Study of manufacturing resource perception and process control of a radio-frequency-identification-enabled decentralized flexible manufacturing system

Abstract

The fusion of mechatronics, communication, control, and information technologies has introduced new automation paradigms into the production environment. Plug-and-play flexible manufacturing systems will become suitable approaches in the future for the development of modular, flexible, and reconfigurable manufacturing systems, addressing the requirements of global markets. This article introduces a Petri Net model-driven methodology for the development, validation, and operation of a radio-frequency identification-enabled decentralized flexible manufacturing system. After analyzing the manufacturing processes and resources of a flexible manufacturing system, the manufacturing resources are classified into active and passive resources. Each active resource is equipped with a radio-frequency identification reader and each passive resource is banded with a radio-frequency identification tag. Real-time state and behavior logic models are built for manufacturing resources based on extended Petri Nets; the models are translated into XML and integrated with the manufacturing resources. In this fashion, each manufacturing resource becomes an autonomous agent, and it can make self-decisions and update its status through the twinned models. In this manner, automatic perception and process control are realized. Finally, the effectiveness and feasibility of the method are verified in an experimental system.

Keywords

Radio-frequency identification decentralized flexible manufacturing system Petri Net resource perception process control

Introduction

Market competition is becoming increasingly fierce with the need to accommodate diversity and individual orientations. This requires production technologies to focus on small and medium size and to turn to part/product families of increasing variety.¹ Manufacturers have to improve product quality and production efficiency while reducing production costs,^2,3 but traditional centralized production systems are not agile or flexible enough to satisfy the personalized requirements of the modern market.⁴ Moreover, their construction always runs the risk of huge investments and long lead times. Once the system is installed and operational, changes are often complex and difficult. Worse yet, in centralized hierarchical organizations, the whole system may shut down with a single failure at a single point.¹

As a result, research is looking closely at plug-and-play decentralized flexible manufacturing systems (FMS). They are more adaptive for multi-type and small-quantity production and provide low-cost, high quality, and diversified products through their adaption, agility, and modularization. Multi-agent manufacturing systems (MAS) and holonic manufacturing systems (HMS) are examples of a distributed FMS. But as Marik and McFarlane⁵ have noted, most MAS and HMS research has focused on theory. The adaptivity and quick response of a distributed FMS depend on the efficiency of direct real-time information collection.⁶ Such kinds of FMS are not generally because of the difficulty acquiring real-time manufacturing process information reliably and automatically identifying manufacturing resources.

With the development of Internet of Things (IoT) technology, especially the growth of radio-frequency identification (RFID) technology and its applications in manufacturing systems, there are increasingly reliable methods to collect information and identify resources automatically in real time. The distributed FMS based on IoT technology has become an important research trend.⁷ In an RFID-enabled manufacturing system, a tag is attached to every product/part which contains its basic information (e.g. unique ID, name, state). The RFID reader is able to read and/or write the data in the tag when the tag is in the readable range of the reader.⁸ Thus, the product and process data can be collected, transferred, treated, and stored in systems. A great deal of research has considered the use of RFID technology for manufacturing systems. McFarlane et al.⁹ applied it to enhance intelligent control in an experimental assembly line. Qiu¹⁰ proposed an RFID-based framework for integrating an automated production control system and a management information system in a factory. Chen et al.¹¹ proposed an enterprise application integration framework based on RFID and agent technologies. Zhang et al.¹² built an ubiquitous manufacturing environment using an RFID-enabled gateway technology, in which RFID technology enables real-time traceability, visibility, and interoperability to improve the performance of shop floor planning, execution, and control. Huang et al.¹³ proposed a reactive, model-based approach to monitor important elements by estimating their most likely states according to RFID information and a constraint-based model. Liu et al.¹⁴ proposed a multi-agent real-time production management and control system for a Loncin motorcycle assembly line, integrated with an RFID-enabled real-time data capture system. The results showed productivity and quality improved. Zhong et al.¹⁵ presented an RFID-enabled real-time manufacturing execution system (RT-MES), in which RFID devices are deployed systematically on the shop floor to track and trace manufacturing objects and collect real-time production data. Zhang et al.¹⁶ presented a multi-agent-based real-time production scheduling method for an RFID-enabled ubiquitous shop floor environment, with RFID deployed at value-adding points to form a machine agent for the collection and processing of real-time shop floor data. Wang et al.¹⁷ proposed a hybrid-data-on-tag-enabled decentralized control system for flexible smart workpiece manufacturing shop floors. Zhong et al.¹⁸ suggested a visualization approach for RFID-enabled shop floor logistics big data from cloud manufacturing. Using an innovative RFID-cuboid model to reconstruct the RFID raw data given the production logic and time series. The literature surveyed by these authors indicates the need to employ RFID technology in a manufacturing control system. In most work, by introducing an RFID system, a physical product/part becomes a recognizable object, real-time shop floor data can be collected, production status can be analyzed, and production performance can be enhanced.

