Scholarly article on topic 'Agent-based manufacturing execution systems for short-series production scheduling'

Agent-based manufacturing execution systems for short-series production scheduling Academic research paper on "Computer and information sciences"

CC BY-NC-ND
0
0
Share paper
Academic journal
Computers in Industry
OECD Field of science
Keywords
{MES / "Manufacturing execution systems" / "ISA 95" / "Intelligent agent" / "Short-series production management"}

Abstract of research paper on Computer and information sciences, author of scientific article — Rafal Cupek, Adam Ziebinski, Lukasz Huczala, Huseyin Erdogan

Abstract This paper presents the architecture of Agent-based Manufacturing Execution Systems dedicated for short-series production support. The functional models are based on the ANSI/ISA-95 (IEC/ISO 62264) standard. The workflow and information exchange for Manufacturing Operations Management are defined by ISA 95 and implemented under a dynamic Agent-based environment. The proposed system is organised as a fully heterarchical architecture, without a central administration or system orchestrators. Unlike most of the existing agent software that are based on Java, the proposed solution is based on Microsoft’s Model-View-Controller and was created under the ASP.NET technology. Holons, which collect information from the real production system, are a Cyber Physical part of the application. Agents process information using Internet services that are available for human users and for the other agents as well. The proposed approach has been verified on the use case of the system that was created to support the production of electronic devices in the Prototyping Department of Continental Ingolstadt. The system model, applied communication mechanisms and examples of agents are presented in this paper. The research part of this paper is focussed on simulation-based planning for a short-series production schedule. The simulation results can be used to support the decision-making process.

Academic research paper on topic "Agent-based manufacturing execution systems for short-series production scheduling"

Contents lists available at ScienceDirect

Computers in Industry

journal homepage www.elsevier.com/locate/compind

Agent-based manufacturing execution systems for short-series production scheduling

Rafal Cupek^*, Adam Ziebinskia, Lukasz Huczalaa, Huseyin Erdoganb

a Silesian University of Technology, Institute of Informatics, Gliwice, Poland b Conti Temic microelectronic GmbH, Ingolstadt, Germany

CrossMark

ARTICLE INFO

ABSTRACT

Article history:

Received 14 December 2015

Received in revised form 14 July 2016

Accepted 22 July 2016

Available online 3 August 2016

Keywords: MES

Manufacturing execution systems ISA 95

Intelligent agent

Short-series production management

This paper presents the architecture of Agent-based Manufacturing Execution Systems dedicated for short-series production support. The functional models are based on the ANSI/ISA-95 (IEC/ISO 62264) standard. The workflow and information exchange for Manufacturing Operations Management are defined by ISA 95 and implemented under a dynamic Agent-based environment. The proposed system is organised as a fully heterarchical architecture, without a central administration or system orchestrators. Unlike most of the existing agent software that are based on Java, the proposed solution is based on Microsoft's Model-View-Controller and was created undertheASP.NET technology. Holons, which collect information from the real production system, are a Cyber Physical part of the application. Agents process information using Internet services that are available for human users and for the other agents as well. The proposed approach has been verified on the use case of the system that was created to support the production of electronic devices in the Prototyping Department of Continental Ingolstadt. The system model, applied communication mechanisms and examples of agents are presented in this paper. The research part of this paper is focussed on simulation-based planning for a short-series production schedule. The simulation results can be used to support the decision-making process. © 2016 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND

license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction

Nowadays, control systems provide very detailed information about the underlying production process. This information is further used by Business Intelligence systems, which are localised on the Enterprise level. Moreover, decisions taken on the business level have to be executed by control systems. Manufacturing Execution Systems (MES) are service-oriented interfaces that connect the world of business operations with the world of production. The classical MES are defined by a static hierarchy of services and data structures, which makes them very difficult to modify. The change of the production model from mass manufacturing to customised manufacturing and short-series production presents new challenges to MES [1]. The emergence of the concept of Cyber Physical Systems (CPS) enforces changes in the architecture of MES. The automation pyramid is no longer a canon in industrial IT systems. CPS are formed by networks of distributed operating entities providing the production process.

* Corresponding author. E-mail address: rcupek@polsl.pl (R. Cupek).

MES must follow this idea through new architectural solutions. A new approach to the architecture of MES is also indispensable to support agile manufacturing. The authors propose a heterarchical MES architecture based on a multi-agent system that is designed to work inside and cooperate with CPS.

In the case of mass production, there was enough time to find the best scenario for the production process in terms of the best production technology and manufacturing operations schedule as well as to find the optimal setup parameters for particular production devices. In addition, the manual actions performed by operators were more stable and the final results more repeatable. In the case of short series manufacturing, the production technology is often changed, production tools have to be adjusted to specific products and the process organisation must follow these changes in order to avoid or reduce losses resulting from non-productive time gaps [2]. Moreover, time-to-market and product development time have become critical aspects of innovation processes. In such a case, the benefits from MES that support manufacturing can be far more important for enterprises than in the case of mass production. CPS improve the availability of information about the progress of the production process that is required by MES. On the other hand, MES can help CPS in the

http://dx.doi.org/10.1016/j.compind.2016.07.009

0166-3615/© 2016 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/ ).

planning and the organisation of the manufacturing. A proper cooperation between MES and the CPS is particularly important in the case of short series manufacturing.

Research results show that trade-offs are made not only between time, quality and expense but also that trade-offs relate to when additional development expenses are incurred, including cross-functional integration (both internal and external) that substantially impacts on product profitability through a mix of direct and mediated effects [3]. Such integration cannot be reached in an effective way with MES support. The above-mentioned factors mean that Manufacturing Execution Systems should follow the changes in manufacturing. MES should no longer be a closed, fixed IT software, but must be created as flexible and open sets of services that interact with the physical production system. Contemporary MES should follow real production and should be self-adaptive in order to support changes in short series manufacturing. The authors propose to improve MES adaptability by changing its architecture from hierarchical to heterarchical software based on holons and agents.

Cyber Physical Systems are places in which the embedded world meets the Internet world [4]. They deploy embedded cyber capabilities and join them with the physical world, including humans, infrastructure and platforms, which transform interactions with the physical world. In the case of MES dedicated for short-series production, there are also different kinds of actors. Human users such as production managers, the staff involved in production optimisation, the logistic team or quality managers are interested in fast and precise information about production progress, realisation of orders and possible production problems. Other actors are production facilities such as machinery and equipment, which need the information necessary for the effective realisation of the production process and the proper validation of the created products. The products themselves are also active participants in the system since they collect valuable information about the actual production parameters that can affect product utilisation and its future development.

CPSs are glued by web services that are available via the Internet. Cyber Physical Systems have the ability to interact with and expand the capabilities of the physical world through computation, communication and control. They are key enablers for future technology developments [5]. Although some MES functionalities are realised internally, some services need to interact with suppliers and customers and have to be available externally. Such an interaction is especially important in the case of short-series production that needs closer cooperation between suppliers, producers and consumers since the production chain must be more flexible than is the case in mass production. The proposed MES architecture is based on Internet services and binds them with CPS to support both the production and later the use of a product. This paper focuses on the new opportunities and research challenges related to agent-based MES architecture. The authors propose agent-based architecture for flexible and heterarchical MES. The proposed MES architecture is composed of agents that offer and execute virtual services and holons, which are physical process interfaces that ensure the materialisation of services. Together they build a bridge between Cyber Physical and Manufacturing Execution Systems.

The novelty of the presented approach is the new model of MES that are based on a triple-layer heterarchical network of agents that perform the required services and join the physical production environment with high-level decisional systems. The internal layer of this model is close to the physical part of the system and reflects the requirements of the products and production processes. It ensures links to the production devices by means of holons. The middleware represents the dependencies between processes and services. The outer layer forms an external interface to other

systems and human users. Such an approach is more flexible and more resistant to implementation errors. Each order is under the care of autonomous agents that support the human participants from one side and represents them in the cyber world. The authors present an activity model of the proposed system. In order to facilitate its application, the model complies with the third part of ANSI/ISA-95 (IEC/ISO 62264-3) [6] standard. The proposed MES architecture has been verified practically on the real example of the production system used by the Prototyping Department of Continental Ingolstadt.

The new functional model requires a new approach to system architecture. Since individual orders are executed independently and have different product supervisors, it is very difficult to take arbitrary decisions in the event of conflicts when accessing resources. In the case of the production that is carried out by the Prototyping Department of Continental Ingolstadt, the classical MES models that are used by Conti's other departments cannot be applied. The main obstacle is that the background knowledge of the production staff cannot be dynamically included in the business models that are represented by MES. In the classical solutions, every change in a model requires a reconstruction of the responsible part of the MES software. In the analysed use case, this is not possible due to the number of variants of production -practically every order is considered as another variant of production. In such a case, a central decision support system must be replaced by local support and the central system must be replaced by the appropriate IT architecture. The authors decided to apply an agent/holon-based solution. Such an architecture, which is described in section 3 and illustrated by the use cases in section 4, allows human users of the system to benefit from distributed schedule planning through a simulation of the execution of the production schedule. From another side, the short-series production that is being considered requires many changes in its organisation. Since, there are different possible variants of production, the holon-based model is more flexible and scalable than the central model of the production that is based on a static facility layout.

