Product-Driven Generation of Action Sequences for Adaptable Manufacturing Systems

Proceedigs of the 15th IFAC Symposium on Proceedigs of the 15th IFAC Symposium on Information Control Problems in Manufacturing Proceedigs the IFAC on Proceedigs of of the 15th 15th IFAC Symposium Symposium on Information Control Problems in Manufacturing Available online at www.sciencedirect.com May 11-13, 2015. Ottawa, Canada Information Control Problems in Manufacturing Information Control Problems in Manufacturing May 11-13, 2015. Ottawa, Canada May 11-13, 11-13, 2015. 2015. Ottawa, Ottawa, Canada Canada May

ScienceDirect

IFAC-PapersOnLine 48-3 (2015) 1502–1508

Product-Driven Generation of Action Product-Driven Generation of Action Product-Driven Generation of Action Sequences for Adaptable Manufacturing Sequences for Adaptable Manufacturing Sequences for Adaptable Manufacturing Systems Systems Systems Nadine Nadine Nadine Nadine

Keddis Gerd Kainz Alois Zoitl Keddis Gerd Kainz Alois Zoitl Keddis Keddis Gerd Gerd Kainz Kainz Alois Alois Zoitl Zoitl fortiss GmbH fortiss GmbH fortiss Guerickestr. 25 fortiss GmbH GmbH Guerickestr. 25 Guerickestr. 25 80805Guerickestr. Munich, Germany 25 80805 Munich, Germany 80805 Munich, Germany {keddis, kainz, zoitl}@fortiss.org 80805 Munich, Germany {keddis, kainz, zoitl}@fortiss.org {keddis, {keddis, kainz, kainz, zoitl}@fortiss.org zoitl}@fortiss.org Abstract: In times of fast changing markets and short product life-cycles manufacturing Abstract: times of fast changing markets short product life-cycles manufacturing Abstract: In times of markets and short life-cycles manufacturing systems haveIn be adaptable and able to supportand a variety in products and product volumes. Abstract: Into times of fast fast changing changing markets and short product product life-cycles manufacturing systems have to be adaptable and able to support a variety in products and product systems have have totobe bebeadaptable adaptable and able able to supportbetween a variety varietydifferent in products products and should product volumes. Production hasto product-driven andto switching products bevolumes. possible systems and support a in and product volumes. Production has product-driven and between different products should be Production has to to be be product-driven and switching switching between different products should that be possible possible with little manual intervention. We suggest an action sequence generation approach tailors Production has to be product-driven and switching between different products should be possible with little manual intervention. We suggest an action generation approach that tailors with little manual intervention. We suggest an action sequence generation approach that tailors the control programs of resources to product needs.sequence The approach requires a model of the with little manual intervention. We suggest an action sequence generation approach that tailors the control programs of resources to product needs. The approach requires a model of the the control programs of resources to product needs. The approach requires a model of available resources with their capabilities and internal material flow, the material flow between the control programs oftheir resources to product needs. material The approach requires a model of the the available resources with capabilities and internal flow, the material flow between available resources resources with their capabilities capabilities and internal material flow, the material flow between between resources as well aswith a product description. Aninternal action material sequenceflow, can the thenmaterial be generated out of available their and flow resources as well as a product description. An action sequence can then be generated out of resources as a description. An sequence can generated out of these models and as later translated into an executable sequence. The be action sequence resources as well well as a product product description. An action actionaction sequence can then then be generated outcan of these models and later translated into an executable action sequence. The action sequence can these models and later translated into an executable action sequence. The action sequence can be automatically downloaded and executed on a resource. The approach is evaluated on an these models and later translated into an executable action sequence. The action sequence can be automatically downloaded aa resource. The approach is evaluated on an be downloaded and executed on educational production system and withexecuted industrialon be automatically automatically downloaded and executed oncomponents. a resource. resource. The The approach approach is is evaluated evaluated on on an an educational production system with industrial components. educational production production system system with with industrial industrial components. components. educational © 2015, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved. Keywords: Flexible manufacturing systems, Information technology, Machine code, Keywords: Flexible manufacturing systems, Information Information technology, Machine Machine code, Keywords: manufacturing systems, Model-based control, Planning Keywords: Flexible Flexible manufacturing systems, Information technology, technology, Machine code, code, Model-based control, Planning Model-based Model-based control, control, Planning Planning 1. INTRODUCTION This paper proposes an approach for automatic generation 1. INTRODUCTION INTRODUCTION This papersequences proposes an an approach for automatic automatic generation 1. This paper proposes for of action forapproach production resources generation to handle 1. INTRODUCTION This papersequences proposes an approach for automatic generation of action for production resources toProduchandle of action sequences for production resources handle current problems in the manufacturing domain.to of action sequences for production resources handle current problems in the the manufacturing domain.toProducProduccurrent problems in manufacturing domain. tion resources in this context refer to hardware entities current problems in the manufacturing domain. ProducAfter the peak of mass production in 1955 there has been tion resources inproduction this context context refer toare hardware entities thetowards peak of of mass mass production production in 1955 1955 there Institute, has been been tion resources in this