productive and value-adding activities, it can be concluded that an overall higher utilization leads to a higher production rate. That means that a higher utilization leads to a higher efficiency. The objective of a high overall utilization means that each work station has to be utilized as good as possible. The utilization will be high if the waiting times (starving) and non-productive periods (e.g. blocking or machine breakdowns) at all work stations are low. The lead time and transported distance of a single product or job become secondary objectives since longer lead times or transportation are accepted if they would result in a higher system utilization. However, lead time is still an important indicators which can also be used as an optimization objective. 3.2. Elements of MMS The main elements (holons) of MMS are products and work stations with buffers. The work stations can be placed on the shop floor according to a defined layout. Work stations can have buffers (with a defined capacity) to store products prior to processing in order to reduce idle times of work stations. Another element of MMS is the material flow which is not a physical element but indicates the routing of products through the MMS. Fig. 2 illustrates the four basic elements of MMS. A product type Pi requires a finite sequence of processing steps which are called work packages. Eq. (1) describes the work packages of product type Pi where mi is the total number of work packages of Pi . WP Pi = {WP i1 , WP i2 , . . ., WP imi }

(1)

Since multiple product types can be produced in a MMS, the work stations must be able to process all WP of all product types. As a consequence, at least one work station must be able to process a required work package. A work station is capable of performing a set of work packages for different products. The allocation of work packages to work stations is based on the available manufacturing equipment of the work stations. The allocated work packages of a particular work station is a subset of all existing WP of all product types. Work stations can be in different states. If a work station is active and not processing it is in the idle state. If a work package is currently processed, the work station is in the processing state. If a finished product cannot be forwarded because no work station offering the next work package is available or has remaining buffer capacity, the work station is in the blocked state. If a work station is under maintenance or if a machine breakdown occurred, the work station is in the not available state. The material flow represents the movement of products from the start position through different work stations to the finish position. It can be described by the movement time between two positions which depend on distance and velocity. The distances between the work stations is determined by the coordinates of the work stations in the factory layouts. The velocities of the movements depend for example on the used material handling devices, obstacles, or surface textures and slopes. These four elements can be used for describing the layout and the operation of MMS.

Fig. 2. Basic elements of MMS.

Please cite this article in press as: Schönemann M, et al. Simulation of matrix-structured manufacturing systems. J Manuf Syst (2015), http://dx.doi.org/10.1016/j.jmsy.2015.09.002

G Model JMSY-421; No. of Pages 9

ARTICLE IN PRESS M. Schönemann et al. / Journal of Manufacturing Systems xxx (2015) xxx–xxx

4

Fig. 3. Comparison of a line and a MMS configuration.

3.3. Layout and operation of MMS The layout of MMS is different compared to traditional sequential flow shop layouts. MMS layouts must not provide a unidirectional material flow but allow flexible positioning of work stations and variable transport routes. Fig. 3 shows an exemplary comparison of an assembly line configuration with a MMS. The line configuration consists of four work stations while the MMS configuration consists of nine work stations. In this example both system configurations can be used for the manufacturing of two product types with four work packages each. In the line configuration, each work station can perform one work package for each product type (e.g. WS1 can perform the first work package of product type 1 WP11 and product type 2 WP21 ). In the MMS, each work station can perform two different work packages for each product type. The number and redundancy of WS in this example is very high for only two different product types. That means that for a real case the number of WS could be reduced or the number of product types could be increased while maintaining a high utilization. In the assembly line configuration, a product can only be forwarded if the next work station is available and not processing another product. Assuming the processing at the second work station (WS2 with WP12 and WP22 ) requires a longer processing time than at the first work station (WS1 with WP11 and WP21 ), products cannot be forwarded from WS1 to WS2 since WS2 is still processing the previous product. In this scenario, WS1 is blocked and WS3 would run out of products (starving). This problem could not be avoided with buffers in front of all work stations. It would only be delayed and high buffer capacities would be required. In a MMS, however, this problem of blocking and starting can be avoided since MMS can offer more than one WS for each WP. Consequently, products can choose from more than one suitable work station. This is illustrated in Fig. 4. Based on the exemplary MMS, products are placed into work stations. The product icons present information about the product type (top), the individual ID of the product instance (middle), and the current work package (bottom). In the given example the work package WP22 of product 4 is currently processed on work station WS2 . Product 4 is of type P2 . The work package WP22 of product 3 is currently processed on work station WS4 . At work station WS1 , product 6 was just completed and product 2 was completed at WS5 . Product 7 is currently loaded from the buffer to WS1 . Products 2 and 6 have to decide where to move next, which is indicated by the question mark. The arrows indicate the possible material flows to suitable work stations. The work stations WS6 , WS7 , and WS8 are in idle state and immediately available. WS2 , WS4 , and WS9 are currently processing and could only offer a spot