This paper is organised as follows. Section 2 introduces the related works, including agent-based MES and interfaces between MES and ERP systems. Section 3 presents the proposed architecture of the system. It describes MES activities using the ISA95 models. It also gives some more details about the implemented business models, the architectural models of agents and holons and presents some details of agent-agent and agent-human communication. It also presents some details related to the application of MES for the electronic device prototype production lines. Section 4 focuses on the system application for the simulation of the short-series production carried out by the Prototyping Department of Continental Ingolstadt. The simulation results give some performance metrics of production that can be used for planning manufacturing activities. Finally, section 5 presents the conclusions.

2. Related work

Nowadays, the rapidly changing environment requires rapid changes in manufacturing systems. Industries must adapt their manufacturing systems to maximise their productivity and the profitability of production. Customers increasingly require a shorter time to market. The changes include shorter product life-cycles, increasing requirements for quality, increasing the customisation of products, the faster implementation of advanced technology and optimising the cost of energy. These expanding options affect materials, processes and interfaces to product models and often the resulting products must be produced in a number of variants. In this section, the authors review the

requirements of the computer-aided manufacturing systems with a special focus on MES functionality. The authors compare MES conceptual models given by standardisation organisations with new architectural concepts. The authors place a special emphasis on a research to create a heterarchical agent and holon-based architectures that support the functionality of the execution of manufacturing. The traditional design of the IT architecture used by industry is hierarchical and cannot easily be adapted to changes that are implemented in production. Such problems have also been found at Continental Ingolstadt. Continental uses a classical MES to support mass production and therefore any change in the underlying process forces the introduction of new patches in the MES. This means that the existing system cannot be used by the Prototyping Department. This is also a challenge for other production departments. One of the objectives related to product customisation is "One part for production", which means that all, even a very short series of products, should be profitable. This has forced research on a new architectural approach. Holons and agents are one of the options that are being considered to meet the requirements of short-series production.

An Advanced Manufacturing Production System is capable of furnishing a mix of products in small or large volumes with both the efficiency of mass production and the flexibility of custom manufacturing in order to respond rapidly to customer demands and the desired quality [7]. Based on the perspective of 2025 [2] "the production of goods and services will therefore have to address mass customisation and become localised and networked to be closer to customers, to respond to local demand and to decrease costs", these trends have been recommended as critical paths.

Modifications to dimensions, functions and materials in products or components after the product design has been released are defined as Engineering changes (ECs) [8-10]. Manufacturers establish engineering change management (ECM) that controls the processes and associated product data for ECs in order to maintain product data consistency. ECs are intensive activities, but their efficiency is low. Engineers spend 30-50% of their efforts on ECM and only 8.5% of their efforts represents value-added activities [11]. A direct link between scheduling and execution is required in automated production systems. Enterprise Resource Planning (ERP) systems do not allow real time control of production operations. MES allow a link to real time control devices to be configured, although they must operate in conjunction with other systems. Additionally, MES have a great overlap in the functionality that is realised via detailed scheduling. Advanced Planning Systems (APS) allow this problem to be avoided due to the capacity to be linked directly to the real-time control devices. The same functionality can be obtained by using agents and holons, which additionally allows the dynamic reconfiguration of machines to occur. Engineering changes (EC) allow the configuration of the machines to be changed, although making improvements and changes often takes a long time.

The assembly process quality control allows the quality of the products to be ensured based on the data about the materials, process and errors. The errors may be propagated to the next process so the quality of the products may have large fluctuations [12,13]. An effective method to improve the quality of products is assembly process adaptive control [14]. In order to better solve the problem of production line stops to change material for orders or an error, a system could be developed to allow selfish behaviour and adaptive decision-making in distributed execution control and emergent scheduling [15,16].

Advanced Planning Systems (APS) are required in production companies. These allow scheduling and production planning to be managed in order to optimise human resources and materials [17].

The implementations of an APS in a company system are often made with ad hoc applications that complement several functionalities of the existing Enterprise Resource Planning (ERP) systems [18-20]. In areas where information automation is possible, ERP provides an integrated platform to manage the business. Some degree of system customisation is required for ERP systems [21,22].

The traditional design of manufacturing control systems is hierarchical and uses the top-down method that defines the functions and modules in which the modules can only communicate with their parent and child modules. A hierarchical control system cannot react to changes and therefore hierarchical control cannot achieve adaptable control. Using technologies, holons and multi-agent systems allows these problems to be avoided.

However, agile and flexible production in high-tech and knowledge-intensive industries cannot be achieved without flexible MES that are able to support frequent changes in the production profile and the realisation of the many production variants that are characteristic in mass-customised production [23]. A formal specification framework for manufacturing execution systems can be found in [24]. Moreover, a holistic approach to the optimisation of mass production customisation at the MES level is indispensable. Currently, there are few holistic models for MES. There are a number of standards in industrial automation, which makes it difficult to analyse the whole enterprise perspective starting from the machine level, production line level or the shop floor level. A classical MES is defined by its static hierarchy, which makes it very difficult to modify them [25]. This has required research on a new approach to the architecture of an MES that has to manage the huge stream of information that is exchanged between business and production systems.

MES were adopted by industry as an "on-line extension of the planning system with an emphasis on executing or carrying out the production with a planned or sequenced list of Work Orders, methods to schedule those Work Orders into Work Stations, control of inventory assignment and management of material movement" [26,27]. In the classical model of an automation pyramid, the MES is placed as a set of interfaces between the physical production subsystems and the high-level decision systems [28]. Today, MES are one of the main factors in enterprise integration

The commonly accepted definitions of the functions and data managed by MES can be found in the set of documents that is managed by MESA International (Manufacturing Enterprise Solutions Association), which is a worldwide not-for-profit community of manufacturing companies, information technology hardware and software suppliers, system integrators, consulting service providers, analysts, editors, academics and students [29]. According to MESA's definition, an MES supports production in the following activities: job scheduling, launching the orders, responding to random events, adjusting production plans, tracing product genealogy, managing production quality and managing maintenance activities. The above-mentioned areas are systematised into conceptual and functional models, which are described the in ANSI/ISA-95 (IEC/ISO 62264) norms that are the international standard for the integration of enterprise and control systems [30].

ISA95 models are an interface between the business processes and manufacturing processes. It separates these processes through a clear demarcation of responsibilities and functions and joins them via a well-defined communication interface that is realised by Business to Manufacturing Markup Language (B2MML), which is an XML implementation of the ANSI/ISA-95 family of standards. It also defines the functions and information flow inside an MES system. ISA95 defines a common terminology and a consistent set of models that are described by UML diagrams, which are

understandable in both domains. ISA95 allows the main challenges in manufacturing to be addressed:

• The move from the vertical enterprise model to the production-oriented horizontal model where manufacturing is controlled by the removal of a product

• Service-based manufacturing where production is carried out by anonymous chain service providers

• Production dynamisation where market volatility makes it necessary to continuously adjust the service profile

• Adaptive manufacturing where machines are connected directly to the market

• Production individualisation on the customer's market where products are individually tailored to the needs of the customer

On the conceptual level ISA95 refers to the classical production system pyramid and addresses its different aspects via five parts as shown in Fig. 1.

ISA-95 defines the MES data structure and MES services that are related to manufacturing operations: defining the product, forecasting production, managing production capability and evaluating production performances. ISA-95 consists of models and terminology and describes the information that is exchanged between the systems for sales, finance and logistics and the systems for production, maintenance and quality. This information is structured in the form of UML (Unified Modelling Language) models, which are the basis for the development of standard interfaces between ERP and MES systems.

ISA-95 is built on the object-oriented model that defines the interface between control systems and business application. It also defines the services that are required for the manufacturing support that is designed according to the object-oriented model [30]. These services are not only based on information exchange but also on the aggregated data or the history of the realisation of the process that has to be managed by database systems. The connection between MES and control systems is outside the scope of ISA95. This link can be established by different communication standards. In order to find the most flexible solution that can be used in the case of short-series production, this work is focussed on the holonic manufacturing concept.