refer hardware entities that execute steps andto called resources tion can resources inproduction this context refer toare hardware entities After peak mass in there has aAfter shiftthe customization (Iacocca After the peak of mass production in 1955 there has been that can execute steps and called resources a shift shiftHutowards towards mass customization (Iacocca Institute, can execute production steps and resources throughout this paper. Resources haveare tocalled provide predethat can execute production steps and are called resources a1991; Institute, et al., mass 2008; customization Koren, 2010). (Iacocca The revolution of that a shift towards mass customization (Iacocca Institute, throughout this that paper. Resources have to provide provide prede1991; Hu et al., 2008; Koren, 2010). The revolution of throughout this paper. Resources have to predefined interfaces give access to available control code throughout this paper. Resources have to provide prede1991; Hu et al., 2008; Koren, 2010). The revolution of manufacturing is continuing with2010). a trend towards person1991; Hu et al., 2008; Koren, The revolution of fined fined interfaces interfaces that that give access access to available available control code manufacturing is continuing with a trend towards persongive to control code implementations to execute actions (Zoitl et al., 2013). fined interfaces that give access to available control code manufacturing is continuing with a trend towards personalized products (Koren, 2010). This revolution is driven manufacturing is continuing with a trend towards personimplementations to execute actions (Zoitl et al., 2013). alized products (Koren, 2010). This revolution is driven driven implementations to execute et al., 2013). Additionally, the approach is actions based on(Zoitl models of resource implementations to execute actions (Zoitl et al., 2013). alized products (Koren, 2010). This revolution is by customer demands that vary over time. Current maralized products (Koren, 2010). revolution is driven Additionally,and thematerial approachflows. is based based on models models of resource resource by customer demands that varyThis overDecker, time. Current Current marthe approach is on of capabilities Action sequences for the Additionally,and thematerial approachflows. is based on models of resource by demands that vary over time. markets are saturated (Westk¨ amper 2006) with a Additionally, by customer customer demands that vary and overDecker, time. Current marcapabilities Action sequences for the the kets are saturated (Westk¨ a mper and 2006) with a capabilities and material flows. Action sequences for production can be automatically generated using capabilities and material flows. Action sequences the kets are saturated (Westk¨ a mper and Decker, 2006) with a higher than(Westk¨ demand which suppliers to bea production can be automatically generated usingforthese kets aresupply saturated amper andforces Decker, 2006) with these higher supply than demand which forces suppliers to be production can be automatically generated using these models. The action sequence considers available material production can be automatically generated using these higher supply than demand which forces suppliers to be more flexible in their processes to be economically viable. higherflexible supplyinthan which suppliersviable. to be models. models. The action action sequence considers available material more theirdemand processes to have beforces economically The considers available flow information to sequence ensure that it is valid. For material this, we models. The action sequence considers available material more flexible in their processes to be economically viable. Manufacturing systems nowadays to support heteromore flexible in their processes to be economically viable. flow information to ensure that it is valid. For this, we we Manufacturing systems nowadays have to support heteroflow information to ensure that it is valid. For this, differentiate between external and internal material flow information to ensure that it internal is valid.material For this,flow. we Manufacturing systems have support heterogeneous products with nowadays low volume in to to be able differentiate Manufacturing systems nowadays have toorder support heterobetween external and flow. geneous products with low volume in order to be able differentiate between external and internal material External material flow describes material flowflow. bedifferentiate between external and the internal material flow. geneous products with low volume in order to be able to compete in highly competitive markets. Changeable geneous products with low volume in order to be able External material flow describes the material flow beto compete compete in systems highly competitive competitive markets. Changeable material flow material flow between different resources, whereas the internal material External material flow describes describes the material flow flow beto highly Changeable manufacturing as describedmarkets. by Wiendahl et al. External to compete in in systems highly competitive markets. Changeable tween different resources, whereas internal material flow manufacturing as described by Wiendahl et al. tween different resources, whereas internal material flow is limited to the material flow within a resource. The first tween different resources, whereas internal material flow manufacturing systems as described by Wiendahl et al. (2007) are necessary to cope with turbulent environments. manufacturing systems as described by Wiendahl et al. is is limited limited to the the material flow within resource. The first first (2007) arecharacterized necessary to by cope with turbulent environments. material within aaa resource. The of this paperflow is the automatic generation of is limited to to the material flow within resource. The first (2007) are necessary cope with turbulent environments. They being adaptable, and contribution (2007)are arecharacterized necessary to to by cope with turbulentintelligent, environments. contribution of this paper is the automatic generation of They are being adaptable, intelligent, and contribution of this paper is the automatic generation of action sequences that consider product requirements as contribution of this paper is the automatic generation of They are characterized by being adaptable, intelligent, and versatile (Zor et al., 2010). They are characterized by being adaptable, intelligent, and action sequences that consider product requirements as versatile (Zor (Zor et et al., al., 2010). 