Fig. 4. MMS scenario for illustration of MMS operation.

in the buffers. WS3 is not available. The selection of a next suitable work station depends on the implemented control strategy. 3.4. Control strategies for MMS The MMS control is responsible for the routing of products through the MMS. The most important type of control decision relates to the selection of work stations for each work package of each product. This decentralized decision situation is indicated in Fig. 4 by the question mark. This situation occurs whenever a work package of a product is completed. Then the product has to select the next work station for the following work package. This decision can be based on different criteria which depend on the selected control strategy. It is made by each product instance in a decentralized manner and in a real world application it could be determined for example with a microcomputer within the product handling unit. Information about work stations (e.g. time till availability) could be provided via RFID technology or via a centralized signal from a server which collects the data from the work stations. The biggest advantage of decentralized decision is that changes and modifications regarding the decision functions of individual product instances can be implemented much easier compared to an integrated and holistic centralized control system. For example, it is possible to define different and new decision criteria for different product types or product instances without having to modify the central control logic. Furthermore, a decentralized structure improves the scalability of a MS since adding new work stations does not require an update of defined control strategies. Examples of different control strategies and criteria are shortest distance between work stations or shortest time until processing. 3.4.1. Shortest distance The control strategy shortest distance aims at reducing the distance a product has to travel through the MMS. After the completion of a work package a product selects the next suitable work station based on the distance of the material flow. Not considered in this strategy is the overall lead time of a product (or job) or the waiting times at work stations. Furthermore, the strategy might be favorable if the cost of transportation is high. 3.4.2. Shortest time until processing The control strategy shortest time aims at short waiting times for each product resulting in short lead times and also a high system utilization. A product will select a suitable work station which offers

Please cite this article in press as: Schönemann M, et al. Simulation of matrix-structured manufacturing systems. J Manuf Syst (2015), http://dx.doi.org/10.1016/j.jmsy.2015.09.002

G Model JMSY-421; No. of Pages 9

ARTICLE IN PRESS M. Schönemann et al. / Journal of Manufacturing Systems xxx (2015) xxx–xxx

the processing of the next work package in the shortest period of time. This includes the time for the movement to a work station. As a result, the idle times of work stations are reduced. According to this control strategy, first all work stations are checked whether they can process the next required work package. Next, all suitable work stations are analyzed regarding the time until each station could start processing the product in question. This time until processing depends on the state of the work station (e.g. idle or processing), the amount of other products in the buffer, and on other products having reserved the work station. Also to be considered is the distance and velocity of the movement of the product to the station. If a work station is currently in idle state and it is not reserved by any other product, the time till processing is only determined by the material flow respectively the distance between the product and the WS and the possible velocity of the product. If a work station is currently processing another product, the remaining processing time as well as the processing times of products in the buffer and the times for loading of a product from the buffer has to be considered. If a work station is reserved by another product, the time till processing is at least the time until the prior product is finished. The times until processing of all suitable work stations are compared and the station with the shortest time until processing is selected. If the time of the movement to the station is shorter than the remaining processing time of prior products plus the time for products in the buffer, the product is sent to the buffer. If the time for the movement is longer than the sum of times for products inside the work station and the buffer, the product is routed directly to the station. These two strategies can be explained based on the example in Fig. 4. Product 6 at WS1 has to select a next work station. If the objective is to select the work station with the shortest distance, the product 6 would select WS2 . If the objective is to select the work station with the shortest time until processing, it has to be determined if one of the transportation times to WS6 or WS8 is shorter than the remaining processing time at WS2 or the remaining processing time plus the processing time of product five in the buffer of WS4 . WS5 would also be an option if product 2 would have left the work station already. This example shows that the strategy shortest time until processing requires detailed information about the state of each work station. 3.5. Summary and discussion The concept of MMS addresses the practical implementation of HMS. Based on defined principles and system elements the MMS concept provides a theory for designing and operating intelligent and agile MS. Advantages of MMS are the capability of processing multiple products with large differences in processing times of work packages while maintaining high utilization, high scalability through adding or removing WS depending on the product demand, robustness regarding machine break downs, and the ability to utilize resources with reduced performance due to the individual cycle time of work stations. This means that it is possible to employ resources with lower performance or handicapped workers without causing blocking or starving of other work cells [41]. Also it is not necessary to perform a sequencing of products in order to predefine a specific order of product types. One major challenge of MMS is the realization of the physical flexible material flow between work stations and a logistic concept for material supply. This is in particular true for large products such as cars. Consequently MMS are more suitable for components of products with a high variety such as automotive components (e.g. battery systems, engines) or electronic devices. Another drawback of MMS is that the redundancy of work stations and resources may increase the investment and space requirements for MS. However, one MMS may replace several sequential line MS configurations each producing one product type leading to reduced total investment and