A manufacturing control system for production processes is composed of software modules as well as the different physical elements of the manufacturing environment. The software module and the physical entity, which are bonded by means of an appropriate communication network, represent a holon in a manufacturing system. Each holon is able to reason, make decisions and communicate interactively with other holons. The first holon general architecture was proposed by Christensen [31] in 1994. Holonic manufacturing is based on an autonomous and cooperative entity called a holon. Holons can be used to separate the physical processing itself, that is, the hardware that executes the manufacturing together with the related control software from the supervisory control activities that are necessary for the

effective change of the production profile. A holon combines the advantages of hierarchical and heterarchical organisational structures [10]. It can provide the adaptability and flexibility of heterarchical control by reacting to changes. It also allows the stability of hierarchical control to be maintained [14]. The adaptive properties mean that holons can be responsible for the direct implementation of production variants and the direct support of the rapid and consistent reconfiguration of the machines, robots and supervisory control and data acquisition tools. The example types of holons and the associated knowledge are presented in Fig. 2.

PROSA (product, resource, order, staff, architecture) is one of the reference architectures for a Holonic Manufacturing System (HMS) [32]. A resource holon is based on the characteristics of the tasks that they perform, e.g. transformation or assembly—the product holon reflects the operations performed on raw materials, semi products and finally the end products and the order holon represents requirements defined by the customer. For each type of resource holon, specific heuristics are used to solve the problem of scheduling tasks. These mechanisms are implemented within a multi-agent system, which supports the development of the manufacturing control system.

Based on the idea of a holonic approach [33] and intelligent resources [34] that started with the PROSA project, the new service-oriented MES architecture should be based on distributed services that are realised as an object-oriented model [35] that is supported by self-descriptive and object-oriented communication protocols [25]. In [36] we can find an example of the development of a robot control system for intelligent manufacturing in which the Conceptual Holonic Model was used for the control software. A networked robotised job shop assembly structure composed of a number of robotic resources that are linked by a closed-loop transportation system was described in [37].

Agents that are organised into a heterarchical structure with a high-level of autonomy and co-operation based on the clientserver structure with no fixed relations [38]. This concept allows for high performance against disturbances through global optimisation. The decision making is autonomous and local, without a global view of the system. The functioning of some agents can be modified or new agents can be added to the control system for the expansibility of the system. For the optimisation of time scheduling and resource allocation in various domains of the production [39], the agent-based production planning approach can be applied. The collaboration of the autonomous agents and planning mechanism allows efficient resource management during the production of goods in many manufacturing processes. The efficiency of the production devices can be increased by

Fig. 1. Production system model and ISA 95 scope.

Fig. 2. Holons types and associated knowledge.

calculating the workload and directing the most appropriate device in a specific product line. This result can be achieved through coordination between the planning and production agents.

A Multi-Agent System (MAS) is the most widely used tool for mainstream Holonic Manufacturing System (HMS) research. Agent technology is used to implement HMS applications. The multiagent system paradigm comes from distributed artificial intelligence (DAI) and is characterised by the decentralisation and parallel execution of activities that are based on autonomous entities (agents). It is generally assumed that an agent is an "autonomous component that represents physical or logical objects in the system, able to act in order to achieve its goals and able to interact with other agents, when it does not possess the knowledge and skills to reach alone its objectives".

Holons and agents are very similar concepts [40] and they increase the adaptability and robustness of systems. They are characterised by concepts such as autonomy, proactiveness, coordination and communication in order to adapt to the environment. The differences between HMS and MAS are related to how the information is managed. Physical and information management is clearly separated in holonic systems, but MAS do not distinguish them. Recursivity is very typical of holonic systems, while mobility is common in MAS technologies. Agents in MAS can cooperate or compete in order to achieve their goals, whereas holons always cooperate to solve a problem. MAS allow decentralised and distributed systems to be designed and implemented in which each agent can take its own control decisions and modify its behaviour using information from the environment or its previous experience. Agent-based approaches have been applied in many different areas, such as process control, control manufacturing, electronic commerce, telecommunications and others.

New concepts, which are based on the service orientation of holonic manufacturing systems, can be found in Service Oriented Architecture (SOA), Enterprise Service Bus (ESB), Web Services, Manufacturing Service Bus (MSB), Distributed Intelligence (DI) and product-driven automation, Service-oriented Multi-Agent Systems (SoMAS) and the resource service access model (RSAM). The holonic component-based approach (HCBA), which was proposed by Chirn and McFarlane [41], is a similar approach. The HCBA defines resource holons, product holons and work-in-process (WIP) agents that correspond to order holons. HCBA allows the static and dynamic integration for HMS to be achieved using component-based development. HMS is considered to be an autonomous, cooperative and intelligent component. HCBA is concentrated on the reusability and the reconfigurability of HMS. Logistic systems play a crucial role in the organisation and management of every large enterprise.

In an MES, software agents can perform their tasks for production planning support. The autonomous system is capable of preparing planned material requests and production schedules by examining alternative variants. It also supports the concern about energy cost savings by taking into account the electricity time zones for a plant and the workload of the factory. Moreover, the scheduled production time can also be presented to customers.

An example of an MES application as an HMS and MAS is presented in [25]. A manufacturing system that produces laminated bullet-proof security glass that is ready-to-assemble on vehicles was described. Software agents work collaboratively to assist in the production, planning and sales departments for the generation of the production plans. A multi-agent oriented manufacturing automation architecture, which was focussed on features of the ISA95, was described to facilitate the managing and control of smart automation components in a distributed manufacturing environment [42].

We can find holonic MES that are based on the PROSA architecture augmented with coordination and control mechanisms inspired by natural systems based on stigmergy and future state prediction in [43,44]. The ADACOR MES switches its decisionmaking to a heterarchical structure [45]. We can find a hybrid control architecture that is modified according to new processes and equipment items based on the concept of coalitions of manufacturing components in [46]. Holonic MES require more effort to be set up, but substantially less effort to be reconfigured [26]. A distributed (re)scheduling system can be based on interactions between order and resource agents acting as autonomic managers [47]. Autonomic agents are created from order or resource agents and allow the role of the autonomic manager to be separated from the object being managed. The objects corresponding to physical machines, parts or final products are usually allocated as control functionality by MES. Agent-based MES architecture for Quality Management support and example of it practical realisation in automotive electronics device manufacturing is presented in [48]. Other very important issue is a communication between localised on factory floor level CPS and software operating on business application level. Many of interconnection problems may be solved by application of an object oriented communication standards like OPC UA [49].

The study on the related work presented above clearly states that efficient data exchange is one of the main issues that significantly affect adapting manufacturing systems to the new requirements. A hierarchical control system cannot react to changes and does not allow problems in factory management to be recognised quickly. It communicates with its parent and child modules and does not allow problems in factory management to be recognised quickly. Holons and multi-agent systems allow this problem to be solved. Holons can be used as interfaces between the software platform and existing production tools, which will allow information about various aspects of a production system to be obtained. Multi-agent systems allow data mining algorithms to be used to explore the information that is available in a production system. Therefore, it is possible to perform parallel analyses of various aspects of the production system, including predicting system behaviour, refining the production model, processing and classifying information and preparing supervisory control and data acquisition tools. The presented successful implementations show that the new architectures for manufacturing control systems permit the application of multiagent systems that allow optimisation procedures to be prepared for production by using machine learning methods. Thus, the system allows for the dynamic reconfiguration of the machines and robots.

3. Activity model for agent-based MES for short-series production

Manufacturing becomes more and more complicated when production series are short, products are diversified and production technology is variable. In the case of short-series production, Manufacturing Execution Systems join the cyber physical part of production with virtual manufacturing services and business level operations. Unfortunately, most of the existing architectures follow the paradigm of hierarchical MES placed between the control systems and business application level. They have fixed interfaces to production facilities and a predefined set of services. The main argument in favour of such a solution is the global optimisation of manufacturing operations, but in the case of highly dynamic, short-series production, it is practically impossible to find a global optimum of the production process. Moreover, in the case of agile manufacturing, changes in the production scenario incur costly modifications of the MES services and interfaces. Such

ad hoc changes can generate new errors that increase the total cost of production.

Manufacturing Execution Systems must have the ability to continuously adapt to changes in production process and therefore autonomous intelligent manufacturing systems have become more complicated. Several critical issues are related to modularity and system integration. The use of technology holons and multi-agent systems allows an existing production-driven MES that is dedicated for mass manufacturing to be changed into a product-driven MES that is dedicated for customised mass production and short-series production. The traditional design of manufacturing control systems does not allow for rapidly expanding options in materials, processes, interfaces to product models that have a number of variants. Holons allow the implementation of production variants and the rapid reconfiguration of the machines and robots. Proposed MES architecture supports the capability of production systems. Demand chain management will allow for a significant reduction in waste and will increase the profitability of production systems. Flexible production planning will effect in reducing the setup and changeover time and costs. The strategic target of the proposed MES solution is "one piece flow production", which means the feasibility of short-series production (up to one element) by using the production lines that were previously designed for mass manufacturing.