2010). action that consider product requirements as well as sequences internal and external material flow. The second action sequences that consider product requirements as versatile versatile et in al.,manufacturing 2010). well as as internal and external material flow. make The second second The role (Zor of IT has increased (Vogel- well external flow. The contribution is anand approach to material automatically action well as internal internal and external material flow. make The second The role role of IT in manufacturing manufacturing hasITincreased increased (Vogelcontribution is an approach to automatically action The of IT in has (VogelHeuser et al., 2009). However, current systems are still contribution is an approach to automatically make action sequences executable. The benefit of this is the reduction of The role of IT in manufacturing has increased (Vogelcontribution is an approach to automatically make action Heuser et al., al., 2009). 2009). However, current ITand systems are still still sequences executable. The benefit of this is the reduction of Heuser et However, current IT systems are too inflexible (Zor et al., 2010; Sauer Jasperneite, sequences executable. The benefit of this is the reduction of manual effort when switching between different products Heuser et al., 2009). However, current IT systems are still sequences executable. The benefit of this is the reduction of too inflexible (Zor et al., 2010; Sauer and Jasperneite, manual effort when switching between different products too (Zor 2010; 2011) and require lot al., of manual efforts and whenJasperneite, changes in manual effort when switching different and factory setups. A factorybetween setup refers to products a set of too inflexible inflexible (Zoraa et et al., 2010; Sauer Sauer and Jasperneite, manual effort when switching between different products 2011) and require lot of manual efforts when changes in and factory factory setups.and A their factory setup within refers to to set of of 2011) and lot efforts when changes the production areaarequired (Z¨ ah et al., 2010). setups. A factory setup refers set available resources relations theaaafactory. 2011) and require require lot of of manual manual efforts when Especially, changes in in and and factory setups.and A their factory setup within refers to set of the production are required required (Z¨ ah h et et al., 2010). 2010). Especially, available resources relations the factory. the production are (Z¨ a al., Especially, scheduling and planning software requires a lot of manual available resources their within factory. is demonstrated using modular production the production are required (Z¨ ah et al., 2010). Especially, available resources and and theiraa relations relations within the the system. factory. scheduling and planning planning software requires lotapproaches of manual manual This This is demonstrated demonstrated using modular production production system. scheduling and requires aaa lot of effort to set-up (Kl¨ oppersoftware et al., 2009). Some This is using aa modular system. scheduling and planning software requires lot of manual This is demonstrated using modular production system. effort to set-up set-up (Kl¨ pper et al., al., 2009).generate Some approaches approaches remainder of this paper is structured as follows: Seceffort to (Kl¨ ooopper et Some are available that can automatically schedules. The effort to set-up (Kl¨ pper et al., 2009). 2009).generate Some approaches The remainder remainder ofoverview this paper paper isavailable structured as follows: follows: Secare available that can automatically schedules. The this structured as Section 2 gives anof of is work in the field The this paper structured as follows: Secare available that can automatically generate schedules. Nevertheless, they can cannot react to dynamic of tion remainder are available that automatically generate changes schedules. 2 gives gives anofoverview overview of isavailable available work in theintrofield Nevertheless, they cannot react to dynamic changes of tion 2 an of work in the field of action sequence generation. In Section 3 a brief tion 2 gives an overview of available work in the field Nevertheless, they cannot react to dynamic changes of the scheduling problem and are only suitable for static Nevertheless, they cannot react to changes of of action action sequence sequence generation. In Section Section brief introthe scheduling(Cheeseman problem and are 2005). onlydynamic suitable for static static generation. In 333 aaaonbrief introduction the execution of action sequences resources of actionof sequence generation. In Section brief introthe scheduling problem and are only suitable for environments et al., Additionally, au- of the scheduling problem and are only suitable for static duction of the execution of action sequences on resources environments (Cheeseman et al., al., 2005). to Additionally, au- duction the execution of action on resources is given.of The required models tosequences enable the approach duction of the execution of action sequences on resources environments (Cheeseman et 2005). Additionally, automatic transformation from schedule machine code environments (Cheeseman et al., 2005). Additionally, auis given. given. The in required models to enable enable thegeneration approach tomaticbetransformation transformation from schedule schedule to machine code is The required to the approach are explained Sectionmodels 4. In Section 5 the is given. The in required to enable thegeneration approach tomatic from would beneficial to complement such to toolmachine chains. code tomaticbetransformation from schedule to machine code are are explained Sectionmodels 4. In In Section Section 5 the the would beneficial to complement such tool chains. explained in Section 4. 5 generation are explained in Section 4. In Section 5 the generation would be beneficial to complement such tool chains. would be beneficial to complement such tool chains. Copyright 2015 IFAC 1572Hosting by Elsevier Ltd. All rights reserved. 2405-8963 © 2015, IFAC (International Federation of Automatic Control) Copyright 2015 responsibility IFAC 1572Control. Peer review© of International Federation of Automatic Copyright © 2015 1572 Copyright ©under 2015 IFAC IFAC 1572 10.1016/j.ifacol.2015.06.299