5

operating cost. These tradeoffs and the return of investment have to be evaluated in the process of MMS development [41]. 4. Simulation of MMS MMS are highly dynamic and stochastic systems. These characteristics require production planners to use suitable decision support tools for the planning and evaluation of MMS configurations (e.g. work station layout, logistic concept, allocation of work packages to work stations). Using mathematical optimization methods seem to be not practical due to the strong dynamic interactions of system elements and stochastic effects during MMS operation. More suitable are simulation based tools to analyze the system behavior of MMS and to evaluate the technical and economical feasibility of MMS for a given set of products. More specifically, simulation allows to analyze the effects of decisions (e.g. about the system layout or the allocation of work packages to work stations) on determined performance indicators of a MMS over time. This means that simulation is an important step in the approach of MMS development. Fig. 5 presents the steps of the development approach starting with the analysis of product characteristics and finishing with the realization of a MMS. The simulation concept presented in this paper was developed to support production planners during the development of MMS. It enables to predict the behavior of a planned MMS, evaluate the feasibility of the MMS concept and to determine the optimal system parameters (e.g. number and layout of WS or allocation of WP to WS). 4.1. Developed MMS simulation model architecture A simulation model for MMS has to represent the operation of a MMS considering the system layout, capabilities of WS, buffer sizes, WP of multiple product types, allocations of WP to WS, flexible product routing, different control strategies, and material supply. Also demanded is a dynamic visualization of the MMS operation as well as the presentation and analysis of resulting performance indicators. The two latter functions are important to involve all stakeholder in an industrial environment and to generate results being directly usable in the planning of MMS. The simulation of MMS requires an approach that allows considering the dynamic system behavior and decentralized decision making of individual products or jobs. Furthermore, the capabilities of each WS (offers processing) have to be considered and matched with the demands of products (needs processing). This is not possible with a solely event based discrete process chain simulation since it would not allow to specify properties of single product instances and their individual behavior. In traditional process chain simulation, all products are represented by occurring events such as arrival at work stations which trigger certain actions (e.g. start of a process). Such simulation does not allow describing the decentralized decision making of autonomous acting product instances

Fig. 5. Process of MMS development and realization.

Please cite this article in press as: Schönemann M, et al. Simulation of matrix-structured manufacturing systems. J Manuf Syst (2015), http://dx.doi.org/10.1016/j.jmsy.2015.09.002

G Model JMSY-421; No. of Pages 9 6

ARTICLE IN PRESS M. Schönemann et al. / Journal of Manufacturing Systems xxx (2015) xxx–xxx

Fig. 7. Simplified logic of product flow and search function.