Like all industrial solutions, the proposed MES also has to fit the existing standards accepted by industry. The authors define an MES functional model that complies with the third part of the widely accepted ANSI/ISA-95 (IEC/ISO 62264) standard. Since ISA95 represents a hierarchical vision of MES, this model has been adapted to the heterarchical agent-based architecture. ISA 95 defines the workflow and information exchange for Manufacturing Operations Management. This includes the structure of manufacturing management functions and their interactions with business systems. ISA95 defines these functionalities at a level below enterprise business systems, but above manufacturing control systems and they consist of five parts.

The first partofISA95 "Models and Terminology" [50], describes the interface content between manufacturing control functions and other enterprise functions including the necessary interfaces between an MES and ERP systems. The second part, "Object Model Attributes" [51], defines the object-oriented data models expressed in UML (Unified Modelling Language) diagrams. It standardises the information structure used for the description of personnel, equipment, physical assets, materials and process segments that are defined as the logical group of personnel, equipment and material resources required to perform a given manufacturing operation step. It also standardises the data structures that are used for the definition of the manufacturing operations, production schedule planning, and performance information assessment and production capability verification. The third part of ISA95 defines the "Activity Models of Manufacturing Operations Management" [6]. It includes object-oriented activity models for production operations management, maintenance operations management, quality operations management and inventory operations management. The fourth part, "Object Models and Attributes of Manufacturing Operations Management" [52], defines the information exchange between manufacturing operations. It includes the object model attributes, resource relationship network model, work definition model, work schedule and job list models, work performance model, work capability model, KPI model and work alerts model. The fifth part, "Business to Manufacturing Transactions" [53], specifies how to exchange the objects defined in Part 2 and Part 4. It defines the details of transactions in the form of messages and the verbs used for communication.

The proposed MES functionality will be presented in more detail on the example of the activity models defined by ISA95 for Manufacturing Operations Management. This model distinguishes four main streams of information related to defining the product, determining the production capability, managing the actual production and evaluating the performance metrics. This information flow is collected in and managed by eight main activity sets as shown i Fig. 3. Very similar schemes are used for maintenance operations management, quality operations management and inventory operations management. The eight main activities for this model as defined by ISA95 [6] are listed below.

1. The product definition management activity focuses on information exchange with engineering, R&D and others to develop the site-specific product production rules. This information may include R&D manufacturing definitions that are translated and extended by product definition management into site-specific definitions using local material, equipment and personnel.

2. Production resource management is a collection of activities that manage the information about the resources required by production operations and relationships between resources. The resources include machines, tools, labour, materials and energy. Management of the resources may include local resource reservation systems to manage information about future availability. Information about resources and the relationships between the resources needed for a segment of production must be maintained and provided on an available, committed and unattainable capacity for specific periods of time of the specified resources.

3. Detailed production scheduling is the collection of activities that create the production schedule and determine the optimal use of local resources to meet the production schedule requirements. This includes ordering the requests for minimal equipment set-up or cleaning, merging requests for the optimal use of equipment and splitting requests when required because of batch sizes or limited production rates. Detailed production scheduling takes into account local situations and resource availability.

4. Production dispatching manages the flow of production by dispatching production to equipment and personnel.

5. Production execution management is focussed on the performance of work as specified by the contents of the job list. The production execution management activity includes selecting, starting and moving those units of work (for example, lots, sublots or batches) through the appropriate sequence of operations to physically produce the product.

Operations Operations Operations Operations definition capability request response

_Detailed

__ '__ \scheduling

f Resource t

\managementy \

> > ( Dispatching

i f Definition^ I

\managementy 1

I f Execution X^management

Fig. 3. Activity model of production operations management [7].

6. Production data collection gathers, compiles and manages production data for specific work processes or specific production requests. The managed data may include sensor readings, equipment states, event data, operator-entered data, transaction data, operator actions, messages, calculation results from models and other data of importance in the making of a product.

7. Production tracking prepares the production response to ERP systems. This includes summarising and reporting information about the personnel and equipment actually used to produce a product, the material consumed, material produced and other relevant production data such as costs and performance analysis results.

8. Production performance analysis is focussed on the analysis and reporting of the performance information exposed to business systems. This include analysis of the information about production unit cycle times, resource utilisation, equipment utilisation, equipment performance, procedure efficiencies and production variability. Relationships between these analyses and others may also be utilised to develop KPI reports. This information may be used to optimise production and the use of resources. Such information may be provided on a scheduled basis, it may be provided at the end of production runs or batches or it may be provided on demand.

Although the Manufacturing Operations Management model defined by ISA-95 reflects the most typical services performed on the MES level, it presents them like an external system activity. The information flow is defined as hierarchical and fixed communication streams binding cooperating services in a static way. In fact, such services exist in real production systems, but they are not static. They emerge from the fractioned activities of the many actors participating in the production process. Information streams presented in ISA95 diagrams are composed of many minor facts and events messages exchanged at different steps of manufacturing in practice. Together they form the overall picture given by the ISA95 model, but they do not exist as excluded or independent service sets or communication paths. In such a case, an external MES model can only reflect a selected set of production use cases, but it cannot reflect the actual manufacturing activities. This means that the classical MES can only be used in mass production where the production system topology is relatively fixed; it is not applicable in short-series production where both the production environment and the production rules are highly dynamic.

In order to reflect the actual paths of the realisation of particular services, the authors propose applying an agent-based model for the creation of an MES system. The agents do not form any fixed hierarchical structure as proposed in the ISA95 diagrams, but the system's architecture emerges from the network of cooperating agents that reflect physical activities of the actors involved in the production process. The heterarchical structure of the system arises during the negotiation and establishment of the system's services. The services are started as a result of manufacturing activity on the request from the ERP system or a human user. The established services create links between agents and those links form streams of information exchange that are composed of event messages and query - answer communication. Particular information links remain as long as a given service is useful. Then the service is ended and the information stream disappears. Although the internal network of agents and their cooperation paths is flexible and changeable during the realisation of production, external system's services are well known and reflect the ISA95 models. Obviously, the activities defined by ISA95 are performed, but they are not based on an external hierarchical structure, but emerge from the physical production activities that are reflected and modelled by the agents. This model is presented in Fig. 4.

Fig. 4. Agent-based model of a Manufacturing Execution System.

To separate fixed and well-known system interfaces from a variable network of agents, the model is divided into three spheres. The actual production is performed by real production facilities that are far more fixed compared to the middleware agents that represent the actual manufacturing operations execution. This forms the most internal part of our model that is closely coupled with the physical part of the system. We designed it using holon-based architecture. Holons allow production resources to be split and then merged in a flexible way. It also allows for the easy adaptation of an MES in real production system features. From the other side, holons explore the services that are available in the physical production system of the middleware part of the model. The holons themselves can be composed of other holons and this structure reflects the real dependences that occur in a production system and the interactions between personnel (LR), equipment (MR), materials (M) and process segments composed of personnel, materials and equipment.

The services required on the workplace level are explored by WPA (Work Place Agents). It is possible that a given WPA is not available or that a new WPA was added. This reflects a real situation in which a given machine is broken or a new one was added. However, we make a general assumption than the features of production equipment change far less often than the requirements for production orders so it is worth separating the internal and middleware levels of agents. The information services covered by WPA include the features required for a given production site, as defined by ISA95 on the UNIT's, WORK CELL's and STORAGE UNIT's levels. WPA themselves are created as collections of holons (holarchy) that are created by resources used at a given workplace. Although the internal structure of workplace can be changed (for example, by machine modernisation or changes in the way that the production process is realised), the interface required by MES will remain and will support the ISA95 activity model as well as the information flow defined by them. Services are exposed by a WPA and used by agents placed in the middleware part of the system. In reference to the ISA95 model, WPAs are responsible for production execution as defined for activity "5. Execution management" and for the resource-related operation defined in activity "2. Resource Management". The details of the service implementation depend on the internal sets of holons aggregated by WPA.