RFID, programmable logic controller (PLC), industrial personal computer (IPC), and so on can provide the hardware basis for the memory ability of manufacturing resources, but these manufacturing resources do not have an independent logical reasoning ability, largely because of the inability of the modeling approach to handle logic control of resources. The decentralized FMS is a typical high synchronous and complex system in which different resources affect each other in a specific manufacturing environment. To accomplish decentralized intelligent control of a decentralized FMS, resources must have state memory and logic reasoning abilities. More specifically, each manufacturing resource should be a cyber-physical intelligent entity. A cyber-physical resource is one that integrates its hardware function with a cyber-representation acting as a virtual representation for the physical part.¹⁹ Using modeling technologies to incorporate the behavior logic and real-time status of the manufacturing resources is an efficient approach to complete automatic identification and intelligent control.

Many methods have been used to model discrete-event dynamic systems (DEDS), that is, IDEF, activity network diagram, unified modeling language, but they cannot efficiently model and validate manufacturing systems.²⁰ Model methodologies for manufacturing systems should have a modeling formalism that capture characteristics inherent to manufacturing systems such as concurrency, deadlocks, conflicts, resource sharing. The methodologies should be able to validate the behavioral specifications and analyze the system performance.

The Petri Nets (PN) and its extensions, that is, colored Petri Net (CPN), timed Petri Net (TPN), are based on well-founded mathematical theory and have very good capacity to formally and graphically model, analyze, and validate concurrency, synchronization, resource sharing, and mutual exclusion, and even can be used to memorize, monitor, and supervise manufacturing processes. The PN can not only describe and analyze concurrence and conflict of resources, but also record the status of the system dynamically by tokens (marks in places), making it a common choice in structural modeling, dynamic analysis, and process control for manufacturing systems. The traditional major application of PN in manufacturing system is in modeling and analysis, with a great deal of research available.^21–23 The PN is also used in process control and monitoring.

In 1998, Feldmann and Colombo²⁴ developed a CPN-based control method of material flow. Then in 1999, Feldmann and Colombo²⁵ focused on the development and implementation of feature- and model-based monitoring methods using high-level PN specifications of flexible production systems and embedded discrete-event controllers. Leitão and Colombo²⁰ proposed a methodology for the development of collaborative production systems, using high-level PN. Mendes et al.²⁶ described a solution for the control of service-oriented devices based on modular and specially adapted high-level PN process descriptions of intra- and inter-control activities. Tavares et al.²⁷ developed a PN-based work-flow modeling and control method and applied this method to logistics and manufacturing process control by combining it with an RFID system. Chen²⁸ developed a CPN-based FMS cell controller. Mendes et al.²⁹ introduced an integrated approach to the design, analysis, validation, simulation, and process execution of service-oriented manufacturing systems, using high-level PN formalism to describe the system behavior. Huang et al.³⁰ studied the traceability issue in cyber-physical manufacturing systems from a theoretical viewpoint. They generalized PN models to formulate dynamic manufacturing processes; based on this formulation, they outlined a detailed approach to traceability analysis. Quintanilla et al.³¹ offered a PN-based methodology to increase planning flexibility in service-oriented HMS.

This article proposes an extended PN-based modeling approach to manufacturing resource perception and process control for RFID-based FMS. It shows manufacturing resource perception and process control can be realized using RFID systems. The rest of this article is organized as follows. Section “Process analysis and resource classification of manufacturing systems” explains the process analysis and resource classification of manufacturing systems. Section “Manufacturing resource modeling” elaborates on manufacturing resource modeling. Section “Resource perception and process control during manufacturing process” gives the key steps of resource perception and process control during manufacturing process. Section “Experimental system implementation” describes the experimental system implementation. Conclusions and suggestions for future research are presented in section “Conclusion and future research.”

Process analysis and resource classification of manufacturing systems

Process analysis of manufacturing systems

A manufacturing system is a transition system that converts raw material resources into the customers’ required products using equipment resources (machining equipment, transport equipment, storage equipment, and so on). In a manufacturing system, each manufacturing resource carries out the corresponding activities of machining, transport, storage, and so on in accordance with the specific processing technology and manufacturing environment.