Fig. 6. Classes of the simulation model.

or product handling units. These shortcomings lead to an agent based approach for simulation of MMS in combination with DES. That means that products and work stations (including buffers) are modeled as agents with individual properties which interact at discrete points in time. Fig. 6 presents the main object classes of the developed simulation model with exemplary attributes and functions. The MMS environment contains a defined number of WS and products which have to be produced. Also defined is the maximum number of products that can be processed at the same time within the MMS. This is important in order to avoid overloading of the system which can result in blockings. This parameter depends on the number of available WS and has to be adapted in case of WS breakdowns. The main attributes of a WS are the WP it can process, the processing times of each WP, the buffer size, as well as the x and y position for the location in the environment. Furthermore, each WS contains functions to calculate the current time until availability and its utilization. The latter value is used to determine the utilization of the entire MMS. The main attributes of a product are the product type, the total number and order of WP, the current progress of production (current WP), the priority of the job, and the velocity of the movements through the MMS (e.g. on a transportation device). Each product can call a search function to determine the best suited WS for processing the next WP. This function determines the capable WS for the next WP and selects the best suited WS depending on the control strategy. If two products select the same work station at the same time, priority rules can be applied to resolve this conflict. Within a simulation run, products are transported through the MMS and trigger certain events at discrete points of time. For example, as soon as a product instance is fed into the MMS it calls the search function for selecting the next WS. After this decision, the product moves either directly to the selected WS or to the related buffer. If a product moves to a WS, the WS becomes reserved for this product. This is necessary because other products have to know that the WS is not immediately available. The product informs the WS about its arrival via a message and the processing starts immediately after the arrival. When the processing is completed, the product selects a WS for its next WP. If

Fig. 8. Simplified state-chart of a WS.

all WP are completed, the product moves to the end position. Fig. 7 illustrates the logic of the product flow and the search function. The behavior of a WS is modeled by defining all possible states and the transitions between these states. Fig. 8 presents a simplified illustration of the state-chart of a WS. The most important states are idle, processing, blocked and failure. The two latter states imply that a WS is not available. Other concurrent states describe the buffer usage (empty, used, full) and if a WS is reserved for other products. The reversed state is activated if a WS is selected by a product. This state is required to include products in the decision making which are currently transported and have not yet arrived at a work station. The possible number of reservations is not limited and the transportation time until the WS of each reserved product is considered in the calculation of the time until the WS will be available again. When a product arrives at an idling WS, the WS state shift from idle to processing and – if no other product is on its way to this WS – from reserved to not reserved. After the processing is completed, the product searches a next WS. If the product cannot find another available WS, it is checked whether the next WP of the product can be also processed at the current WS. If this is not the case, the WS is blocked by the product which cannot move anywhere else. As soon as another WS becomes available, the product leaves the WS and the WS processes the next product from the buffer or shifts into idle state. If a work station is also able to process the next work package of this blocking product, it will process next work package and redirect products with reservations to its buffer. If a work station with reservations becomes blocked, the products with reservations search another available WS and move to this WS or its buffer.

Please cite this article in press as: Schönemann M, et al. Simulation of matrix-structured manufacturing systems. J Manuf Syst (2015), http://dx.doi.org/10.1016/j.jmsy.2015.09.002

G Model JMSY-421; No. of Pages 9

ARTICLE IN PRESS M. Schönemann et al. / Journal of Manufacturing Systems xxx (2015) xxx–xxx

7

Important indicators describing the behavior and performance of the MMS are determined during the simulation. Examples are the utilization of each WS and the entire system, the number of blockings per WS, the WP causing blockings, the lead time of each product, and the average lead time of all products. These indicators can be used to determine the efficiency of a MMS configuration as well as the required layout (e.g. numbers of WS and allocation of WP) in order to achieve a high utilization for a given set of product. Not considered in the current model are physical routes of the product flows and the material delivery (e.g. defined paths or obstacles).

Table 1 Processing times tp for WP of P1 and P2 .

