Proceedigs of theOttawa, 15th IFAC Symposium on May 11-13, 2015. Canada Proceedigs of the 15th IFAC Symposium on Information Control Problems in Manufacturing Proceedigs of the 15th IFAC Symposium on Information Control Problems in Manufacturing Available online at www.sciencedirect.com May 11-13, 2015. Ottawa, Canada Information Control Problems in Manufacturing May 11-13, 2015. Ottawa, Canada May 11-13, 2015. Ottawa, Canada
ScienceDirect
AIFAC-PapersOnLine Conceptual for Operational 48-3Model (2015) 1865–1869 A for A Conceptual Conceptual Model for Operational Operational Control in SmartModel Manufacturing Systems A Conceptual Model for Operational Control Control in in Smart Smart Manufacturing Manufacturing Systems Systems Control in Smart Manufacturing Systems Timothy Sprock, Leon F. McGinnis
Timothy Sprock, Sprock, Leon Leon F. F. McGinnis McGinnis Timothy Timothy Sprock, Leon F. McGinnis Georgia Institute of Technology, Atlanta, GA, 30332 email:
[email protected],
[email protected] Georgia Institute of of Technology, Technology, Atlanta, GA, GA, 30332 30332 Georgia Institute Atlanta, Georgia Institute of Technology, Atlanta, GA, 30332 email:
[email protected],
[email protected] email:
[email protected],
[email protected] email:
[email protected],
[email protected] Abstract: Advancements in computer integrated manufacturing and intelligent devices have spurred a revolution in manufacturing. To maximize their potential, smart operational control mechanisms must not only
Advancements in computer integrated manufacturing and intelligent devices spurred Abstract: Advancements in system computer integrated manufacturing andsolve intelligent devicesofhave have spurred aa Abstract: integrate real-time data from operations, but also formulate and a wide variety optimal-control revolution in manufacturing. To maximize their potential, smart operational control mechanisms must not Advancements in computer integrated manufacturing and intelligent devices have spurred a Abstract: revolution in manufacturing. Toand maximize their potential, smart operational mechanisms must not only only analyses quickly and efficiently then translate the results into executablecontrol commands. To support design of
integrate real-time data system operations, but formulate and solve aa wide variety revolution in manufacturing. maximize their potential, smart operational mechanisms must not only integrate real-time data from from To system operations, buta also also formulate and capable solvecontrol wide variety of of aoptimal-control optimal-control smart operational controllers, this paper proposes conceptual model of integrating description of analyses quickly and efficiently and then translate the results into executable commands. To support design of integrate real-time data from system operations, but also formulate and solve a wide variety of optimal-control analyses quickly and in efficiently and then translate the management results into executable support design of the control activities the manufacturing operations level of thecommands. enterprise To with a description smart operational controllers, this paper proposes a conceptual model capable of integrating a description of analyses quickly and efficiently and then translate the results into executable commands. To support design smart operational this interface paper proposes a conceptualanalyses. model capable of integrating a description of the physical systemcontrollers, and an explicit to optimal-control the activities in operations management level of with smart operational controllers, this paper proposes a conceptual model of integrating the control control activities in the the manufacturing manufacturing operations management levelcapable of the the enterprise enterprise with aa description description of of Keywords: control system design, intelligent manufacturing systems, optimal control the physical system and an explicit interface to optimal-control analyses. control activities in the manufacturing operations management level of the enterprise with a description © 2015, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved. of the physical system and an explicit interface to optimal-control analyses. the physical system system and an explicitintelligent interface to optimal-control analyses. Keywords: Keywords: control control system design, design, intelligent manufacturing manufacturing systems, systems, optimal optimal control control Keywords: control system design, intelligent manufacturing systems, optimal control
1. INTRODUCTION