Figure 1 gives the simple manufacturing system with seven buffers (storage equipment: B1, M1_BI, M1_BO, B2, M2_BI, M2_BO, and B3), two machines (machining equipment: M1 and M2), three transporters (AGV1, ROB1, and AGV2), and several workpieces (J1, J2, etc.). To more easily express the proposed approach, the remainder of the article is based on this scenario.

Figure 1.

A simple manufacturing process.

Classification of manufacturing resources

Generalized manufacturing resources refer to the software and hardware used in the overall product manufacturing cycle including equipment, workers, materials, technology, and so on.³² In this article, only four kinds of resources are considered: machining equipment, transport equipment, storage equipment, and workpieces (materials). They are further divided into active and passive resources. Active resources have their own status memory and can make intelligent decisions based on the active perception of the manufacturing environment. They include machining equipment and transport equipment. Passive resources cannot recognize the manufacturing environment actively, but they have their own status memory and can be recognized by active resources. They include workpieces and storage equipment.

To enable active perception and logical reasoning, each active resource has a terminal controller and an RFID reader. The resource state and behavior logic model based on extended PN are stored in the terminal controller; thus, active resources will have state memory and logic reasoning. Each passive resource is given an RFID tag; its state and behavior logic model of extended PN are stored in the tag as XML schema. When the resources begin to communicate, the active resources sense and identify the passive resources using the RFID readers. The active resources acquire the status of passive resources by analyzing the PN models; they decide the next steps by combining their own status and behavior logic with the status of the passive resources. In this way, the automatic perception and process control of manufacturing resources can be accomplished.

Manufacturing resource modeling

To accomplish the perception and logical control of manufacturing resources, these resources must memorize and realize their own status and behavior logic. This means the designer of the system should plant behavior logic and real-time status in them using modeling technologies. In this article, based on the requirements of all kinds of manufacturing resources, the models are built by extending the definition of basic PN and CPN.

Basic concepts of PN and CPN

PN was originally developed by CA Petri³³ in his PhD dissertation in 1962. Since then, the PN has been extended and developed; it is now widely used in modeling, simulating, scheduling, controlling, and monitoring DEDS with asynchrony, synchronicity, and competitive characters.

An ordinary PN is a collection of directed arcs connecting places and transitions. Places may hold tokens. The state or marking of a net is its assignment of tokens to places. In the manufacturing system field, places are often used to represent the status of resources; a token’s presence indicates whether a resource is available, a process is under way, or a condition is true. Transitions are used to model events. Directed arcs are used to represent the causal relations between place and transition.

The modeling of the simultaneous execution of several instances of the same component (e.g. a type of product) is easily modeled by an ordinary PN. However, the modeling of the simultaneous execution of several instances of different components (e.g. different types of products) is complex using an ordinary PN, since it requires the usage of as many PN models as the number of available components (e.g. different types). A CPN can overcome this. A CPN is a backward compatible extension of a PN. CPNs preserve useful properties of PNs and, at the same time, extend initial formalism to allow the distinction between tokens.³⁴ A CPN allows tokens to have data value attached to them. This attached data value is called the token color. Although the color can be arbitrarily complex type, places in CPNs usually contain tokens of one type. This type is called the colorset of the place. More detailed definition and examples can be found in the literature.^21,22

PN-based workpiece modeling

Workpieces are manufacturing system objects. They are materials at first and become products (or parts) gradually through a series of operations (including machining, inspection, handling, storage, and so on). The modeling of the workpiece must express all activities and their logical relationship, related information of external resources, the key parameters, and the results of operations in the manufacturing system. The model should also be able to reflect the real-time status of products. In our approach, each kind of workpiece has a PN model based on its manufacturing process. For example, Figure 2 shows the PN model of workpiece J1 according to the manufacturing process shown in Figure 1. Transitions denote the operations (including machining, transport, testing, and so on.). Every transition is assigned one or more pieces of the machining or transport equipment. This equipment is the set of equipment required to perform the operation. Places denote the storage equipment (e.g. buffers) during the manufacturing process. Each place corresponds to one or more pieces of storage equipment. This equipment is the set of equipment required to store the workpiece.

Figure 2.

Modeling of workpiece J1.

Tokens in place show the state of the workpiece. The directed arcs show the relationship between places and transitions; every directed arc has a function, that is, denoting whether the calculated value is true or false. The calculated value is the condition of whether the directed arc is connected or not; when the value is true, the directed arc is connected. If the directed arc has no function, the directed arc is connected by default. A detailed description of every element in Figure 2 appears in Table 1.