4.2. Model implementation and configuration

all WS are placed in the defined order (WS1 → WS8 ). Each WS has to have the required resources (e.g. tools) since it is not intended to use a WS for more than one WP. If multiple WP require the same resources, these resources have to be available redundantly. For the MMS, the WP have to be allocated to WS according to a specific planning algorithm (not presented in this paper). For example, if multiple WP require the same resources, these WP can be allocated to the same WS. This allows to use expensive resources for both product types resulting in reduced investment. Both MS configurations are modeled with the developed simulation tool. Figs. 9 and 10 show screenshots of both the simulated line and MMS configurations. The utilizations of both MS configurations were analyzed by conducting individual simulation runs in which the manufacturing of 1000 products is simulated with a random distribution of product types without an initial sequencing (fix seed value). This random occurrence of jobs should imitate the worst possible case of unknown demands and without a prior sequencing or sorting of jobs. The stochastic occurrences of machine failures were not included at first. In the first two runs, the line configuration was analyzed with a buffer size (BS) of zero and one respectively. The resulting utilization of the line configuration is presented in Fig. 11. The average utilization of the line configuration with BS = 1 and BS = 0 including rampup and rampdown is around 57%. The plot shows a strong fluctuation of the utilization and that the 100% mark is not reached. This effect is caused by frequent starving and blocking of work stations. Overall this is not a desired system behavior since a lot of WS are not highly utilized. The resulting utilization of the MMS simulation (with BS = 1) is presented in Fig. 12. The utilization of the MMS is higher overall compared to the line configuration. The average utilization is 86.9%. Furthermore the fluctuation of the utilization is much lower

The developed simulation concept was implemented in the software AnyLogic® . This software allows combining agent based and discrete event simulation within one model. Furthermore, the software provides various elements for graphical representation of model elements during a simulation run (see Fig. 10). The state of WS is represented by the background color of a WS icon. The black horizontal bar indicates the progress of the current WP and the green vertical bar indicates the utilization of the WS. The model can be configured regarding the number and the layout of WS, the definition of product types and the required work packages, the allocation of work packages to work cells, the maximum number of products in the system and the desired control strategy, as well as various other parameters defining the system design and operation. The configuration of the model is based on spreadsheets which can be modified with MS Excel®. 5. Concept application This case study presents the application of the developed simulation approach for the comparison of a traditional assembly line configuration and a matrix-structured layout with respect to the system utilization. Both configurations are suitable for the manufacturing of large product volumes with a constant product unit flow. In the given scenario, both MS configurations are assumed to produce two product types P1 and P2 in a random order. P2 requires seven WP while P1 requires these seven plus one additional WP (WP15 ). Both product types have individual processing times for each WP (see Table 1). Prior to the actual simulation, the WP have to be allocated to WS. For the line configuration, each WP is allocated to one WS and

WP

tp of P1 (min)

tp of P2 (min)

1 2 3 4 5 6 7 8

6 5 7 5.5 8 6 10.5 4

6 4 6.5 5.8 6 10 4 –

Fig. 9. Simulation visualization of line configuration.

Fig. 10. Simulation visualization of MMS configuration.

Please cite this article in press as: Schönemann M, et al. Simulation of matrix-structured manufacturing systems. J Manuf Syst (2015), http://dx.doi.org/10.1016/j.jmsy.2015.09.002

G Model JMSY-421; No. of Pages 9 8

ARTICLE IN PRESS M. Schönemann et al. / Journal of Manufacturing Systems xxx (2015) xxx–xxx

Fig. 11. Utilization of the line configuration (excerpt).

Fig. 12. Utilization of the MMS configuration (excerpt).

and an utilization of 100% was reached temporarily. For the given scenario, the MMS is advantageous compared to the line configuration. However, optimization of the system design might lead to further improvements. A histogram of WP causing blockings can provide indications about which WP are not sufficient represented within the system. Fig. 13 presents this histogram for the initial MMS configuration. In the given scenario, WS often got blocked because products could not be forwarded to WP7 . This effect is identical for both product types. Enabling one other WS to also perform WP7 may resolve this blocking issue. This improvement measure is analyzed within a simulation run and the new resulting histogram is shown in Fig. 14. The figure shows that adding WP7 to another WS leads to less overall blockings. These results in and a slightly higher total utilization of the system (87.5%). However, since the improvement is rather small (0.5%), it is important to evaluate the benefit and the higher investments for the new required equipment. Further simulation runs were performed in order to analyze the robustness of the two configurations regarding machine failures. It is assumed that the robustness of the MMS is higher and the impact of failures on the utilization lower compared to the line configuration. To support this assumption, variables were implemented for the mean time to failure (MTTF) and the mean time to repair (MTTR)

Fig. 13. WP beeing blocked.