The middleware part of the system is composed of the network of agents that are responsible for the services focussed on the realisation of a given production order. They also have to execute the activities and requested message flows defined by ISA95. The information flows defined by ISA 95 arise from the interactions between the order supervisor and production system participants that are responsible for the execution of the order. In the proposed

system, every order is represented in the middleware part by one Order Agent (OA). The Order Agent reflects all of the requirements related to the production process agreed by the customer during the activity "1. Definition Management". It is also responsible for collecting information about the operation results supported by WPA described in ISA95 model as activity "6. Data collection". Before actual production can begin the necessary resources have to be available, confirmed and reserved. In classical manufacturing, the operations related to the reservation of resources, job planning and its distribution are done by human actors - a logistics team and a production team that prepare the production scheme as well as by a human product supervisor. In the proposed system, the activity of the human product supervisor is supported by a Supervisor Agent (SA). Each Order Agent receives support from an individual Supervisor Agent. This OA-SA pair of agents cooperates from the time of the order confirmation to its finalisation. SA supervises the execution of required production steps and plans them as required in the activities defined for "3. Detailed scheduling". Afterwards SA sends OA to the equipment and human operators for the given tasks as defined for "4. Dispatching" activity. The OA is interested in the proper and timely execution of a given order, including the detailed parameters required by production contract. The OA also has to collect the actual results of production. The SA represents the point of view of the product supervisor. It supports the cost-effective and time-efficient execution of the production required by the OA by interacting with the WPAs. The SA is responsible for preparing the production schedule for a given OA and resource reservation by communicating with the relevant WPAs.

The Order Agents collect data about the realisation of order (Activity 6) but such raw information has to be analysed in a broader context, including production performance that is focussed by the PA (Performance Agents) under the activity "8. Performance Analysis" or the quality issues detected and analysed by QA (Quality Agents). Although quality management is a part of a separate ISA95 diagram, QA work results are used in activities related with "7. Tracking", which is carried out by an OA. Unlike in the case of an OA - SA pair, the QA and PA agents are not tightly bound with a given production order or a given WPA. They reflect selected aspects of the production performance expressed by different KPIs and different aspects of quality management activities. Therefore, QA and PA operate in a global way and focus on the realisation of many production orders. They collect and process information from different SA and OA. These services are also used for the global analyses that are required by the ERP system, but some of the results are used directly in the Operation response information prepared by OA.

The most external layer of the proposed system provides interfaces to both other computer systems and human users as well. It allows high-level MES services to be split into agents operating in middleware and to collect feedback information and send it to external systems using the interfaces. A computercomputer interface follows B2MML (Business to Manufacturing Management Language). This external layer is also composed of agents, but as in the case of a physical production system, they are bound with a given system user or computer system interface.

One of required system functionalities is reflected by Product Supervisor Agents (PSA) that support a real production supervisor. Another functionality is the Customer Agents (CA) that create an interface between the production system and customers. Each customer communicates with a dedicated Customer Agent that handles all of the orders commissioned to the MES by a given customer. The main task for the Customer Agent is to mediate between a human customer and the cyber physical production system. Each CA supports all of the production tasks ordered by a given customer and who is personalised by user identifier. From

the other side, a CA ensures callback communication via e-mail or an SMS sent to the customer. Since one customer may have many orders in the production system and one supervisor may handle many orders under production, the relations between the PSA-SA and CA-OA are one-to-many. PSA supervise all of the operations related to SA activity and sends information to the responsible personnel if necessary. From another side, the PSA allows a human user to manage all of the orders under his/her supervision by communication with relevant OA. Each human product supervisor has a corresponding PSA that supports the execution of orders supervised by a given product supervisor. Avery similar solution is in the case of a CA. One CA is dedicated to a customer and represents his/her interests. The CA allows for the creation of new orders but also monitors the realisation of the order progress.

A CA selects a PA taking into account the order details and from that moment, the PSA-CA pair of agents cooperates on the external level. This goal of this cooperation is the preparation of an order by verifying the fusibility of the required technology or components and the raw order scheduling based on the average load of the production system. All of the decisions are made by human actors, but the PSA and CA support them in finding the best trade-off that reflects the point of view of both the customer and the producer. A detailed description of the planned production process is a part of an order and consists of the production path between Workplaces including any possible variants and parameters for the operations that are performed on a given production process. A ProductSu-pervisorAgent (PSA) verifies the feasibility of a required process and in the event that it is feasible presents the expected costs. A CustomerAgent can accept the proposed conditions for the execution of an order or can search for another ProductSupervisor-Agent. A real customer receives information about the ability or inability of the MES to realise a given order as well as the expected costs. After the negotiation process is finished, the customer confirms or cancels the order by Customer Agent and the moment of acceptance starts the physical production process, which is supported by the middleware agents, including OA and SA pair, and agents supporting quality (QA), performance (PA) and others.

The third level of agents allows information from different orders to be aggregated. In the proposed system, external service realisation is split between middleware agents that realise the process in parallel. Moreover, some services span a long time horizon and therefore, they are supported by different generations of middleware agents. The selected examples of aggregated knowledge are Quality Management issues, System Performance Management or Production Costs Management. They are often used as an external services that are required by an ERP system.

Referring to other manufacturing systems pointed out by T.J. Williams in the Purdue enterprise reference architecture [54] that was later adopted as the reference for the Activity Models of Manufacturing Operations Management defined by ISA95-3 [6], it is possible to extend the proposed system with other agent-based interfaces that can be located in the outer layer. The actual operations list should support the requirements that are given for the requested business operations and defined by systems that are located outside of the MES such as Order Processing, the Production Cost Account, Product Shipping Administration, Product Inventory Control, Marketing & Sales, Research Development and Engineering, Procurement, Material and Energy Control.

4. Selected case studies and practical results

The proposed MES architecture was practically verified on the real example of the production system used by the Prototyping Department of Continental Ingolstadt. This department realises two types of automotive electronic production. One is the production of the prototypes of new electronic devices for the

automotive market. This kind of production obviously has a low volume. It also needs a great deal of flexibility and requires many changes related to product development. The second kind of production is the short-series production of the electronic boards used in Advanced Driver Assistance Systems. In this case, car electronics are often produced in a short series with many variants. The Prototyping Department supports both kinds of production. Production is split between the production of SMD (Surface Mounted Devices), which is focussed on the electronics components and the backend production that is related to the assembly of the housing and connectors.

Although the Prototyping Department is located in Ingolstadt, it executes orders sent by different departments of Conti, which are located all over the world. This means that the considered MES architecture has to exchange information with different IT systems and also has to follow different business models. On the one hand, the analysed production system requires a significant commitment of staff that have to track customer requirements and solve the problems that occur during production. On the other hand, the system has to exchange information with the different IT systems and databases that exist in the company. In order to ensure the high reliability of these processes, the authors organised the MES architecture as a cyber-physical system that includes humans as well as the infrastructure and platforms that transform the interactions with the physical world. This has also been taken into account when designing the IT infrastructure. The presented solution works as s web application that is integrated with and based on Continental's domain policy. All of the services that involve humans are available via a web browser and are prepared in different language versions. It is not necessary for a user to install any additional software. Moreover, all of the required business logic is separated in the database, which makes the system more flexible and easy to maintain. In this paper, the authors only focus on the activities that are implemented in the detailed production scheduling that is supported by MES.

In the considered use case, the MES operates as an interface between product designers, order supervisors, production managers, production staff, production equipment used and the electronic devices produced. In the proposed architecture all of the above actors are supported by agents that provide the functionalities required on the MES level. The relations between agents follow the real dependencies that occur during production. Since the Prototyping Department executes many orders in parallel, the MES has to be able to solve any conflicts of interest between actors that reflect the real decisions that have to be taken during the production process. To make the right decision in both the cyber and physical parts of the system, all agents must receive timely and appropriate information.

Production operations management is an interactive process since the product manager has to contact the customer and agree on both the technical issues and logistical aspects of production. Afterwards, orders are qualified for execution that is composed of several steps described by the product specifications. The logistics team is responsible for preparing all of the necessary materials using the BOM (Bill of Materials) together with the technical documentation for the production process, then the actual production is executed by the SMD line and finalised by the backend line. All of the above operations in the case of Prototyping Department have to be flexible. Sometimes the BOM or SMD operations can be changed to reflect production needs, and sometimes they are changed to produce in a more effective way (for example, some materials can be replaced by others that are available). The tests of the product also have to be flexible. The test path depends on the product specification from one side and on the selected production technology from the other side.

In order to illustrate the different variants of production, the implemented Work Place Agents (WPA) that were used in the test part of the Prototyping Department and possible production flows are presented in Fig. 5. In the first step, all of the PCBs that comprise a given order are tested at the Automatic Optical Inspection (AOI) station. PCBs that have errors are sent for additional analysis, which is done by human operators at the Repair (REP) station. Whenever possible, errors are fixed at the REP and the appropriate information is stored in the production logs. Then, the electrical connections between the parts are tested at the Flying Probe (FP) station. In the case of errors, the PCB returns to the REP station and then the FP tests are repeated. The maximum number of repetitions is a parameter of a given order. In a case in which it is equal zero (no rework allowed); the testing process is similar as that in the case of mass production. When it is more than zero, the production paths can be different for particular PCBs in a given order. An order can be released after of the all constituent PCBs are tested. This is typical when many parallel orders are being processed simultaneously by a test part.