Table 1.

Elements of workpiece model in Figure 2.

Element	Description	Associated apparatus
P1	J1 is in B1 and waiting to be transported	B1
P2	J1 is in M1_BI and waiting to be machined by M1	M1_BI
P3	J1 is qualified upon inspection and in M1_BO after being finished in M1	M1_BO
P4	J1 is disqualified upon inspection and in M1_BO after being finished in M1	M1_BO
P5	J1 is in M2_BI and waiting to be machined by M2	M2_BI
P6	J1 is in B2	B2
P7	J1 is in M2_BO after being finished in M2	M2_BO
P8	J1 is in B3	B3
T1	J1 is being moved by AGV1 from B1 to M1_BI	AGV1
T2	J1 is being machined by M1	M1
T3	J1 is being moved by ROB1 from M1_BO to M2_BI	ROB1
T4	J1 is being moved by ROB1 from M1_BO to B2	ROB1
T5	J1 is being machined by M2	M2
T6	J1 is being moved by AGV2 from M2_BO to B3	AGV2
G1	Qualified upon inspection	−
G2	Disqualified upon inspection	−

As illustrated in Figure 3, to show the complete state of a workpiece, a workpiece model based on PN should give the total information of the workpiece, including machining equipment, transport equipment, storage equipment, starting and ending time of each activity, the key technical standard, the record of quality inspection, and so on.

Figure 3.

Information contained in a workpiece model.

CPN-based equipment resource model

Machining equipment, transport equipment, and storage equipment are the important resources during the production process. In a FMS, a single piece of equipment can be used for a variety of products. Thus, as the types of workpieces increase, the FMS becomes more and more complicated, and a PN-based model becomes difficult to build. A CPN can overcome the problem better than a PN.

Model of machining equipment

Machining equipment is the most important resource to complete the workpieces. This article assumes the equipment can get the workpiece from the input buffer and release it to the output buffer. Thus, machining processing includes three stages: (1) getting a workpiece from the input buffer and clamping, (2) machining, and (3) releasing the workpiece to the output buffer.

In Figure 4, the CPN-based machining equipment model includes three colorsets: M is the equipment colorset, denoting the type of equipment; J is the workpiece colorset, denoting the type of workpiece; and P is the process colorset, denoting the type of process. The model also defines three variables. m is the equipment variable, denoting the equipment used; j is the workpiece variable, denoting the workpiece machined; and p is the process variable, denoting the process used. The machining equipment will interact with workpieces and storage equipment because machining equipment is an active resource. To accomplish interaction, the place in the model is divided into state-place and port-place. State-place is shown by a solid circle and reflects the real-time status of the machining equipment. Port-place is shown by a dashed circle and reflects the status of the workpiece and storage equipment which interacts with the machining equipment. The information on status comes from the relevant workpiece and storage equipment. The main process of the model is as follows:

Step 1. The equipment perceives the arriving workpiece and reads its status information; the information is inserted into the port-place IMP1 by its RFID reader. A colored token (j, p) in IMP1 shows workpiece j is waiting for machining, and the machining process is p.

Step 2. The equipment’s state is checked. If there is a token in MP0, the equipment is ready and the system can move to the next operation; otherwise, the system keeps waiting.

Step 3. The equipment gets the machining process from the process database according to the colored token (j, p). If it exists, the system will get it. If not, the system will alarm and revert to a waiting state.

Step 4. The system begins to machine when the workpiece, the equipment, and the machining process are ready. Transition MT3 is triggered, and (1) a colored token (m, j, p) is updated into state-place MP3, indicating the process p of workpiece j is executed by equipment m. (2) The input buffer space of the equipment is released; specifically, a token (j) is added into port-place IMP2, showing the storage space of a workpiece j is added to the input buffer.

Step 5. Machining is completed. (1) At this point, the output buffer of the equipment is checked. If there is no needed storage space, the system keeps waiting. Otherwise, transition MT6 is triggered; the workpiece is moved to the output buffer; a token (j) is reduced from IMP4, indicating the storage space of a workpiece j is reduced in the output buffer. (2) A token (j, p) is added to IMP3 to show workpiece j is finished process p; the state of workpiece j is also updated. (3) The equipment is released, and a token (m) is added to MP0.

Figure 4.

Model of machining equipment.