Traditionally, the planning and control aspects of operations research are divided into three broad classifications reflecting INTRODUCTION the planning and aspects operations Smart manufacturing1. utilize plant-wide integration of Traditionally, 1.systems INTRODUCTION Traditionally, theplanning planning horizons: and control controlstrategic, aspects of of operations their respective tactical, and 1. INTRODUCTION Traditionally, the planning andbroad control aspects of operations research into three classifications reflecting intelligent devices and sensorsutilize to drive production agility and research are are divided divided intocontrol three broad classifications reflecting Smart manufacturing systems plant-wide integration of operational. However, in the manufacturing domain Smart manufacturing systems utilize plant-wide integration of research are divided into three broad classifications reflecting their planning horizons: strategic, tactical, and efficiency. To achieve this level of integration, industrial their respective respective planning strategic, and Smart manufacturing utilize plant-wide integration of extends intelligent devices and sensors to production agility and beyond the scope ofhorizons: operations research, tactical, requiring the intelligent devices(ICS), andsystems sensors to drive drive production agility and their respective planning horizons: strategic, tactical, and operational. However, control in the manufacturing domain control systems and supervisory control and data operational.ofHowever, control inand theprocess manufacturing domain intelligent devices and sensors to drive production agility and efficiency. To achieve this level of integration, industrial integration enterprise business control systems. efficiency. To achieve systems this levelinof particular, integration,have industrial operational. However, control in the manufacturing domain extends beyond the of research, requiring the acquisition (SCADA) been extends beyond the scope scope of operations operations research, requiring the efficiency. To achieve this level of integration, industrial control systems (ICS), and supervisory control and data The ISA-95 standard (www.isa-95.com) partially fills this control systems (ICS), and supervisory control and data extends beyond the scope of operations research, requiring the integration of enterprise business and process control systems. established to support the exchange of control and data signals integration of enterprise business and process control systems. control systems (ICS), systems and supervisory controlhave and been data requirement, and consists of models and terminology that can acquisition (SCADA) in particular, acquisition (SCADA) systems in particular, have been integration of enterprise business and process control systems. The ISA-95 standard (www.isa-95.com) partially fills this at the device-level (Galloway and Hancke, 2013). Extending Theused ISA-95 standard which (www.isa-95.com) partially fills this acquisition particular, havesignals been be established to support the exchange control and data to determine information must be exchanged established to(SCADA) support thesystems exchangeinof of control and data signals The ISA-95 standard (www.isa-95.com) partially fills requirement, and consists of models and terminology that can these successes to (Galloway enterprise-wide integration and decisionrequirement, and consists of models and terminology that this can established to support the exchange of control and data signals at the device-level and Hancke, 2013). Extending between systems associated with sales, finance and logistics at the device-level (Galloway and of Hancke, 2013). Extending requirement, and consists of models and terminology that can be used to determine which information must be exchanged making will require the definitions control and device to be be used to determine which information must be exchanged at the device-level (Galloway and Hancke, 2013). Extending these successes to enterprise-wide integration and decisionand systems associated withwith production, maintenance and these successes to enterprise-wide integration and decisionbe used to determine which information must be exchanged between systems associated sales, finance and logistics extended and supported at the operations management and betweenISA-95 systemsseparates associated sales, finance and into logistics these successes to enterprise-wide and decisionmaking will require the definitions of control and device to thewith manufacturing domain four making will require the definitions ofintegration control andfuture, device to be be quality. between systems associated sales, finance and logistics and associated with production, maintenance and business planning levels as well. In the smart and systems systems associated withwith production, maintenance and making will require the definitions of control and device to be extended and supported at the operations management and levels: 4) Business Planning & Logistics (ERP systems), 3) extended and supported at the operations management and and systems associated with production, maintenance and quality. ISA-95 separates the manufacturing domain into four operational control mechanisms must not only integrate realquality. ISA-95 separates the manufacturing domain into four extended and supported at the operations management and business planning