INCOM 2015 Nadine Keddis Gerd Kainz Alois Zoitl et al. / IFAC-PapersOnLine 48-3 (2015) 1502–1508 May 11-13, 2015. Ottawa, Canada



process is introduced in detail. Section 6 describes the experimental setup and the evaluation of the approach. Finally, Section 7 concludes the paper.

1503

Module Lever Module Drill Attributes: Material plastic, 8mm hole

2. RELATED WORK There has been much work in the past years to increase software quality and re-usability of control software, e.g., (S¨ under et al., 2006; Eckert et al., 2012; Sorouri et al., 2012; Zoitl and Pr¨ ahofer, 2012). The approaches mainly try to develop design patterns and improve the development of control software to achieve this. However, they focus on control software of a resource only and not on generating the equivalent action sequences. Additionally, they target rigid control software and focus on how to manage variability for different resources. In our case, we also consider that different products and their variants might be produced on the same resource. The resource must be able to support different action sequences and be able to work in different factory setups containing different resources. The goal is to allow the execution of different operations on the same resource depending on the desired product. Nevertheless, such modular concepts for control software are necessary to enable such an approach. They act as a starting point for our approach. There has been much work in the field of recipes in manufacturing as well. They are similar to action sequences but more static and defined manually for each resource. For fixed factory setups and plant structures, standards like ISA 88 were developed by the International Standardization Association (Brandl, 2006). ISA 88 separates the description of production steps from manufacturing resources. The standard provides means to coordinate different resources based on a predefined recipe for a product. Originally this was developed for the process manufacturing, but similar concepts for the discrete manufacturing exist, e.g., PackML guideline (Arens et al., 2006). The NAMUR association gives a recommendation for recipes and defines general requirements of recipes in NE033 (NAMUR, 2003). It gives a recommendation on how to structure recipes for discontinuous processes and mainly focuses on the process industry. It is a good foundation for introducing recipe-based operation. Nevertheless, it does not propose how these recipes can be used from a planning and scheduling perspective. In the field of agent-based planning and scheduling there has been plenty of work to develop suitable platforms and planning strategies, e.g., (Bussmann and Schild, 2001; Gabel and Riedmiller, 2008; Leit˜ ao, 2009; Alexakos et al., 2012; Lepuschitz et al., 2013). Holonic manufacturing tackles similar problems. They are used for coordination and sequencing of manufacturing resources (McFarlane and Bussmann, 2003; Lohse et al., 2005). Such approaches focus on negotiating schedules and allocating resources for the operations. The agents execute available code depending on the negotiated schedule. Z¨ ah et al. (2008) propose an approach to generate schedules using different capabilities. However, they do not discuss how the generated schedules can be translated into machine-readable control commands that can be used to automatically execute them on resources.

Capabilities: * Transport * Drill * Test Height

Module Test Height

Interaction Points Material Flow

Module Rotary Table

Fig. 1. Resource model for a processing station with capabilities and internal material flow information. 3. BACKGROUND The starting point for our approach is a modular programmable logic controller (PLC) program. This is described in detail in the work of Zoitl et al. (2013). We just give a brief overview here. Every component is represented by a module comprising all functions required to control it as well as a set of interfaces to access the functions. Additionally, there is a recipe controller that is responsible for coordinating the execution of the module programs. Moreover, each program has a startup code, called program outline, which includes the basic control of a resource with its locking mechanisms. It also triggers the recipe controller that executes the currently loaded action sequence. Each command in an action sequence consists of a flag in the beginning to indicate whether this is a branch instruction or not. This is followed by a set of flags to determine the branch. Afterwards, the command is specified using an identifier for the used module, interface, and parameters. The execution of action sequences is similar to how CPUs function.

4. SYSTEM MODELING In order to generate action sequences based on the product requirements, a factory-independent description of the product has to be provided. Additionally, a description of the factory and the available resources is necessary. As proposed in previous work (Keddis et al., 2013, 2014) the resources are modeled based on their capabilities. Capabilities of each resource express the processing steps that are supported and can be executed by this resource. We just give a brief example of the resource model, for more information see (Keddis et al., 2013). In addition to modeling capabilities, the internal material flow of the resource has to be modeled. The internal material flow describes the internal structure of a resource and how it is actually built to be able to support the processing steps. Moreover, a factory model is provided that represents how the different resources interact within the factory. Combining this information with a model for the control software of the resource can then enable an automatic generation of a product-based action sequence that can be directly executed on the resource. In the following, the different required models are presented in detail.

1573

INCOM 2015 Nadine Keddis Gerd Kainz Alois Zoitl et al. / IFAC-PapersOnLine 48-3 (2015) 1502–1508 May 11-13, 2015. Ottawa, Canada

1504

Input Point input1

Rotation Plate P1

Rotation Plate P2

Rotation Plate P3

Verify Orientation P1

Drill P1

Rotation Plate P6

Rotation Plate P5

4.2 Modeling Factories

Input Point input2 Rotation Plate P4

Lever P1

Lever P2

Overlapping positions Input and Output positions Material Flow

Output Point output1

Fig. 2. Resulting internal material flow information for a processing station. 4.1 Modeling Resources A resource model consists of all capabilities of a resource as well as its internal material flow. A resource is composed of several modules that offer a specific capability. Fig. 1 illustrates an example of such a model. The processing station is composed of a drill module, a testing module, a rotary table module, and a lever module. All modules are composed of actuators and sensors. The rotary table and the lever provide a transport capability. The drill provides a drill capability. The testing module provides a test orientation capability. The internal material flow is used to describe how material can get from one place to another within a resource. It describes the logical flow of material. Since we are looking at adaptable manufacturing systems, this is necessary to be able to use different modules of a resource for different products. To ensure that the paths between the modules used in an action sequence exist, the internal material flow has to be modeled. Therefore, each module in the resource model additionally has at least one position. Modules that can transport material have several positions, describing all the places the material can be transported to and from. Furthermore, the links between different positions of one module can be unidirectional or bidirectional. This depends on the type of the module. A conveyor belt for example can transport in one or both directions. Positions of different modules can overlap to model that the material can be handed over from one module to the next. Overlapping positions are physically the same position but logically they indicate that material can move in both directions between modules (bidirectional flow). In the example in Fig. 1 the rotary table has 6 positions, the lever 2, and the other modules have one position each. Both the rotary table and the lever have a unidirectional material flow between their different positions. The testing module and the rotary table overlap in the second position of the rotary table. The drill and the rotary table overlap in the third position of the rotary table. Finally, the lever and the rotary table overlap in the fourth position of the rotary table. As a last step in the modeling process, the inputs and outputs of the resource have to be mapped to a position of a module. This is similar to the concept of overlapping positions. The only difference is that material can only be moved in one direction. It is either transported from the input to the module or from the module to the output. Fig. 2 shows the resulting material flow graph for the processing station.