Fig. 14. WP beeing blocked, after improvement.

of each work station. These variables are determined based on probability distributions. In general, any distributions can be used in order to describe the individual behavior of the modeled resources. In this case, a Weibull function is used for modeling the MTTF since it is flexible and can describe different distributions depending on the defined shape and scale parameters [42,43]. The MTTR was described by uniform distributions in order to equally account short and long repair times. This modified simulation model was used to perform 100 runs for each configuration with random seed values. The histograms of the utilizations are shown in Fig. 15. The results show that the MMS configuration still achieves a higher utilization due to the redundancy of work stations and the adaptive control strategy. However, the results also show that the deviation of the utilization is larger compared to the line configuration. This may be explained by the fact that the WS in the simulated line configuration are already often blocked so that machine failures are not causing a significant difference. Overall, the case study has shown the general feasibility of the MMS concept and that it can lead to better utilizations of manufacturing systems. The simulation further provides results which can be used for an economic comparison of the MS configurations. On the one hand, the MMS configuration achieves a higher utilization and provides higher flexibility compared to the assembly line configuration. MMS might also produce a higher number of different product types which may not be possible on a sequential assembly line. On the other hand, it requires more work stations and technologies for data acquisition, information exchange, decision making and flexible product handling. This leads to a higher required investment compared to the assembly line configuration. The analysis of overall advantage of different configurations and control strategies can be further supported by the simulation model. It can be used to determine the output during a longer time period and it can show the effects of adding or removing work stations to or from the different MS configurations.

Fig. 15. Utilizations of line and matrix configuration.

Please cite this article in press as: Schönemann M, et al. Simulation of matrix-structured manufacturing systems. J Manuf Syst (2015), http://dx.doi.org/10.1016/j.jmsy.2015.09.002

G Model JMSY-421; No. of Pages 9

ARTICLE IN PRESS M. Schönemann et al. / Journal of Manufacturing Systems xxx (2015) xxx–xxx

6. Conclusion and outlook MMS can be a promising case for the application of decentralized control solutions, which are motivated in the research field of intelligent manufacturing systems and cyber physical systems. MMS enable to achieve a high system utilization while producing multiple product types with unknown demands. The developed simulation approach enables production planners to design and evaluate MMS configurations regarding different objectives. The agent based structure of the simulation model allows considering individual product characteristics of product instances which is an extension to traditional event-driven process chain simulations. This functionality can also be used for considering for example the influence of processes on product parameters or the embodied energy of each product. Further research will deal with the developing and improvement of algorithms for the MMS layout and for the allocation of work packages to work stations. The developed simulation model should be extended toward the consideration of transportation equipment and material supply logistics. Also, the integration of energy demands of manufacturing equipment seems promising since the principles of MMS offer the opportunity for intelligent energy efficient operation strategies. This includes but is not limited to routing of products to less energy intensive equipment or shutdown of redundant and not required equipment. Furthermore, important is the economic evaluation of MMS in comparison with traditional manufacturing system configurations. This evaluation has to answer the question whether a higher invest in a higher number of work stations and redundant equipment required for MMS is compensated by a higher utilization and by the fact that a MMS is capable of manufacturing multiple product types. References [1] Koren Y. The global manufacturing revolution: product–process–business integration and reconfigurable systems. Wiley; 2010. [2] Halubek P, Herrmann C. Design of mixed model assembly lines – simulation based planning support. In: 44th CIRP Conf Manuf Syst. 2011. [3] Hayes RH, Wheelwright SC. Link manufacturing process and product life cycles. Harv Bus Rev 1979;(January–February):133–40. [4] Greschke P, Schönemann M, Thiede S, Herrmann C. Matrix structures for high volumes and flexibility in production systems. Procedia CIRP 2014;17:160–5, http://dx.doi.org/10.1016/j.procir.2014.02.040. [5] Putnik G, Sluga A, ElMaraghy H, Teti R, Koren Y, Tolio T, et al. Scalability in manufacturing systems design and operation: State-of-the-art and future developments roadmap. CIRP Ann Manuf Technol 2013;62(2):751–74, http://dx.doi.org/10.1016/j.cirp.2013.05.002. [6] Koren Y, Shpitalni M. Design of reconfigurable manufacturing systems. J Manuf Syst 2010;29(4):130–41, http://dx.doi.org/10.1016/j.jmsy.2011.01.001. [7] Nazarian E, Ko J, Wang H. Design of multi-product manufacturing lines with the consideration of product change dependent inter-task times, reduced changeover and machine flexibility. J Manuf Syst 2010;29(1):35–46, http://dx.doi.org/10.1016/j.jmsy.2010.08.001. [8] Wang W, Koren Y. Scalability planning for reconfigurable manufacturing systems. J Manuf Syst 2012;31(2):83–91, http://dx.doi.org/10.1016/j.jmsy.2011.11.001. [9] Bi ZM, Lang SYT, Shen W, Wang L. Reconfigurable manufacturing systems: the state of the art. Int J Prod Res 2008;46(4):967–92, http://dx.doi.org/10.1080/00207540600905646. [10] Park H-S, Tran N-H. An autonomous manufacturing system based on swarm of cognitive agents. J Manuf Syst 2012;31(3):337–48, http://dx.doi.org/10.1016/j.jmsy.2012.05.002. [11] Gunasekaran A. Editorial Design and implementation of agile manufacturing systems. Int J Prod Econ 1999;62:1–6. [12] Langer G, Alting L. An architecture for Agile Shop Floor control systems. J Manuf Syst 2000;19(4):267–81, http://dx.doi.org/10.1016/S0278-6125(01)80006-6. [13] Elkins Da, Huang N, Alden JM. Agile manufacturing systems in the automotive industry. Int J Prod Econ 2004;91(3):201–14, http://dx.doi.org/10.1016/j.ijpe.2003.07.006.