levels as well. In the future, smart Manufacturing Operations Management (MES) 2) business planning levels as well. In formulate the future, quality. ISA-95 separates the manufacturing domain into four levels: 4) Business Planning & Logistics (ERP systems), 3) time data from system operations, but also and smart solve levels: 4) Business Planning & (PLC, Logistics (ERP systems), 3) business planning levels as well. In the future, smart operational control mechanisms must not only integrate realManufacturing Control Systems DCS, SCADA), and 1) operational controlofmechanisms must not only integrate reallevels: 4) BusinessOperations Planning & Management Logistics (ERP (MES) systems), 3) 2) atime wide variety optimal-control analyses quickly and Manufacturing Manufacturing Operations Management (MES) 2) operational control mechanisms must not only integrate realsystem operations, but also formulate and Intelligent devices. time data data from from system operations, but results also formulate and solve solve Operations Management (MES) 2) Manufacturing Control Systems (PLC, DCS, SCADA), and 1) efficiently and then translate the into executable Manufacturing Control Systems (PLC, DCS, SCADA), and 1) data variety from system operations, but also formulate and solve atime of analyses quickly and a wide wide variety of optimal-control optimal-control analyses quickly and Intelligent Manufacturing Control Systems (PLC, DCS, SCADA), and 1) devices. commands. Much of the existing literature on the implementation of a wide variety of optimal-control analyses and Intelligent devices. efficiently and translate into executable efficiently and then then translate the the results results intoquickly executable Intelligent devices. controlof for manufacturing systems has focused on efficiently and then translate the results into executable commands. literature on the of However, commands.there are several major challenges to realizing this Much Much of the the existing existingdetails literature on Levels the implementation implementation of implementation-level (ISA-95 1 &focused 2) from the commands. Much of the existing literature on the implementation of control for manufacturing systems has on vision for smart operational control mechanisms. Foremost, in control for manufacturing systems has focused on However, are major to realizing this computer for science and software engineering perspectives. With However, there thereindustrial are several several major challenges challenges tothe realizing this implementation-level control manufacturing systems has focused on details (ISA-95 Levels 1 & 2) from the contemporary engineering practice prevailing implementation-level details (ISA-95manufacturing Levels 1 & 2) from the However, there operational are several major to realizing this vision control mechanisms. Foremost, in the evolution towards advanced systems vision for for smart smart operational controlchallenges mechanisms. Foremost, in implementation-level details (ISA-95 Levels 1 & 2) from the computer science and software engineering perspectives. With paradigm neglects to conceptually or operationally separate computer science and software engineering perspectives. With vision for smart operational control mechanisms. Foremost, in contemporary industrial engineering practice the (AMS), much has been accomplished in designing and contemporary industrial engineering practice the prevailing prevailing computer science and software engineering perspectives. With the evolution towards advanced manufacturing systems the model of the plant from the model of the control of that the evolution towards systems, advancedincluding manufacturing systems contemporary industrial engineeringor the prevailing paradigm neglects to operationally separate managing these complex research in topics paradigm neglects to conceptually conceptually orpractice operationally separate the evolution manufacturing systems much has been accomplished in and plant. Moreover, optimal-control analyses, and analysis in (AMS), (AMS), muchistowards has beenadvanced accomplished in designing designing paradigm neglects to conceptually or operationally separate the model of the plant from the model of the control of that such as: how the control network organized (Dilts et and al., the model of the plant from the model of the control of that (AMS), much has been accomplished in designing and managing these complex systems, including research in topics general, that will be essential to provide on-demand decision managing theseshould complexcontrol systems,networks including be research in topics the model of the plant from the model of the control of that plant. Moreover, optimal-control analyses, and analysis in 1991), how implemented plant. Moreover, optimal-control analyses, analysis in managing theseis systems, including research in topics such as: the network organized (Dilts et al., support are currently often purpose-built to and answer specific as: how