After modeling the resource capabilities and internal material flow, instances of the resource models can be combined to model a whole factory. Additionally, external material flow information has to be added to the model to ensure that only feasible action sequences are generated. The factory model stores information about the current factory setup with available resources and their relations. The relations are represented by connections between different resources. This is the required external material flow information that is necessary to ensure that the material can later on get to the right resource. Each connection is a link between two interaction points, each belonging to a different resource. The interaction points for the processing station are illustrated in Fig. 1. The connection also has a direction that is inferred based on the types of interaction points that are linked together. Connecting an output of resource X with an input of resource Y results in a unidirectional connection from X to Y . Bidirectional connections are a result of connecting input/output points of different resources. The arrow above the interaction points in Fig. 1 indicates whether the material flow is bidirectional or unidirectional. In the latter case it also depicts the direction. The graph in the lower left box in Fig. 3 depicts a factory model with external material flow information represented by the arrows in the graph. The factory model can be automatically generated at run-time as described in our previous work (Keddis et al., 2013).

4.3 Mapping Resource Interfaces to Control Software Executable action sequences can only be automatically generated when there is a defined interface with the corresponding control software for each module. For each required operation there has to be a mapping to the commands that have to be executed on the resource. Since we use a modular concept, the operations are specific to a module and thus, are attached to the module description. For each operation that is supported by a module, there is an executable command attached that can be downloaded to the control device of the resource in case this operation is required for the current product. For the processing station example there are interfaces related to each of the modules. The rotary table provides a transport interface with two positions as parameters. Whenever a transport operation using the rotary table is required, two positions are passed to the module and the corresponding control code is executed. The first position in that case describes from where to transport the material while the second position corresponds to the position where the material should go to. Similarly, the lever module provides an interface to its transport operation. The drill and the testing module offer an interface to their operation. This is a drill interface and a test height interface respectively. Each module implements an interface that contains the execution of the operation on this specific module. Two resources that both have two different types of e.g. a drill module will use the same interface to access the drill functions, but might have two different implementations of this interface.

1574

INCOM 2015 Nadine Keddis Gerd Kainz Alois Zoitl et al. / IFAC-PapersOnLine 48-3 (2015) 1502–1508 May 11-13, 2015. Ottawa, Canada



4.4 Product Description The product is described by its required production steps and their dependencies. Each of the production steps is described as a capability. The capabilities are the same as used in the resource model to enable an automatic mapping of production steps to resources. A simplified example is illustrated in Fig. 3 in the upper left box. 5. GENERATING ACTION SEQUENCES In adaptable manufacturing systems several products can be produced using the same resources. If it is possible, the factory setup is reused for the different products and only the action sequences are updated accordingly. In some cases a change in the factory setup might be necessary. Changes include adding or removing resources, or changing the position of a resource within the factory. It is also possible to use different modules of a resource for different products. In order to automatically generate action sequences for the different products, two aspects have to be considered during the generation process: the product description and the factory setup. These can then be combined to automatically generate executable action sequences. Fig. 3 illustrates this workflow. 5.1 Automatic Action Sequence Generation Based on the required capabilities defined in the product description and the factory setup, action sequences can be generated automatically. The factory model mentioned earlier contains all the needed information about the factory setup. It includes all available resources with their capabilities and internal material flow description as well as all the connections between different resources that model the external material flow. The first step in the automatic generation of action sequences maps each production step to all the resources providing the required capability. The mapping can be generated by a simple matching algorithm that iterates over all available resources and checks whether they can provide the capability or not. If there is no matching resource for one production step, then the product cannot be produced using the current factory setup. The algorithm stops in that case and no action sequences are generated. In case there is a match, the production step is mapped to the corresponding module that supports the capability on the resource. The second step generates a valid action sequence considering the internal and external material flow. To find a valid action sequence we suggest using a branch-andbound complete search with backtracking. The action sequence is generated backwards starting with the last production step of the product description. This ensures that the precedence relations described by the dependencies in the production plan are maintained because no operation is planned before its predecessors have finished. The resource mapping to the production steps introduced earlier is the starting point for the generation process. After finding suitable resources for each operation, we have to check whether there are valid internal and external material flows when these resources are used. The algorithm that generates a valid action sequence considering the external material flow is described in our previous work (Keddis