Because of the highly individualised production variants, which have to reflect the needs of short-series production, the classical MES model that is based on the production order push paradigm cannot be used. In the case of mass production, MES can be well adjusted to the corresponding business models. Any deviation from the model during the production is simply regarded as an error and questionable items are rejected. In the case of short-series production, the process is less reproducible and often different variants of production have to be accepted. In addition, decisions are often taken by the operators during the production, and therefore a hard defined pre-production model cannot be used. In order to reflect the need for cooperation, the authors decided to replace the classical MES software model with the agent-based architecture that was proposed in Section 3. Such a solution ensures the required flexibility of the MES for short-series production. In our use case, we focussed on the structure of the production system used in the Prototype Department, which is relatively simple and consists of a logistics workplace, SMD production line, test and rework stations.

The clients (real customers) submit production requests (short series in our case) into the system. This process is supported by MES services that are available to human users by web browsers that allow for the creation of new production requests, the modification of production orders and the monitoring of production progress. Real customers are supported by Customer Agents (CA). Each production order is monitored by a human product supervisor who is reflected in the system by the Product Supervisory Agent (PSA). The CA finds the most relevant Product Supervisor Agent (PSA). In the case of the presented experiment, there is only one PSA so the PSA selection part is omitted in our

Fig. 5. Work Place Agents implemented in the test part of the Prototyping Department of Continental Ingolstadt.

presentation. Next, the PSA creates a Supervisory Agent (SA) that is responsible for the realisation of a given order. After the decision about the start of production, each order in production is attended to by the Order Agents (OA) that represents all of the client's requirements related to the given production order. One OA is created for each order (product) under production. The linked pair of SA-SA starts the actual execution of a given order. It is responsible for tracking all activities related to the physical production of the given order (product).

The link to the production devices is established by means of Work Place Agents based on holons, which makes the system more flexible and which is also extremely important in the case of short-series production. Holons operate directly on the level of the production devices and are the interfaces to the proposed multiagent MES. In our case, many of WPAs contain LR holons (Labour Resource) that represent human operators. Since the system was implemented as a Microsoft MVC (Model-View-Controller) application and was created under the ASP.NET technology, the user interface benefits from HTML5 in terms of its mobility and easily accessible user interface. An example of such an interface is the Automatic Optical Inspection Work Place WPA.AOI. Although optical tests are performed and the results are evaluated by an automatic vision recognition system, the final error assessment is made by a human operator and any error is confirmed by a web browser interface as shown in Fig. 6

The simplified data model of our system is presented in Fig. 7. The base instance of the agent system is a Holon that contains a SignalR client object that connects to the central SignalR HUB service. Each holon can provide simulation data based on the user parameters. The two general types of holons are defined as Agents - responsible for processing an order request and the interaction with a client; and WorkPlaces - perform specific production process operations with a close connection to a machine and an operator.

Fig. 8 illustrates an example of the communication method between agents. In the first phase, the agents log into the system. In the next phases, the messages that are involved with workplace queues, orders requests, assigning orders, executing operations, the status of the execution orders and operations and releasing orders to production can be seen.

Holons collect information from the real production system and are a Cyber Physical part of the application. Agents process the information using the Internet services that are available for human users and for the other agents as well. From the other side,

Continental^

The Future in Motion

AP-08 (Automatische Optische Inspektion)

Entered Order: SON22000167

Enter Order ID:_

AOI Data

Description

Checked: StartTime: EndTime:

AOI Errors I Historical View |

Save & Exit

08.06.201511:18:02 08.06.201514:16:55

N0. PBC Number Position Defect Code Defect Side Defect Count

I ( I 01: Lotpaste wenig ») |T: Top » I

2 MIN438204 R19 16: Grabstein / Tombeston T 1

3 MIN438286 R51 02: Bauelement fehlt T 1

4 MIN438286 R43 02: Bauelement fehlt T 1

5 MIN438286 C79 02: Bauelement fehlt T 1

6 MIN438285 R51 02: Bauelement fehlt

Fig. 7. Test case data model (simplified view).

Fig. 6. Operator interface on the holon's side on the example of an Automatic Optical Inspection.

Fig. 8. An example interaction between agents.

WPAs provide an interface from the MES services to the actors responsible for the execution of a given production operation at a given workplace. This interface can be oriented to machine-machine communication, but also to human-machine-human communication.

The example interaction between agents during the subsequent process steps (PS) and the MESSAGE exchanges are illustrated in Fig. 8. During the first steps, which are shown in Fig. 8, (PS1-PS9) agents register themselves with the communication HUB. This process supports the flexibility of a system that doesn't use any predefined configuration. The only fixed point that is known to the agents is the address of the HUB. In the step PS10, the client agent

(CA) sent an order for production that will be supported by an order agent (OA). The CA doesn't have any knowledge about the actual production facilities and is only focussed on the requirements that are defined by the order. Knowledge about the possible paths of production is shared between the supervisor agent (SA) and order agent (OA) that are responsible for preparing the production scenario (PS11-PS24). The following steps (PS25-PS57) reflect the simulation of the production. Finally, the simulation results are sent back to the client agent (CA) in step PS58.

The cooperating entities (represented as columns in Fig. 8) are:

• HUB is the communication centre used as a router for message exchange between agents. It allows the necessary parameters to be sent but unlike RPC (Remote Procedure Call), it does not support a message response. The HUB supports the following communication modes: point-point, multi-point and broadcast. It was created based on the SignalR communication library;

• CA represents a Client Agent. In our example, we simulate only one Client Agent that generates a list of orders. In real cases there are many Client Agents linked with real clients by the authentication mechanism. The authentication system is based on a client's username and password. Each CA is linked with a given business model. The business models are defined based on the CA's employment place;

• OA is an Order Agent. Because we show only one order realisation path in Fig. 7, we omitted the order ID. In a real system, the OA is identified by a unique identifier that is composed of the order type and order number. Each OA has an individual production path that depends on the order type and individual production parameters including the production volume. In the presented example, we simulate these parameters based on real orders;

• SA is a Supervisor Agent. In our example, we simulate only one Supervisor Agent that supports all of the orders represented by the OAs. In real cases, there are many Supervisor Agents and CAs select one of them. The SA load is one of the selection parameters;

• WPA represents Work Place Agents. In the case of our simulation, we use six WPA examples that reflect the real workplaces that exist in the Prototyping Department; REP represents the Reparation Work Place, X-ray represents the X-ray tests, FP represents the Flying Probe Tests, AOI represents the Automatic Optical Inspection tests, SMD represents the SMD production line and LOG represents the Logistics section.

The presented research experiment is focussed on the simulation support for short-series production planning. The actual production is still maintained manually, but the Product Supervisor receives support in finding the best scenario of production scheduling and dispatching. For the purpose of the presented simulation example, the production process routes and product parameters were randomly generated for the Order Agents. In the case of real production, the actual production scenarios are used to find the optimal production paths and schedules for the execution of orders in the production system. The simulation results can be used to support the decision-making process. Thanks to this approach, the best solution that can be found for production can be presented to the Product Supervisor. Continuous feedback to the simulation Agent is provided by the production operation monitoring while all of the executed operations are reported by the Holons. All bottleneck problems are instantaneously visible on the holons level and can be solved by production rescheduling without having to wait for the final realisation of the order.

The simulation results are presented in Figs. 9-11. We have defined the simulation goal as the verification of the production

Fig. 9. Number of orders in the system and WPA queue lengths during the first experiment.

Number of orders 100

/ • Orders in the system

_/_ • Orders generation

j • SMD

_/ • AOI

FP XRay

/y-1 i Sv

t ' 'VH^—uC^ *— -» ■»», , it , TffTiT

18:53:34 19:00:46 19:07:58 19:15:10 19:22:22 19:29:34 19:36:46

Fig. 10. Number of orders in the system and WPA queue lengths during the second experiment.

system behaviour during the peak load of production orders. We performed two sets of experiments, each one with the goal of one hundred orders for production. The orders were generated in randomly with a normal distribution, but with different parameters of the average time interval between orders being generated by Customer Agent and the average processing time by the Work Place Agents. The simulation parameters are presented in Table 1. In both cases, the orders have a random number of produced parts between 1 and 10 and were generated as three different types with different production paths (with equal probability).

VED, SPO, TRP represent examples of different models of production. Models define the data that is collected during order preparation and realisation but also indicate the required

Order number

Fig. 11. Comparison of the service time for both simulation scenarios.

Table 1

Simulation parameters.