Model of transport equipment

Transport equipment is used to move workpieces from one place to another. Transport equipment has the capability of active perception and logical thinking; it can perceive the arrival of the workpiece which must be transported and determine the transport path according to the status of the workpiece. There are four steps in the transport process: (1) getting the workpiece, (2) transporting, (3) releasing the workpiece, and (4) resetting.

The CPN-based transport equipment model shown in Figure 5 includes three colorsets: T is the equipment colorset, denoting the type of equipment; J is the workpiece colorset, denoting the type of workpiece; and Pa is the transport path colorset, denoting the path of transport. The model also defines three variables: t is the equipment variable, denoting the equipment used; j is the workpiece variable denoting the workpiece transported; and pa is the transport path variable, denoting the path used. Similar to the machining equipment model, the transport equipment model has state-place and port-place. The main process of the model is as follows:

Step 1. The equipment perceives the arriving workpiece and reads its status information, and the information is then inserted into the port-place ITP1 by its RFID reader. A colored token (j, pa) in IMP1 shows workpiece j is waiting for transport and the transport path is pa.

Step 2. The equipment’s own state is checked; if there is a token in TP0, the equipment is ready and the system can carry out the next operation. Otherwise, the system keeps waiting.

Step 3. The equipment gets the transport path from the path database according to the colored token (j, pa). If the path exists, the system will get it; if not, the system will show an error and return to a waiting state.

Step 4. The system begins to transport when the workpiece, the equipment, and the transport path are ready. Transition TT3 is triggered, and (1) a colored token (t, j, pa) is updated into state-place TP3, indicating the transport path pa of workpiece j is executed by equipment t. (2) The input buffer space of the equipment is released; specifically, a token (j) is added to port-place ITP2, showing the storage space of a workpiece j is added to the input buffer.

Step 5. Transport is completed. (1) At this point, the output buffer of the equipment is checked. If there is no enough needed storage space, the system keeps waiting. Otherwise, transition TT6 is triggered, and the workpiece is moved to the output buffer so that a token (j) is reduced from ITP4, showing the storage space of workpiece j is reduced in the output buffer. (2) A token (j, pa) is added to ITP3 to show the workpiece j has finished the transport path pa; the state of workpiece j is updated as well. (3) The equipment is reset, and a token (t) is added into TP5.

Step 6. The reset activity of transport equipment is finished. Transition TT7 is triggered, and a token (j) is added to TP0, showing that the reset activity of the transport equipment is successful and the equipment is waiting for the next transport task.

Figure 5.

Model of transport equipment.

Model of storage equipment

Storage equipment refers to space or containers used to store workpieces. Storage equipment is a passive resource providing storage space information to active resources. In Figure 6, the CPN-based storage equipment model has one colorset, J, denoting the type of workpiece. The model also defines one workpiece variable, j, denoting the stored workpiece. The model only has state-place.

Figure 6.

Model of storage equipment.

Resource perception and process control during manufacturing process

The manufacturing process is the process by which the different manufacturing resources interact. Target products are produced through a series of interactive activities, including machining, transport, storage, and so on. In the distributed system, the efficiency of the whole system depends on the perception and interaction efficiency of the various resources. In our proposed model, the method of resource perception and process control in production initialization, material transport, and workpiece machining is added to the manufacturing process as shown in Figure 1.

Initialization

At the stage of production initialization, the type of product is determined by the production plan. We begin by letting J1 denote the product type. Next, the ID of workpiece (#J1011) is generated automatically by the system, and the processing path of the workpiece is downloaded from the process database. The PN-based model of workpiece is built according to its manufacturing process, and the ID, PN-based model, and initial status of the workpiece are written into the RFID tag. Finally, the workpiece is put in the initial buffer (B1) and the state of B1 is updated.

Transport

The interactive process of transporting a workpiece from an initial buffer to an input buffer of M1 is shown in Figure 7. The detailed interactive steps are as follows:

Step 1. After production initialization, the workpiece is put into buffer B1. The transport equipment (AGV1) perceives it and reads the PN model from its tag; the state of workpiece #J1011 is analyzed by the AGV1 to decide whether it is to be moved by AGV1 or not. If the workpiece is to be moved, step 2 is taken. If not, the system stops.

Step 2. The transport path for #J1011 is acquired; the destination is M1_BI, namely, the input buffer of M1. The state of the workpiece is modified from “waiting for transport” to “in transport.” The token in P1 is deleted, and transition T1 is triggered; the border of transition T1 remains bold to indicate transport is in progress until the end of transport. The actual starting time, equipment, and transport path are updated into the workpiece model. The state of B1 is modified by adding a colored token, J1, to BP0 in the CPN-based model of B1. Finally, the state of the AGV1 is updated from waiting for transport to in transport.