how iscomplex the control control network organized (Dilts al., plant. Moreover, optimal-control analyses, and analysis in such general, that will be essential to provide on-demand decision [SCADA, (Galloway and Hancke, 2013)], how can et general, that will be essential to provide on-demand decision such as: how is the control network organized (Dilts etlegal al., 1991), how should control networks be implemented questions, with an implicit system model and many possible 1991), how should control networks be implemented general, that will be essential to provide on-demand decision support are currently often purpose-built to answer specific sequences of controller actions be generated [MPSG, (Smith support are currently often purpose-built to answer specific 1991), how should control networks be implemented [SCADA, (Galloway and Hancke, 2013)], how can legal analysis implementations on and theanswer question, the [SCADA, (Galloway and Hancke, 2013)], how can legal support are currently oftendepending purpose-built to specific questions, with an implicit system model many possible et al., 2003)], how to generate PLC code [IEC,[MPSG, (Vogel-Heuser questions, with anthe implicit system model and many possible [SCADA, (Galloway and Hancke, 2013)], how can(Smith legal sequences of controller actions be instance data, and solver. Finally, because of the semantic sequences of controller actions be generated generated [MPSG, (Smith questions, with an implicit system model and many possible analysis implementations depending on the question, the et al., 2005)], and how to use automata and formal language analysis implementations depending on the such question, the et sequences of controller actions be generated [MPSG, (Smith al., 2003)], how to generate PLC code [IEC, (Vogel-Heuser gap between operations research models, as job al., 2003)], how generate and PLCstructure code [IEC, (Vogel-Heuser analysis implementations depending on theof question, the et instance and the Finally, because the theory to derive theto existence offormal controllers, and instance data, data, andIT-based the solver. solver. Finally, because the semantic semantic al., 2003)], how to generate PLC code and [IEC, (Vogel-Heuser et 2005)], and how to use automata language scheduling, and models, such as the of manufacturing et al., 2005)], and how to use automata and formal language instance data, and the solver. Finally, because of the semantic gap between operations research models, such as job how to define a controllable language for discrete event gap between operations research models, such as job et al., 2005)], to useand automata andof language theory to derive the existence structure controllers, and execution system (MES), models, there is such little integration theory to(Ramadge deriveand thehow existence and structure offormal controllers, and gap between research models, such between as job systems scheduling, and IT-based as and Wonham, 1989). scheduling, andoperations IT-basedand models, such as the the manufacturing manufacturing theory to derive the existence and structure of controllers, and how to define a controllable language for discrete event control analysis methods control execution tools. how to define a controllable language for discrete event scheduling, and IT-based models, such as the manufacturing execution system (MES), there is little integration between execution system (MES), there is little integration between systems how to define a controllable language for discrete event (Ramadge and Wonham, 1989). In contrast, the field of operations research is primarily systems (Ramadge and Wonham, 1989). execution system (MES),and there is little integration control analysis methods control execution tools. control analysis methods and control execution tools. between systems (Ramadge and Wonham, 1989). focused on prescriptive analytics, where the notion of control control analysis methods and control execution tools. In In contrast, contrast, the the field field of of operations operations research research is is primarily primarily In contrast, the field analytics, of operations is of primarily focused on prescriptive whereresearch the notion control Copyright © 2015 IFAC 1939focused on prescriptive analytics, where the notion of control focused on prescriptive analytics, where the notion of control 2405-8963 © IFAC (International Federation of Automatic Control) Copyright © 2015, 2015 IFAC 1939Hosting by Elsevier Ltd. All rights reserved. Copyright 2015 responsibility IFAC 1939Control. Peer review©under of International Federation of Automatic Copyright © 2015 IFAC 1939 10.1016/j.ifacol.2015.06.358
INCOM 2015 May 11-13, 2015. Ottawa, Canada 1866
Timothy Sprock et al. / IFAC-PapersOnLine 48-3 (2015) 1865–1869