1505

et al., 2014). In this paper, we focus on integrating the internal material flow as well and generating executable code out of it. The action sequence is generated iteratively. While generating the action sequence, we have to consider the internal material flow in addition to the external material flow in each step. Let’s assume that the required operation in that step is mapped to module m1 on resource r1 . If the previous operation was assigned to a module m2 on the same resource (r1 ), then we only have to check whether there is a valid internal material flow from m1 to m2 . The external material flow is irrelevant in this case. If the previous operation was assigned to a module m3 on a different resource (r2 ), then both the internal and external material flows are relevant. We first check whether we can get from the input of r1 to m1 . If r1 has several inputs, we consider the one connected to r2 . The material can flow from the input to the module if there is a path in the internal material flow graph as illustrated in Fig. 1. The path is determined using a breadth-first-search. Since this returns the shortest path, we ensure hereby that we do not consider overlapping positions and modules and thus the path does not contain unnecessary detours through the resource. For each intermediate module a transport operation is added, so that the material can be transported from the input to m1 . We store these steps in the action sequence. The same procedure is done to determine how to get from m3 to the output of r2 . Then we check if there is a valid external material flow from r2 to r1 . For each intermediate resource, we determine a valid path from the input to the output of this resource and add all intermediate transport operations to the action sequence as done while checking the internal material flow. The previous steps are repeated for each operation in the product description until a valid action sequence is found. If several modules of a resource are required in consecutive steps, we first check whether we can get from the input to the first required module. Then the path between all the required modules is determined. Finally, we check how to get from the last module to the output. The search is stopped whenever there is no valid internal or external material flow and backtracking to the next branch is triggered. There is no valid action sequence when there are no backtracking branches left and no previously evaluated branch in the search tree has led to a valid action sequence. The resulting action sequence is depicted in Fig. 3 in the middle. The color of the ellipse indicates the resource that should execute the operation. The text in the ellipse indicates which operation has to be executed and which module is assigned for it. 5.2 Executable Action Sequences After generating a valid action sequence, it is translated into an executable one. The translation is based on the interfaces that are defined for each module of the resource. Each module includes a mapping from its interfaces that are used in the generated action sequence to the corresponding control code. The planning component informs each resource which modules should be used and which commands are required and the resource defines the required executable action sequences by invoking the respective modules. For each operation in the action se-

1575

INCOM 2015 Nadine Keddis Gerd Kainz Alois Zoitl et al. / IFAC-PapersOnLine 48-3 (2015) 1502–1508 May 11-13, 2015. Ottawa, Canada

1506

Supply Black Case (Stack Magazine Black)

Production Plans Supply Temperature Sensor

Supply Case

Transport from P1 to P2 (Conveyor)

Supply Temperature Sensor

Supply Case

Supply Case Assemble

Transport form P1 to P2 (Conveyor)

Test Orientation Assemble

Test Height

Supply Temperature Sort Sensor

Drill

Supply Sensor (Pick&Place)

Sort

Assemble

Pick_Place.assemble(true);

Store

Assemble (Pick&Place)

Factory Model with Material flow and capabilities Supply_RBS

• Supply Black Case • Supply Silver Case • Supply Red Can

Pick&Place Temperature

• Supply Temperature Sensor • Assemble • Transport

Pick&Place Cover Sorting

IF Workpiece detected THEN Conveyor.transportWP(P1, P2); END IF;

Transport form P2 to P3 (Conveyor)

Conveyor.transportWP(P2, P3);

Transport from P1 to P3 (Conveyor)

Transport from P1 to P2 (Conveyor)

• Supply Cover • Assemble • Transport • Sort Workpieces • Store Workpieces

Store (Storage Black)

Fig. 3. The workflow for generating action sequences. A valid action sequence is generated from the product and factory descriptions. This is later translated into an executable action sequence. quence, the executable is then composed of the required code snippets. For example if we only want to drill on the processing station, the action sequence would include a transport operation on the rotary table from the input position to the drill, a drill operation on the drill, a transport operation on the rotary table from the drill to the lever, and a transport operation from the lever to the output position. The processing station would then translate these four operations to the necessary control code. The control code includes the module that should execute the operation, the operation it should execute, and the parameters that are specific for this product. An example of the translation is depicted in Fig. 3 on the right. In general, this results in one executable action sequence that includes all the control code that is required by a resource. In case a resource is used at two different times in the action sequence, several executable action sequences can be generated for each point in time. Each executable action sequence is then automatically downloaded to the resource when the corresponding operation should start. Afterwards, the resource processes the executable action sequence and performs the operation included in it. The download process is illustrated in Fig. 4.

red, or silver thermometers, black, red, or silver hygrometers, red boxes, black, red, or silver workpieces with different characteristics. The different characteristics allow changing the production steps required for a product to demonstrate that different action sequences are generated for different products. The manufacturing systems can be composed of different Festo stations, conveyor belts, and mobile robots. Each of them provides a set of capabilities to support the production of the above mentioned products. To increase variability, production steps supported by the stations can be included or skipped in the production plan. Optional steps are testing, verifying orientation, drilling, and assembly. Stations are either controlled by a Siemens S7 300 PLC or a Festo CoDeSys PLC. Additionally, a PC is used as a supervisory control. The supervisory control calculates the valid action sequences based on the resource models and the factory model. The models are defined using the Eclipse Modeling Framework (EMF). EMF4CPP is used to generate C++ classes out of the models. Besides, a graphical user interface (GUI) runs on the supervisory control to enable user interaction. The user can specify the type of product that should be Production Control Supply Black Case

6. EXPERIMENTAL SETUP AND RESULTS

Supply Sensor

6.1 Experimental Setup

Assemble

A simplified industrial manufacturing system is used to evaluate the approach. The setup is used for educational purposes and consists of different combinable resources from the Festo modular production system. One possible setup is shown in Fig. 4 on the right 1 . Different products can be produced within the manufacturing system: black, 1

The experimental setup and the results can be seen in the following video: http://www.youtube.com/watch?v=Tkcv-mbhYqk.