Step 3. When workpiece #J1011 arrives at M1_BI, the CPN-based model of MI_BI is read by the AGV1 and its status is analyzed. Then, the AGV1 determines whether to place workpiece #J1011. If M1_BI can hold #J1011, step 5 is taken; otherwise, step 4 is taken.

Step 4. The system checks why M1_BI cannot hold #J1011. If the reason is insufficient capacity, the workpiece will wait until M1_BI has enough space before taking step 5. If the reason is that M1_B1 cannot accept the type of J1, the system will send out an exception.

Step 5. Workpiece #J1011 is put into M1_BI. Its state will be modified from in transport to end of transport, and the information on transport end time is written into the workpiece model. The state of M1_BI is modified and workpiece #J1011 waits for M1.

Step 6. When the AGV1 returns to the original location, transition TT7 is triggered, and the state of the AGV1 is modified from resetting to ready.

Figure 7.

Interaction diagram of workpiece transport.

Machining

When the M1 perceives workpiece #J1011 has arrived at M1_BI, M1 checks its own state using its own CPN model. If its state is not in the ready condition, the workpiece will wait in M1_BI; otherwise, the machining operation is carried out using the following steps:

Step 1. The PN model #J1011 is read by M1 and its state is analyzed. If #J1011 is not suitable to machine in M1 because of its type and state, the system sends out an exception. If it is suitable, the corresponding machining process is downloaded from the process database. If this is not successful, the system sends out an exception. If everything is ready, the machining transition is triggered.

Step 2. The workpiece state is modified from waiting for machining to being machined, and the machining equipment and starting time are written into the workpiece’s tag. Then, the equipment state is modified from the ready condition to in machining; the state of M1_B1 is modified as well.

Step 3. After the machining activities, the workpiece is checked to see whether it qualifies or is disqualified. If it is qualified, its state is modified from being machined to successful completion; if not, its state is modified as completed but scrapped. The state of M1_BO is checked to determine whether there is enough storage space for #J1011. If there is enough storage space, step 5 is taken. If not, step 4 is taken.

Step 4. The system checks the cause for #J1011’s rejection. If the reason is that the storage space is occupied, the workpiece will keep waiting until it is released. Then, step 5 can be taken. If the reason is that M1_BO cannot accept this type of J1, the system will send out an exception.

Step 5. The transition to the completion of machining is triggered. #J1011 is put in M1_BO, and the states of M1, #J10111, and M1_BO are modified. Finally, information on the completion time and the machining quality results is written into the PN model.

Experimental system implementation

To verify the availability and feasibility of the proposed method, we next build an experimental system. As shown in Figure 8, the experimental system contains two robots, two AGVs, one buffer, two machines. We opt for a high-frequency, non-source, and passive RFID system; the reasons for choosing this RFID system include its large memory capacity and the ease of controlling the identified area. In this system, every active resource (machines or transporters) is equipped with a terminal control computer connected to a RFID reader to read or write information from the tag attached to a passive resource. Every passive resource has an RFID tag. As shown in Figure 8, the process has seven steps, arranged according to the logical sequence of activities and involved operators. The steps are as follows:

Step 1. Basic information and configuration are managed, where all information on products, processes, equipment, and so on is managed and configured. The product bill of material tells the manufacturing system what to manufacture and assemble, the process contains information on how to manufacture and assemble, and the equipment declares where the production task is performed and who is in charge.

Step 2. All resources are modeled applying the suggested approach to information on products, processes, equipment, and so on. The models are then translated into XML for ease of reading, writing, and storing.

Step 3. The product models are written into the RFID tags, and the equipment models are stored in corresponding terminal controllers in the initialization processes.

Step 4. When the workpiece with its twinned model and attached tag is inputted into the manufacturing system, the equipment (machines or transporters) sense and identify it by the attached RFID readers; the model in XML schema is parsed, and appropriate operations are carried out by the equipment.

Step 5. When the models of the resources are parsed, the real-time status of resources can be monitored visually. The equipment’s real-time status is shown in its terminal controllers. The workpiece’s real-time status can be shown in two different ways: one appears in the active resource controller when the workpiece is being operated on; the other is shown in a specialized monitoring terminal, that is, a watchdog.

Step 6. When the equipment finishes the operations of a workpiece, it updates the state of the workpiece and its own state according to the operation results.