has a different flavor than that implied by the AMS literature. Whereas control at ISA-95 levels 1 & 2 provide the capability to turn a server on and off, optimal control models at ISA-95 level 3 try to find the optimal operating policy, that is, rules for turning the server on and off that result in the lowest long-run cost (Tadj and Choudhury, 2005). This distinction separates the control responsibilities between ISA-95 levels 1&2 and level 3; level 3 is where operations research models can and should be implemented. If “smart manufacturing” is to be broadly realized, the requirements identified in ISA-95 for Level 3 controllers must be augmented with a specific implementable architecture that supports the real-time development of optimal policies and their integration with both Level 4 planning and level 2 execution systems. What is needed is a unified model encompassing both “plant”, i.e., the resources and processes controlled at ISA levels 2 and 1, and the level 3 controller. Our approach to addressing these challenges is to use a single modeling language (OMG SysML™) to construct a standard representation of the system that explicitly formalizes plant and control separation for the domain. The resulting reference architecture for smart manufacturing provides the missing bridge between system data and analysis models, i.e., between the ISA level 2 controller implementations and the ability to implement operations research models in the ISA level 3 controller. While there are existing models that capture the structure and behavior for the manufacturing plant (cf. section 2), there remains a need for an explicit model of operational control that can bridge between system models, analysis tools, and execution tools. In this paper, we discuss these challenges in more detail and derive a set of functional requirements for a controller capable of operational control for the smart manufacturing domain. The remainder of this paper is organized as follows: sections 1.1 and 1.2 discuss the challenges to smart manufacturing control in more depth. Then section 2 introduces a formal model of the base system, or plant, which supports the design and implementation of reusable analysis, including optimalcontrol analysis. Section 3 introduces a model of operational control that bridges between analysis and execution tools and explicitly incorporates the separation of plant and control. Sections 4 and 5 discuss the results of this modeling methodology and future directions. 1.1. Separation of Plant and Control In the literature on operational control, optimal control policies are designed using implicit models of the plant often coupled with simplified or idealized behaviors; e.g. using a simplified back-order clearing scheme to maintain tractability and develop efficient algorithms to solve the simplified inventory control problem (Deshpande et al., 2003). Conversely, resource investment models assume a default control policy, such as inventory allocation in the newsvendor formulation for supply chain operation. Historic separation between the specification of plant and control has resulted in poor interoperability between the two areas of research; e.g. whereas solving the newsvendor problem identifies an optimal
resource investment decision given a default allocation policy, the inventory rationing policy assumes a fixed resource investment. This can lead to a sub-optimal solution, where greater resource investment is suggested. Not only is there a modeling gap between formalisms but there is a theoretical modeling concern – the lack of explicit models of control separated, yet coupled to, models of the plant. Implementing operational control requires some external process to be applied to the base system, or plant, to guide its evolution. Controlled petri nets are a class of petri nets with external enabling conditions called control places, which allow an external controller to influence the progression of tokens in the net (Holloway et al., 1997). The switch place of an Extended Petri Nets uses information, or “control logic", external to the petri net model to resolve conflicts that arise[at the switch place] (Valavanis, 1990). In controlled Markov chains (Cassandras and Lafortune, 2008), action is taken to control the probabilities that affect the evolution of the chain. Similar to Markov chains, abstract state machines (ASMs) can implement a collection of `next-state functions' that would define a set of control functions for the domain and express how the transition system evolves (Reisig, 2012). These external functions contain the necessary logic for resolving control-driven questions, such as where to route a token next or which token to service next. However, there is not a definitive way to construct these functions as they “follow no general rules, but depend on the logic of the specific problem at hand and the desired system behavior" (Valavanis, 1990). 