Transport

Sort

Fig. 4. A simplified action sequence is divided into an executable action sequence for each resource and downloaded to it.

1576

INCOM 2015 Nadine Keddis Gerd Kainz Alois Zoitl et al. / IFAC-PapersOnLine 48-3 (2015) 1502–1508 May 11-13, 2015. Ottawa, Canada



produced in the GUI and trigger the generation of action sequences that are afterwards automatically executed. Several products can be selected at the same time. If the selected products cannot be produced on the current setup, this is displayed in the GUI by highlighting the missing capabilities. Missing links between resources can also be highlighted in the GUI. 6.2 Evaluation For the evaluation we produce different products that require different modules of the resources to be active. For each product a valid action sequence can be found within seconds without manual configuration or programming. Product descriptions included up to ten production steps and the factory setup was composed of up to seven resources. Since the number of machines that have to be considered in each step is much smaller than the number of operations, the search space is small. Branches can be bound as soon as there is no resource that provides the capability, there is no external material flow, or there is no internal material flow. This can quickly limit the search and only a few branches of the search tree reach the last level. After generating a valid action sequence, this was successfully translated into executable action sequences. The translation was done automatically and no further manual changes were necessary to execute the action sequences. When changes in the factory setup occurred, this was automatically reflected in the generated action sequence in form of a changed number of transport operations for different modules on different resources. In order to use this approach, the resource models and the product models have to be created manually once in the beginning. The factory model can be automatically generated as described in previous work (Keddis et al., 2013). Afterwards, the models can be used for different products and factory setup. This procedure is less-error prone than traditional engineering because models can be thoroughly tested and reused later. The program outline and the recipe controller for the resource have to be programmed once manually and later the generated action sequence can be downloaded and executed automatically. The demonstration showed that the proposed approach is suitable for adaptable manufacturing systems where there are frequent switches between different products and factory setups, because it reduces manual effort, while ensuring that the production can be executed correctly. The approach still has to be evaluated on large problems. 7. CONCLUSION In this paper we proposed an approach to automatically generate action sequences for adaptable manufacturing systems. The approach takes the requirements of the product into consideration and generates an action sequence that can be executed directly in the given manufacturing system. For this, a valid action sequence is generated based on resource and factory models, as well as mapping from interfaces to control software for each module. Thus, the action sequence generation adapts to changes in the manufacturing system in addition to changed product descriptions. It is suitable for adaptable manufacturing systems that have to deal with lots of product variants and different

1507

products. A simplified industrial setup was used to evaluate the approach. For different products and factory setups different action sequences were generated without manual effort and user interaction. The action sequences were automatically downloaded to the resources when needed and the production was triggered and executed without further intervention. The contribution in this work is an adaptable planning algorithm that adapts resource control programs depending on product requirements and factory topology. Currently, the approach is limited to small problems. The next step in our work involves evaluating the approach on larger problems. Additionally, we intend to expand the approach to other PLC systems. Currently, the approach was tested with Siemens S7 only. Other technologies should be supported in the future. For this, the interpretation and execution of the generated recipes must be implemented on the different PLCs. AKNOWLEDGMENT This approach is a result of the research and development project “SpeedFactory”, which was funded by the German Federal Ministry for Economic Affairs and Energy as part of the technology programme “Autonomics for industry 4.0” and supervised by the German Aerospace Center (DLR) under Grant No. 01MA13002B. REFERENCES Alexakos, C., Georgoudakis, M., Kalogeras, A., and Likothanassis, S. (2012). Adaptive Manufacturing Utilizing Ontology-driven Multi-Agent Systems: Extending Pabadis’ Promise Approach. In IEEE International Conference on Industrial Technology (ICIT), 42–47. IEEE. Arens, D., Hopfgartner, T., Jensen, T., Lamping, M., Pieper, M., and D., S. (2006). Packaging Machine Language V3.0 Mode & States Definition Document. OMAC Motion for Packaging Working Group. Brandl, D. (2006). Design Patterns for Flexible Manufacturing. ISA. Bussmann, S. and Schild, K. (2001). An Agent-based Approach to the Control of Flexible Production Systems. In IEEE International Conference on Emerging Technologies and Factory Automation (ETFA), volume 2, 481–488. IEEE. Cheeseman, M., Swann, P., Hesketh, G., and Barnes, S. (2005). Adaptive Manufacturing Scheduling: A Flexible and Configurable Agent-based Prototype. Production Planning & Control, 16(5), 479–487. Eckert, K., Hadlich, T., Frank, T., Fay, A., Diedrich, C., and Vogel-Heuser, B. (2012). Design Patterns for Distributed Automation Systems with Consideration of Non-Functional Requirements. In IEEE International Conference on Emerging Technologies & Factory Automation (ETFA). IEEE. Gabel, T. and Riedmiller, M. (2008). Adaptive Reactive Job-Shop Scheduling with Reinforcement Learning Agents. International Journal of Information Technology and Intelligent Computing, 24(4). Hu, S., Zhu, X., Wang, H., and Koren, Y. (2008). Product Variety and Manufacturing Complexity in As-