Customer Agent Simulation 1 Simulation 2

Average time interval 10 5

Standard deviation 5 1

Number of orders 100 100

Simulation 1

LOG SMD AOI FP XRay

Average processing time 3 15 6 5 4

Standard deviation 1 5 2 1 2

Simulation 2

LOG SMD AOI FP Xray

Average processing time 2 5 4 3 4

Standard deviation 1 2 1 1 2

Order type Routing

VED LOG SMD AOI FP Xray

SPO LOG SMD AOI FP

TRP LOG SMD FP

(Automatic Optical Inspection), FP (Flying Probe tests) and X-ray (X-Ray tests). Because different types of orders were generated during the simulation, particular test places received a different number of orders for processing. Moreover, the production volume for a given order was also randomly generated, but with a normal distribution, which reflects a given type of order. Fig. 9 presents two bottlenecks in the production process.

In the first phase, we can see that the time of order preparation by the logistics department is more important while in the second phase of the experiment, it is the time of production on the SMD line. Test stations showed a low load (queue length) in our experiment. In a real system, one test station is used for more than one production line.

During the second experiment, we lengthened the average value of the time interval between orders and shortened average service time for WPA slightly (especially for the SMD line). The results of the second simulation are presented in Fig. 10. We can see that in this case, Logistics was able to process the incoming orders, but we still had a bottleneck on the SMD line.

The service time for a given order is also an important parameter of production (especially for customers). In Fig. 11, we can compare the service time in the case of our first and second experiment. Although, the time increases with the number of orders in both cases service, during the second experiment the behaviour of the production system is more stable and more predictable.

The two use cases presented above show how the proposed approach allows the process of production scheduling to be improved. Individual orders have different execution paths and different service times for various steps of production. The proposed method allows the expected load on the system to be estimated with accuracy for a particular Work Place load. The expected execution times of individual orders are calculated by the simulation. The results of the first experiment (Fig. 9) show not only the expected time for the realisation of each order but also allow a lengthening queue of orders at the SMD production line (SMD Work Place) to be detected. A long queue may require additional storage space to be prepared for waiting orders and such a problem is clearly visible in advance thanks to the earlier simulation. The results of the second experiment (Fig. 10) show that in the event of an increased throughput of SMD Work Place, the Logistics Work Place will not be able to keep up with the preparation of orders for production. In such a case, for example, the production manager can verify how the introduction of work on the second shift in the logistics will balance the load of the production system via the next simulation run.

5. Conclusions

production procedures. In Table 1 we can see different routing paths for the tests performed on VED, SPO, TRP. Order types are used in a simulation to reflect the real production processes that are related to a given type of order.

In Fig. 9, we can see the time diagram that presents the number of orders generated (Orders generation) from the start point of the first simulation (before the start the system was empty). After the peak load of CA activity, the generation of orders was stopped. The time scale presents the simulation time (the real average processing time at given workplaces were hidden at the request of the Prototyping Department).

The orders in a system trend indicate how production copes with orders. The number of orders queuing at a given Work Place is presented as LOG (Logistics), SMD (SMD production line), AOI

The authors propose an innovative IT architecture for an MES dedicated for short-series production that is based on a multiagent environment implemented in the Microsoft Visual Studio MVC (Model-View-Controller) environment. The authors used the ISA95 activity model to make the MES applicable in a real production environment. The proposed IT architecture is based on the push information flow model, which allows for a shorter reaction time and a reduction in unproductive periods. Agents support distributed information processing and the appropriate human reactions that are necessary in the event of production problems. Although the selection of the technology was forced by the requirements specified by the industry, it allowed the authors to verify the new features and communication technologies, which are not very typical in agent-based systems. The tests performed show a simulation of the behaviour of the production system during the peak load of production orders.

The proposed approach is based on a distributed agent-based model that allows for the decentralisation of production planning that is particularly advantageous in the case of short-series production. In the case of mass production, the optimisation of the production schedule is mostly a part of the activity of the Enterprise Resource Planning Systems or is carried out by dedicated Advanced Planning Systems. In the case of short-series production, such an approach would not be effective due to the many specific, production-side and technology dependent parameters that are inaccessible at the level of the business systems. Production can be optimised locally during its execution. However, there is a lack of supporting tools that can help to predict the impact of the decisions taken by the operating management. The proposed system is an attempt to fill this gap. In section 4, the authors show that even small changes in the way production is implemented can result in a big change to the overall performance of the entire production system. The proposed solution transfers part of the production scheduling task from the business planning level to the business execution tasks that are carried out by the Manufacturing Execution Systems. Since there are many players and many goals at this level, the globally optimised, but not accurate central scheduling, is replaced by local optimisation that is focussed on individual orders that is provided by Order Agents. As is shown by the use cases, the output of the simulation is input into the local decisions, which leads to the further optimisation of short-series production.

Acknowledgements

This work was supported by the European Research Area - FP7-PEOPLE-2013-IAPP AutoUniMo project "Automotive Production Engineering Unified Perspective based on Data Mining Methods and Virtual Factory Model" (grant agreement no: 612207).

References

[1] S.S. Shipp, N. Gupta, B. Lal, J.A. Scott, C.L. Weber, M.S. Finnin, M. Blake, S. Newsome, S. Thomas, Emerging global trends in advanced manufacturing, DTIC Document, (2012).

[2] European Commission, A Manufacturing Industry Vision 2025, European Commission (Joint Research Centre) Foresight study, 2013.

[3] R.C. McNally, M.B. Akdeniz, R.J. Calantone, New product development processes and new product profitability: exploring the mediating role of speed to market and product quality: the mediating role of speed to market and product quality, J. Prod. Innov. Manag. 28 (2011) 63-77, doi:http://dx.doi. org/10.1111/j.1540-5885.2011.00861.x.

[4] R. Poovendran, Cyber-Physical systems: close encounters between two parallel worlds [Point of view], Proc. IEEE 98 (2010) 1363-1366, doi:http:// dx.doi.org/10.1109/JPROC.2010.2050377.

[5] R. Baheti, H. Gill, Cyber-physical systems, Impact Control Technol. 12 (2011) 161-166.

[6] The International Society of Automation, ANSI/ISA-95. 00. 03-2013 (IEC 62264-3 Modified) Enterprise-Control System Integration-Part 3: Activity Models of Manufacturing Operations Management Approved 8 July 2013 (2013).

[7] A. Rajhans, S.-W. Cheng, B. Schmerl, D. Garlan, B.H. Krogh, C. Agbi, A. Bhave, An architectural approach to the design and analysis of cyber-physical systems, Electron. Commun. EASST 21 (2009), doi:http://dx.doi.org/10.14279/tuj. eceasst.21.286.

[8] I.C. Wright, A review of research into engineering change management: implications for product design, Des. Stud. 18 (1997) 33-42, doi:http://dx.doi. org/10.1016/S0142-694X(96)00029-4.

[9] T.A.W. Jarratt, C.M. Eckert, N.H.M. Caldwell, P.J. Clarkson, Engineering change: an overview and perspective on the literature, Res. Eng. Des. 22 (2011) 103124, doi:http://dx.doi.org/10.1007/s00163-010-0097-y.

[10] N. Do, Integration of engineering change objects in product data management databases to support engineering change analysis, Comput. Ind. 73 (2015) 6981, doi:http://dx.doi.org/10.1016/j.compind.2015.08.002.

[11] J. Tav£ar,J. Duhovnik, Engineering change management in individual and mass production, Robot. Comput.-Integr. Manuf. 21 (2005) 205-215, doi:http://dx. doi.org/10.1016/j.rcim.2004.07.017.

[12] W. Liu, R. Ning, J. Liu, K. Jiang, Mechanism analysis of deviation sourcing and propagation for mechanical assembly, Jixie Gongcheng XuebaoChinese, J. Mech. Eng. 48 (2012) 156-168.

[13] D. Liu, P.Jiang, Fluctuation analysis of process flow based on error propagation network, Chin. J. Mech. Eng. 46 (2010) 14-21.

[14] X. Wang, M. Liu, M. Ge, L. Ling, C. Liu, Research on assembly quality adaptive control system for complex mechanical products assembly process under uncertainty, Comput. Ind. 74 (2015) 43-57, doi:http://dx.doi.org/10.1016/j. compind.2015.09.001.

[15] M. Rolon, E. Martinez, Agent-based modeling and simulation of an autonomic manufacturing execution system, Comput. Ind. 63 (2012) 53-78, doi:http://dx. doi.org/10.1016/j.compind.2011.10.005.

[16] M. Rolon, E. Martinez, Agent learning in autonomic manufacturing execution systems for enterprise networking, Comput. Ind. Eng. 63 (2012) 901-925, doi: http://dx.doi.org/10.1016Zj.cie.2012.06.004.

[17] H. Stadtler, Supply chain management and advanced planning—basics, overview and challenges, Eur. J. Oper. Res. 163 (2005) 575-588, doi:http:// dx.doi.org/10.1016/j.ejor.2004.03.00 .

[18] M.C. Vidoni, A.R. Vecchietti, An intelligent agent for ERP's data structure analysis based on ANSI/ISA-95 standard, Comput. Ind. 73 (2015) 39-50, doi: http://dx.doi.org/10.1016/j.compind.2015.07.01 .