Step 7. After the workpiece is finished, the operation record is stored in the database for future tracing. The record will be analyzed so that the manufacturing system can be improved.

Figure 8.

The experimental system and its implementation.

Perception-decision-execution-perception closed-loop control is realized in the operation of the experimental system. Places map with a kind of control signal, knowledge data and state information; they are connected with corresponding PLC blocks, data elements in databases, or document directories. Their tokens are the values in the PLC blocks, datum in the databases or documents in the directories. Once the places of tokens are changed and their subsequent transitions are checked, if their fire condition is satisfied, they will be fired. Then, the tokens will changed in their subsequent place according to the execution results, and the corresponding values of the PLC blocks, datum in the databases or documents in the directories will be changed—the cycle continues until no transition can be fired.

In the manufacturing system development phase, the suggested approach improves efficiency while reducing cost and time. The development of a manufacturing system requires three steps: (1) modeling each kind of resource, (2) simulating, validating and modifying the models until they are approved, and (3) connecting with hardware if the models are approved. In other words, the manufacturing system can be validated before its implementation.

In the manufacturing system operation phase, the approach improves flexibility and reconfigurability. When the system needs to introduce a new resource belonging to an existing resource class, it only requires the instantiation of one more object and the addition of a new token in the corresponding PN model. The introduction of a new resource not belonging to an existing resource class requires the development of a new PN model that represents its dynamic behavior. The remotion of a resource only requires the remotion of the token associated with the resource in the corresponding models. The modification of the resource requires the modification of the PN model and its parameters and probably the modification of the associated models.

But there are two main limitations of the approach when it is applied to real industries. The first one is the memory capability of the RFID tag and the tag cannot store the whole needed information. In the experimental system, only the essential information is stored in tags. The second one is the real-time control capability. In the experimental system, only the unreal-time and on–off controls are mapped to the places of CPN models and the real-time control processes are mapped to specific PLC code blocks.

Conclusion and future research

The rapid diversification of global markets is changing traditional production modes, with distributed FMS becoming increasingly popular. RFID technology is accelerating this trend, as its applications in manufacturing systems have the potential to benefit the building of distributed and real-time factories. This article presents an extended PN-based modeling approach to manufacturing resource perception and process control for an RFID-based FMS. In the proposed approach, the real-time state and behavior logic models of manufacturing resources are built using extended PNs based on their classification. These models are integrated with the RFID system and manufacturing physical resources. In this way, the traditional manufacturing resources become smart objects with “brains”; they can “perceive,”“think,” and “control themselves” according to their attached real-time state and behavior logic models. Future research will apply the approach to more complex manufacturing systems; a specific goal is to find an exception-handling mechanism.

Footnotes

Acknowledgements

The authors thank the referees for their helpful comments and suggestions, which contributed to a significant improvement of the original version of this article. A special thanks goes to Dr Elizabeth Thompson for her edit.

Academic Editor: Murat Uzam

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Gratitude is also extended to the support from China National Science-technology Support Plan Project under Grant No. 2015BAF08B02, Beijing Jiaotong University Talent Fund Project under Grant No. 2012RC0003, and Fundamental Research Funds for the Central Universities of China under Grant No. 2015JBM078 for their support.

References

Colombo

Schoop

Neubert

An agent-based intelligent control platform for industrial holonic manufacturing systems. IEEE T Ind Electron 2006; 53: 322–337.

Liu

Wang

Ling

. RFID-enabled real-time manufacturing operation management system for the assembly process of mechanical products. Int J RF Technol 2015; 6: 185–205.

Ostgathe

Zaeh

. System for product-based control of production processes. In: Proceedings of the 2013 IEEE workshop on computational intelligence in production and logistics systems (CIPLS), Singapore, 16–19 April 2013, pp.138–144. New York: IEEE.

Leitão

Agent-based distributed manufacturing control: a state-of-the-art survey. Eng Appl Artif Intel 2009; 22: 979–991.

Marik

McFarlane

Industrial adoption of agent-based technologies. IEEE Intell Syst 2005; 20: 27–35.

Zaeh

Ostgathe

A multi-agent-supported, product-based production control. In: Proceedings of the 2009 IEEE international conference on control and automation, Christchurch, New Zealand, 9–11 December 2009, pp.2376–2383. New York: IEEE.

Zheng

Jiang

Qiao

. Challenges and frontiers of manufacturing systems. Chin J Mech Eng 2010; 46: 124–136.

Barenji

Hashemipour

A multi-agent RFID-enabled distributed control system for a flexible manufacturing shop. Int J Adv Manuf Tech 2014; 71: 1773–1791.