1577

INCOM 2015 Nadine Keddis Gerd Kainz Alois Zoitl et al. / IFAC-PapersOnLine 48-3 (2015) 1502–1508 May 11-13, 2015. Ottawa, Canada

1508

sembly Systems and Supply Chains. CIRP AnnalsManufacturing Technology, 57(1), 45–48. Iacocca Institute (1991). 21. Century Manufacturing Enterprise Strategy: An Industry-Led View. Technical report, Iacocca Institute, Bethlehem, Pennsylvania. Keddis, N., Kainz, G., Buckl, C., and Knoll, A. (2013). Towards Adaptable Manufacturing Systems. In IEEE International Conference on Industrial Technology (ICIT), 1410–1415. IEEE. Keddis, N., Kainz, G., and Zoitl, A. (2014). Capabilitybased Planning and Scheduling for Adaptable Manufacturing Systems. In IEEE International Conference on Emerging Technologies & Factory Automation (ETFA). IEEE. Kl¨ opper, B., Sondermann-W¨ olke, C., Romaus, C., and V¨ ocking, H. (2009). Probabilistic Planning Integrated in a Multi-level Dependability Concept for Mechatronic Systems. In IEEE Symposium on Computational Intelligence in Control and Automation (CICA), 104–111. IEEE. Koren, Y. (2010). The Global Manufacturing Revolution. John Willey & Sons, New Jersey. Leit˜ ao, P. (2009). Agent-based Distributed Manufacturing Control: A State-of-the-Art Survey. Engineering Applications of Artificial Intelligence, 22(7), 979–991. Lepuschitz, W., Groessing, B., Merdan, M., and Schitter, G. (2013). Evaluation of a Multi-agent Approach for a Real Transportation System. In IEEE International Conference on Industrial Technology (ICIT), 1273–1278. IEEE. Lohse, N., Hirani, H., Ratchev, S., and Turitto, M. (2005). An Ontology for the Definition and Validation of Assembly Processes for Evolvable Assembly Systems. In The 6th IEEE International Symposium on Assembly and Task Planning: From Nano to Macro Assembly and Manufacturing (ISATP), 242–247. doi: 10.1109/ISATP.2005.1511480. McFarlane, D.C. and Bussmann, S. (2003). Holonic Manufacturing Control: Rationales, Developments and Open Issues. In Agent-based manufacturing, 303–326. Springer. NAMUR (2003). NE 033 Requirements to be met by Systems for Recipe-Based Operations.

Sauer, O. and Jasperneite, J. (2011). Adaptive Information Technology in Manufacturing. In CIRP Conference on Manufacturing Systems. Madison, WI, USA. Sorouri, M., Patil, S., and Vyatkin, V. (2012). Distributed Control Patterns for Intelligent Mechatronic Systems. In IEEE Conference on Industrial Informatics (INDIN). IEEE. S¨ under, C., Zoitl, A., and Christoph, D. (2006). Functional Structure-based Modelling of Automation Systems. International Journal of Manufacturing Research, 1(4), 405–420. Vogel-Heuser, B., Kegel, G., Bender, K., and Wucherer, K. (2009). Global Information Architecture for Industrial Automation. atp. Westk¨amper, E. and Decker, M. (2006). Einf¨ uhrung in die Organisation der Produktion. Springer. Wiendahl, H.P., ElMaraghy, H., Nyhuis, P., Z¨ ah, M.F., Wiendahl, H.H., Duffie, N., and Brieke, M. (2007). Changeable Manufacturing – Classification, Design and Operation. CIRP Annals-Manufacturing Technology, 56(2), 783–809. Z¨ah, M.F., Beetz, M., Shea, K., Reinhart, G., Stursberg, O., Ostgathe, M., Lau, C., Ertelt, C., Pangercic, D., R¨ uhr, T., et al. (2008). An Integrated Approach to Realize the Cognitive Machine Shop. In Proceedings of the 1st International Workshop on Cognition for Technical Systems, 6–8. Z¨ah, M.F., Reinhart, G., Ostgathe, M., Geiger, F., and Lau, C. (2010). A Holistic Approach for the Cognitive Control of Production Systems. Advanced Engineering Informatics, 24(3), 300–307. Zoitl, A. and Pr¨ahofer, H. (2012). Guidelines and Patterns for Building Hierarchical Automation Solutions in the IEC 61499 Modeling Language. IEEE Transactions on Industrial Informatics, PP(99), 1. doi: 10.1109/TII.2012.2235449. Zoitl, A., Kainz, G., and Keddis, N. (2013). Production Plan-Driven Flexible Assembly Automation Architecture. In Industrial Applications of Holonic and MultiAgent Systems, 49–58. Springer. Zor, S., G¨orlach, K., and Leymann, F. (2010). Using BPMN for Modeling Manufacturing Processes. In Proceedings of 43rd CIRP International Conference on Manufacturing Systems, 515–522.

1578