9

[14] Leitão P, Restivo F. ADACOR: a holonic architecture for agile and adaptive manufacturing control. Comput Ind 2006;57(2):121–30, http://dx.doi.org/10.1016/j.compind.2005.05.005. [15] Dominici G. The holonic production system: a multi agent simulation approach. iBusiness 2010;02(03):201–9, http://dx.doi.org/10.4236/ib.2010.23025. [16] Gerber C, Siekmann J, Vierke G. Holonic multi-agent systems. Tech. Rep. 681, Deutsches Forschungszentrum für Künstliche Intelligenz GmbH, Kaiserslautern; 1999. [17] Leitão P, Restivo FJ. Implementation of a Holonic Control System in a Flexible Manufacturing System. IEEE Trans Syst Man Cybern C: Appl Rev 2008;38(5):699–709. [18] The National Institute of Standards and Technology, Cyber-Physical Systems Homepage, URL: http://www.nist.gov/cps/. [19] Kagermann H, Wahlster W, Helbig J. Recommendations for implementing the strategic initiative INDUSTRIE 4.0. Tech. Rep., April, Acatech – National Academy of Science and Engineering; 2013. [20] Global Research and Business Innovation Program, Intelligent Manufacturing Systems, URL: www.ims.org/. [21] Banks J, Carson JS, Nelson BL, Nicol DM. Discrete-event system simulation. 5th ed. Upper Saddle River, NJ: Prentice Hall; 2010. [22] Jahangirian M, Eldabi T, Naseer A, Stergioulas LK, Young T. Simulation in manufacturing and business: a review. Eur J Oper Res 2010;203(1):1–13, http://dx.doi.org/10.1016/j.ejor.2009.06.004. [23] Negahban A, Smith JS. Simulation for manufacturing system design and operation: literature review and analysis. J Manuf Syst 2014;33(2):241–61, http://dx.doi.org/10.1016/j.jmsy.2013.12.007. [24] Law AM. Simulation modeling and analysis. 4th ed. Boston: McGraw-Hill; 2007. [25] Chan FTS, Chan HK. A comprehensive survey and future trend of simulation study on FMS scheduling. J Intell Manuf 2004;15:87–102. [26] Shirazi B, Mahdvi I, Solimanupur M. Development of a simulation-based intelligent decision support system for the adaptive real-time control of flexible manufacturing systems. J Softw Eng Appl 2010;03(07):661–73, http://dx.doi.org/10.4236/jsea.2010.37076. [27] Borshchev A, Filippov A. From system dynamics to agent based modeling. In: 22nd Int Conf Syst Dyn Soc. 2004. [28] Bagdasaryan A. Discrete dynamic simulation models and technique for complex control systems. Simul Model Pract Theor 2011;19(4):1061–87, http://dx.doi.org/10.1016/j.simpat.2010.12.010. [29] Leit ao P, Barbosa J, Trentesaux D. Bio-inspired multi-agent systems for reconfigurable manufacturing systems. Eng Appl Artif Intell 2012;25(5):934–44, http://dx.doi.org/10.1016/j.engappai.2011.09.025. [30] Leitão P. Agent-based distributed