1.2. Gap Between Analysis and Execution Tools In next-generation smart manufacturing operations management systems, the execution tools such as MES must be tightly integrated with a suite of analysis tools which provide real-time optimal control feedback to the manufacturing environment. The challenge is that there are several broad formalisms for modeling this domain: mathematical (math programming and statistics), behavioral (simulation models), and IT-driven models (Min and Zhou, 2002). Each formalism has its own semantics, resulting in limited interoperability between analysis tools themselves and limited integration with execution tools. The mathematical representation of the analysis domain can be formulated by removing the domain specific semantics of the problem. In fact, the same abstraction process allows simulation models and IT-driven models to be mapped into mathematical models. However, this mapping is not bidirectional. The mathematical abstraction does not have behavioral semantics and thus cannot support simulation modeling without adding information. Simulation is important beyond its analysis capabilities because it provides a method for a controller to evaluate its control decisions on an internal model of the system before implementing the decision. Moreover, with the context of the problem stripped from the math programming formulation, it is challenging, perhaps precarious, to implement the output of the optimization. For example, if MES is designed to implement decisions based on the optimization analysis output, a vector X, then it will
1940
INCOM 2015 May 11-13, 2015. Ottawa, Canada
Timothy Sprock et al. / IFAC-PapersOnLine 48-3 (2015) 1865–1869
implement that decision regardless of the implicit content; e.g whether the output is for a resource assignment policy or a state-based admission policy. Because very few analysis tools are designed using the same formalism, the ability to directly exchange information between analysis tools is limited. There is an explicit modeling gap in this domain between system models, analysis models, and analysis tools. One possible way to bridge this gap is to construct an intermediate abstraction that provides a complete representation of the system and enables the construction of a translator between each of the native formalisms for each analysis model. 2. A SMART MANUFACTURING REFERENCE ARCHITECTURE At a high level, the operational control problem reduces to ``For given levels of capacity for available capabilities, how should a controller respond to requests for service, which require a specific capacity of a particular functional capability?'' This section briefly introduces an abstract reference architecture for smart manufacturing systems that provides a common, yet explicit, language to specify these control problems within the context of the system model. To construct a bridge between models of the system and a diverse collection of analysis models, an intermediate formalism at a high-level of abstraction is useful. The token flow network (TFN) (Thiers, 2014) provides an abstraction of the structure underlying many systems, including manufacturing systems, material handling systems, and supporting supply chains. The basic network definition constructed in the TFN captures the structure (nodes) and relationships between nodes (edges). This basic definition is extended to support the flow of physical goods, information, and resources through a network of flow nodes and flow edges. Finally, the network definition also supports conversions which take place at process nodes. These specialized nodes are organized by sequencing dependencies and form process networks which are essential for capturing the behavior of dynamic systems. The definition of the token flow network can be extended with the additional semantics needed to express the operational control problem, including a common description of the system, addressing its structure and behavior, its owned and shared resources, and its interaction with other systems. The structure of the plant is a logical aggregation defined by the product, process, resource, facility paradigm (Figure 1). This paradigm is intended to be a complement to other domain reference architectures such as CMSD (Lee et al., 2011), but provide
1867
a more abstract approach that may be applicable to domains beyond manufacturing. While most of the control activities that are discussed throughout this paper are focused on an individual system controlling its own plant, the system model also needs to incorporate the system’s interactions with the other systems in its ecosystem. Extensions of this nature are intended to be a bridge between formal models of relationships (Smith, 1980), such as client-server architecture, and soft models of relationships, such as the contracting literature from the operations management domain (Tsay et al., 1999). 