[19] K. Bokovec, T. Damij, T. RajkoviC, Evaluating ERP Projects with multi-attribute decision support systems, Comput. Ind. 73 (2015) 93-104, doi:http://dx.doi. org/10.1016/j.compind.2015.07.004.

[20] A.J. Zoryk-Schalla, J.C. Fransoo, T.G. de Kok, Modeling the planning process in advanced planning systems, Inf. Manage. 42 (2004) 75-87, doi:http://dx.doi. org/10.1016/j.im.2003.06.005.

[21] M. Bonev, L. Hvam, J. Clarkson, A. Maier, Formal computer-aided product family architecture design for mass customization, Comput. Ind. 74 (2015) 5870, doi:http://dx.doi.org/10.1016/j.compind.2015.07.006.

[22] M.A. Rothenberger, M. Srite, An investigation of customization in ERP system implementations, IEEE Trans. Eng. Manag. 56 (2009) 663-676, doi:http://dx. doi.org/10.1109/TEM.2009.2028319.

[23] M.F. Zaeh, M. Beetz, K. Shea, G. Reinhart, K. Bender, C. Lau, M. Ostgathe, W. Vogl, M. Wiesbeck, M. Engelhard, C. Ertelt, T. Rühr, M. Friedrich, S. Herle, The cognitive factory, in: H.A. ElMaraghy (Ed.), Chang. Reconfigurable Manuf. Syst., Springer, London, London, 2009, pp. 355-371 http://link.springer.com/ 10.1007/978-1-84882-067-8_20 (accessed 10.12.15).

[24] M. Witsch, B. Vogel-Heuser, Towards a formal specification framework for manufacturing execution systems, IEEE Trans. Ind. Inform. 8 (2012) 311-320, doi:http://dx.doi.org/10.1109/TII.2012.2186585.

[25] P. Blanc, I. Demongodin, P. Castagna, A holonic approach for manufacturing execution system design: an industrial application, Eng. Appl. Artif. Intell. 21 (2008) 315-330, doi:http://dx.doi.org/10.1016/j.engappai.2008.01.00 .

[26] T.J. Williams, P. Bernus, J. Brosvic, D. Chen, G. Doumeingts, L. Nemes, J.L. Nevins, B. Vallespir, J. Vlietstra, D. Zoetekouw, Architectures for integrating manufacturing activities and enterprises, Comput. Ind. 24(1994) 111 -139, doi: http://dx.doi.org/10.1016/0166-3615(94)90016-7.

[27] M. McClellan, Introduction to manufacturing execution systems (2001) 1-7.

[28] T. Sauter, The continuing evolution of integration in manufacturing automation, IEEE Ind. Electron. Mag. 1 (2007) 10-19, doi:http://dx.doi.org/ 10.1109/MIE.2007.357183.

[29] Manufacturing Enterprise Solutions Association, MESA, (n.d.). http://www. mesa.org/.

[30] International Society of Automation, ANSI/ISA-95, (n.d.). http://isa-95.com/.

[31] M. Rolon, M. Canavesio, E. Martinez, Agent based modelling and simulation of intelligent distributed scheduling systems, Comput. Aided Chem. Eng., Elsevier, 2009, pp. 985-990 http://linkinghub.elsevier.com/retrieve/pii/ S1570794609701646 (accessed 10.12.15).

[32] W. Shaojun, W. Gang, L. Min, G. Guoan, Enterprise resource planning implementation decision & optimization models, J. Syst. Eng. Electron. 19 (2008) 513-521, doi:http://dx.doi.org/10.1016/S1004-4132(08)60115-2.

[33] D. Trentesaux, Distributed control of production systems, Eng. Appl. Artif. Intell. 22 (2009) 971-978, doi:http://dx.doi.org/10.1016/j. engappai.2009.05.001.

[34] A. Maka, R. Cupek, M. Wierzchanowski, Agent-based modeling for warehouse logistics systems, IEEE (2011) 151-155, doi:http://dx.doi.org/10.1109/ UKSIM.2011.37.

[35] H. Van Brussel, Holonic manufacturing systems, the international academy for production engineering, in: L. Laperriere, G. Reinhart (Eds.), CIRP Encycl. Prod. Eng., Springer, Berlin, Heidelberg, 2014, pp. 654-659 http://dx.doi.org/ 10.1007/978-3-642-20617-7_6556 (accessed 10.12.15).

[36] L. Rannanjärvi, T. Heikkilä, Software development for holonic manufacturing systems, Comput. Ind. 37 (1998) 233-253, doi:http://dx.doi.org/10.1016/ S0166-3615(98)00101-8.

[37] T. Borangiu, P. Gilbert, N.-A. Ivanescu, A. Rosu, An implementing framework for holonic manufacturing control with multiple robot-vision stations, Eng. Appl. Artif. Intell. 22 (2009) 505-521, doi:http://dx.doi.org/10.1016/j. engappai.2009.03.001.

[38] D.M. Dilts, N.P. Boyd, H.H. Whorms, The evolution of control architectures for automated manufacturing systems, J. Manuf. Syst. 10 (1991) 79-93, doi:http:// dx.doi.org/10.1016/0278-6125(91)90049-8.

[39] N. Aissani, B. Beldjilali, D. Trentesaux, Dynamic scheduling of maintenance tasks in the petroleum industry: a reinforcement approach, Eng. Appl. Artif. Intell. 22 (2009) 1089-1103, doi:http://dx.doi.org/10.1016/j. engappai.2009.01.014.

[40] P. Leitäo, Agent-based distributed manufacturing control: a state-of-the-art survey, Eng. Appl. Artif. Intell. 22 (2009) 979-991, doi:http://dx.doi.org/ 10.1016/j.engappai.2008.09.005.

[41] J.-L. Chirn, D. McFarlane, Application of the holonic component-based approach to the control of a robot assembly cell, in: 2000.

[42] K. Nagorny, A.W. Colombo, U. Schmidtmann, A service- and multi-agent-oriented manufacturing automation architecture, Comput. Ind. 63 (2012)813-823, doi:http://dx.doi.org/10.1016/j.compind.2012.08.003.

[43] Hadeli, P. Valckenaers, M. Kollingbaum, H. Van Brussel, Multi-agent coordination and control using stigmergy, Comput. Ind. 53 (2004) 75-96, doi:http://dx.doi.org/10.1016/S0166-3615(03)00123-4.

[44] P. Valckenaers, H. Van Brussel, Holonic manufacturing execution systems, CIRP Ann. - Manuf. Technol. 54 (2005) 427-432, doi:http://dx.doi.org/10.1016/ S0007-8506(07)60137- .

[45] P. Leitao, A.W. Colombo, F.J. Restivo, ADACOR: a collaborative production automation and control architecture, IEEE Intell. Syst. 20 (2005) 58-66, doi: http://dx.doi.org/10.1109/MlS.2005.2.

[46] J.A.B. de Oliveira, Coalition based approach for shop floor agility -a multiagent approach (2003) https://run.unl.pt/handle/10362/2483 (accessed 10.12.15).

[47] J.H. Christensen, Holonic manufacturing systems: initial architecture and standards directions, Proc 1st Euro Wkshp Holonic Manuf. Syst. (1994).

[48] R. Cupek, H. Erdogan, L. Huczala, U. Wozar, A. Ziebinski, Agent based quality management in lean manufacturing, in: M. Núñez, N.T. Nguyen, D. Camacho, B.

Trawiñski (Eds.), Comput. Collect. Intell., Springer International Publishing, Cham, 2015, pp. 89-100 http://dx.doi.org/10.1007/978-3-642-20617-7_6556 (accessed 10.12.15).

[49] R. Cupek, A. Maka, OPC UA for vertical communication in logistic informatics systems, IEEE (2010) 1--4, doi:http://dx.doi.org/10.1109/ETFA.2010.5640978.

[50] The International Society of Automation, ANS1/1SA-95.00.01-2010 (1EC 622641 Mod) Enterprise-Control System Integration - Part 1: Models and Terminology (2010).

[51] The International Society of Automation, ANS1/1SA-95.00.02-2010 (1EC 622642 Mod) Enterprise-Control System Integration - Part 2: Object Model Attributes (2010).

[52] The International Society of Automation, ANS1/1SA-95.00.04-2012 Enterprise-Control System Integration - Part 4: Objects and attributes for manufacturing operations management integration (2012).

[53] The International Society of Automation, ANS1/1SA-95.00.05-2013 Enterprise-Control System Integration - Part 5: Business-to-Manufacturing Transactions (2013).

[54] Theodore J. Williams, The Purdue enterprise reference architecture, Comput. Ind. 24 (2) (1994) 141-158.