manufacturing control: a statesurvey. Eng Appl Artif Intell 2009;22(7):979–91, of-the-art http://dx.doi.org/10.1016/j.engappai.2008.09.005. [31] Monostori L, Váncza J, Kumara SRT. Agent-based systems for manufacturing. CIRP Ann Manuf Technol 2006;55:697. [32] Gössinger R. Opportunistische koordinierung bei der werkstattfertigung. Gabler Verlag; Dt. Univ-Verlag; 2000. [33] Baumgärtel H, Bussmann S, Klosterberg M. Multi-agent coordination of material flow in a car plant. In: 2nd Conf Pract Appl Intell Agents Multi-Agent Technol. 1997. p. 227–36. [34] Jana TK, Bairagi B, Paul S, Sarkar B, Saha J. Dynamic schedule execution in an agent based holonic manufacturing system. J Manuf Syst 2013;32(4):801–16, http://dx.doi.org/10.1016/j.jmsy.2013.07.004. ˚ V, Vrba P, Leit ao P, editors. Holonic and multi-agent systems for manu[35] MaAík facturing, Vol. 6867 of Lecture Notes in Computer Science. Berlin, Heidelberg: Springer; 2011., http://dx.doi.org/10.1007/978-3-642-23181-0. [36] Barbosa J, Leitao P. Simulation of multi-agent manufacturing systems using Agent-Based Modelling platforms. In: 9th IEEE Int Conf Ind Informatics. 2011. p. 477–82, http://dx.doi.org/10.1109/INDIN.2011.6034926. [37] Bal M, Hashemipour M. Implementation of holonic scheduling and control in flow-line manufacturing systems: die-casting case study. Prod Plan Control 2011;22(2):108–23, http://dx.doi.org/10.1080/09537281003664983. [38] Sallez Y, Berger T, Raileanu S, Chaabane S, Trentesaux D. Semi-heterarchical control of FMS: from theory to application. Eng Appl Artif Intell 2010;23(8):1314–26, http://dx.doi.org/10.1016/j.engappai.2010.06.013. [39] Cardin O, Castagna P. Using online simulation in Holonic mansystems. Eng Appl Artif Intell 2009;22(7):1025–33, ufacturing http://dx.doi.org/10.1016/j.engappai.2009.01.006. [40] Sallez Y, Berger T, Trentesaux D. A stigmergic approach for dynamic routing of active products in FMS. Comput Ind 2009;60(3):204–16, http://dx.doi.org/10.1016/j.compind.2008.12.002. [41] Greschke P, Herrmann C. Das Humanpotenzial einer taktunabhängigen Montage. ZWF –Zeitschrift für wirtschaftlichen Fabrikbetr 2014;2014(10):687–90. [42] Abernethy RB. The new Weibull handbook – reliability & statistical analysis for predicting life, safety risk, support costs failures and forecasting warranty claims. 5th ed. North Palm Beach, FL; 2006. [43] Birolini A. Reliability engineering – theory and practice. Berlin, Heidelberg: Springer; 2010.

Please cite this article in press as: Schönemann M, et al. Simulation of matrix-structured manufacturing systems. J Manuf Syst (2015), http://dx.doi.org/10.1016/j.jmsy.2015.09.002