3. A CONCEPTUAL FRAMEWORK FOR OPERATIONAL CONTROL IN MANUFACTURING SYSTEMS In addition to capturing the structure and behavior of the system plant, the reference architecture must also include a control model to provide interaction and influence over the activities executed in the smart manufacturing system. Each independent and controllable node in the system must have a controller, and while there have been several architectures proposed for the controller itself (Davis et al., 1992; Shirazi et al., 2010), to the best of our knowledge there has not been any attempt to formalize the definition of and provide a language for operational control for manufacturing systems. The new technical idea proposed here is a canonical abstraction, or language, for the design and analysis of control in smart operations management systems. Fundamentally, the controller must evaluate the current trajectory of its base system and decide when and which types of corrections are needed to maintain the system along its planned trajectory. To accomplish this task, the controller must select, formulate, and solve a set of control problems and then implement the resulting control action. To resolve ambiguous terminology in the literature, the abstraction proposed here incorporates a question-driven organization of the control problems that are addressed in the related literature (Section 3.1). To answer these fundamental control questions, the controller uses the abstract reference architecture for smart manufacturing systems discussed in Section 2 to construct a consistent definition of the control problem. This consistent problem and system definition enables a uniform interface to connect optimization tools to solve the control problem (Section 3.2), and a common expression and implementation of the output and solution to the control problem (Sections 3.3 and 3.4, respectively). 3.1 A Metamodel of Operational Control Questions
Figure 1: Abstract Plant Model
In order to be able to design the control mechanisms of these systems, we need to be able to describe the control activities that are being executed by the system controller at the operational management level. Considering the operations research and control literature, it can argued that a small collection of distinct control problems has been addressed. One description of these activities is the theory of controllable queueing systems which specifies the control of admission, servicing, scheduling, and routing jobs in queues and networks (Tadj and Choudhury, 2005). 1941
INCOM 2015 May 11-13, 2015. Ottawa, Canada Timothy Sprock et al. / IFAC-PapersOnLine 48-3 (2015) 1865–1869
1868
Admission control problems decide which tasks to service, possibly rejecting tasks to maintain system stability or short flow times. Service control problems, sometimes called dispatch, decide when a task gets serviced or which task gets serviced next. Scheduling control decides which resource will service a task and when. Finally, routing control decides where a task goes after it is done being serviced in this system. Also, there is at least one additional question that deals with changing the state of the resources in the system, including maintenance, set-up/tear-down, and re-positioning activities.
Which tasks get serviced? (Admission/Induction) When {sequence, time} does a task get serviced? (Sequencing/Scheduling) Which resource services a task? (Assignment/Scheduling) Where does a task go after service? (Routing) What is the state of a resource? (Services can it provide)
of the output of optimization solvers that adequately expresses the optimal control decisions to be implemented. 3.4 Optimal Control: Implementing Prescriptive Analysis Results The finite state machine (FSM) provides a formal specification for implementing the prescriptive analysis results; i.e., translating the output of optimization into control code. These models of computation are constructed from primitive definitions of states, transitions, events, and guards, which is important since the state/guard/transition paradigm is consistent and well-specified with respect to the definitions of policies. 4. RESULTS
Figure 2 A question-driven model of control However, one prominent issue is that the names of the control problems may vary by author or domain; whereas one author may call the control activity “servicing" another may call it “dispatching" which may also be used to refer to a “scheduling" policy. To resolve this discrepancy of word choice across the domain, a fundamental set of questions that the controller must be able to answer is proposed (Figure 2). 3.2 Formulating the Control Problem: An Interface to Solvers To support real-time, online decision-making where a control mechanism is required (or given the option) to solve an optimal control problem at any given decision point, the controller must maintain a representation of the current state of the system and be able to formulate the control problem to be solved. This formulation is then passed across an interface to an appropriate optimization solver. One solution to formulating optimal control problems that provides reusability and extensibility is to use the strategy pattern (Gamma et al., 1995) which defines a family of algorithms, encapsulates each one, and makes them interchangeable. The result is that the controller has a family of scheduling algorithms and can choose which solver to implement based on its environment or existing operational and performance constraints. 3.3 Optimal Control: Expressing Prescriptive Analysis Results Formulating the control problem to be shared with a solver typically requires translating the problem into mathematical semantics, thereby stripping away domain specific semantics from problem. This creates an obvious difficulty, in that the solution to the submitted problem will be returned in the same abstract semantics. Therefore, we need a formal specification
A canonical abstraction of operational control has the potential to create a bridge between system models, analysis methods, and implementation tools by establishing a common language for expressing the control problems in the smart manufacturing domain. This common language can be used to create an interface between system models and analysis models as well as enterprise-wide control integration. The language also provides an intermediate abstraction that facilitates interoperability between the distinct formalisms used by math programming tools and discrete event simulation tools. Finally, the resulting model of control links control activities at the enterprise level, including strategic and tactical planning, to the device-level control layer. 5. FUTURE WORK The framework proposed here is informed by important prior efforts, particularly the manufacturing system models represented by CMSD (Lee et al., 2011) and the Davis and Jones work on controller architecture (Davis et al., 1992). This paper extends and augments that prior work by: (1) proposing and explicit integration of plant and control through a bridging abstraction; (2) proposing a novel approach to specifying the functional requirements for ISA-95 level 3 optimizations; and showing how a unified semantic model can be created using OMG SysML™. Clearly, a great deal of work remains to be done, to formalize the proposed framework, to create instantiations of the decision support optimizations, and to demonstrate the resulting control implementations. What is not in question is the need for a more comprehensive modeling abstraction which integrates plant and control, in order to achieve smart manufacturing, and provide the foundation for smart manufacturing in general. REFERENCES Cassandras, C.G., Lafortune, S., 2008. Introduction to discrete event systems. Springer. Davis, W., Jones, A., Saleh, A., 1992. Generic architecture for intelligent control systems. Comput. Integr. Manuf. Syst. 5, 105–113. Deshpande, V., Cohen, M.A., Donohue, K., 2003. A threshold inventory rationing policy for service-differentiated demand classes. Manag. Sci. 49, 683–703.
1942
INCOM 2015 May 11-13, 2015. Ottawa, Canada
Timothy Sprock et al. / IFAC-PapersOnLine 48-3 (2015) 1865–1869
Dilts, D., Boyd, N., Whorms, H., 1991. The evolution of control architectures for automated manufacturing systems. J. Manuf. Syst. 10, 79–93. Galloway, B., Hancke, G.P., 2013. Introduction to industrial control networks. Commun. Surv. Tutor. IEEE 15, 860–880. Gamma, E., Helm, R., Johnson, R., Vlissides, J., 1995. Design Patterns - Elements of Reusable Object-Oriented Software. Addison-Wesley. Holloway, L.E., Krogh, B.H., Giua, A., 1997. A survey of Petri net methods for controlled discrete event systems. Discrete Event Dyn. Syst. 7, 151–190. Lee, Y.-T.T., Riddick, F.H., Johansson, B.J.I., 2011. Core Manufacturing Simulation Data – a manufacturing simulation integration standard: overview and case studies. Int. J. Comput. Integr. Manuf. 24, 689–709. doi:10.1080/0951192X.2011.574154 Min, H., Zhou, G., 2002. Supply chain modeling: past, present and future. Comput. Ind. Eng. 43, 231–249. Ramadge, P.J.G., Wonham, W.M., 1989. The Control of Discrete Event Systems. Proc. IEEE 77, 81–98. Reisig, W., 2012. The expressive power of abstract-state machines. Comput. Inform. 22, 209–219. Shirazi, B., Mahdavi, I., Solimanpur, M., others, 2010. Development of a simulation-based intelligent decision support system for the adaptive real-time control of flexible manufacturing systems. J. Softw. Eng. Appl. 3, 661. Smith, J., Joshi, S., Qiu, R., 2003. Message-based Part State Graphs (MPSG): a formal model for shop-floor control implementation. Int. J. Prod. Res. 41, 1739– 1764. Tadj, L., Choudhury, G., 2005. Optimal design and control of queues. Top 13, 359–412. Thiers, G., 2014. A Model-Based Systems Engineering Methodology to Make Engineering Analysis of Discrete-Event Logistics Systems More CostAccessible. Georgia Institute of Technology, Atlanta, GA. Valavanis, K.P., 1990. On the hierarchical modeling analysis and simulation of flexible manufacturing systems with extended petri nets. Syst. Man Cybern. IEEE Trans. On 20, 94–110. Vogel-Heuser, B., Witsch, D., Katzke, U., 2005. Automatic code generation from a UML model to IEC 61131-3 and system configuration tools, in: Control and Automation, 2005. ICCA’05. International Conference on. IEEE, pp. 1034–1039.
1943
1869