McFarlane

Sarma

Chirn

. Auto ID systems and intelligent manufacturing control. Eng Appl Artif Intel 2003; 16: 365–376.

10.

Qiu

RG.

RFID-enabled automation in support of factory integration. Robot Cim: Int Manuf 2007; 23: 677–683.

11.

Chen

Jwo

JS.

An RFID-based enterprise application integration framework for real-time management of dynamic manufacturing processes. Int J Adv Manuf Tech 2010; 50: 1217–1234.

12.

Zhang

Huang

. Agent-based smart objects management system for real-time ubiquitous manufacturing. Robot Cim: Int Manuf 2011; 27: 538–549.

13.

Huang

Williams

Zheng

Reactive, model-based monitoring in RFID-enabled manufacturing. Comput Ind 2011; 62: 811–819.

14.

Liu

Zheng

Sun

. RFID-enabled real-time production management system for Loncin motorcycle assembly line. Int J Comp Integ M 2012; 25: 86–99.

15.

Zhong

Dai

. RFID-enabled real-time manufacturing execution system for mass-customization production. Robot Cim: Int Manuf 2013; 29: 283–292.

16.

Zhang

Huang

Sun

. Multi-agent based real-time production scheduling method for radio frequency identification enabled ubiquitous shopfloor environment. Comput Ind Eng 2014; 76: 89–97.

17.

Wang

Jiang

Ding

A hybrid-data-on-tag-enabled decentralized control system for flexible smart workpiece manufacturing shop floors. Proc IMechE, Part C: J Mechanical Engineering Science. Epub ahead of print 2015. DOI: 10.1177/0954406215620452.

18.

Zhong

Lan

. Visualization of RFID-enabled shopfloor logistics Big Data in Cloud Manufacturing. Int J Adv Manuf Tech 2016; 84: 5–16.

19.

Leitao

Karnouskos

Ribeiro

. Smart agents in industrial cyber-physical systems. P IEEE 2016; 104: 1086–1101.

20.

Leitão

Colombo

. Petri net based methodology for the development of collaborative production systems. In: Proceedings of the 2006 IEEE conference on emerging technologies and factory automation, Prague, 20–22 September 2006, pp.819–826. New York: IEEE.

21.

Murata

Petri nets: properties, analysis and applications. P IEEE 1989; 77: 541–580.

22.

Jensen

Coloured Petri nets: basic concepts, analysis methods and practical use, vol. 1. Berlin: Springer Science + Business Media, 2013.

23.

Alla

Ladet

Martinez

. Modelling and validation of complex systems by coloured Petri nets application to a flexible manufacturing system. In: Rozenberg

(ed.) Advances in Petri nets 1984. Berlin; Heidelberg: Springer, 1985, pp.15–31.

24.

Feldmann

Colombo

AW.

Material flow and control sequence specification of flexible production systems using coloured Petri nets. Int J Adv Manuf Tech 1998; 14: 760–774.

25.

Feldmann

Colombo

AW.

Monitoring of flexible production systems using high-level Petri net specifications. Control Eng Pract 1999; 7: 1449–1466.

26.

Mendes

Leitão

Colombo

. High-level Petri nets control modules for service-oriented devices: a case study. In: Proceedings of the 34th annual conference of IEEE industrial electronics, 2008 (IECON 2008), 10–13 November 2008, pp.1487–1492. New York: IEEE.

27.

Tavares

ZDS

Jean-Paul

Saraiva

TA.

Elementary Petri net inside RFID distributed database (PNRD). Int J Prod Res 2010; 48: 2563–2582.

28.

Chen

KY.

Cell controller design for RFID based flexible manufacturing systems. Int J Comp Integ M 2012; 25: 35–50.

29.

Mendes

Leitão

Colombo

. High-level Petri nets for the process description and control in service-oriented manufacturing systems. Int J Prod Res 2012; 50: 1650–1665.

30.

Huang

Zhu

Cheng

. A Petri net-based approach for supporting traceability in cyber-physical manufacturing systems. Sensors 2016; 16: E382.

31.

Quintanilla

Cardin

L’Anton

. A Petri net-based methodology to increase flexibility in service-oriented holonic manufacturing systems. Comput Ind 2016; 76: 53–68.

32.

Yao

Jin

. Virtualization and servitization of cloud manufacturing resources. J South China Univ Technol 2013; 41: 1–7.

33.

Petri

CA.

Communication with automata. PhD Thesis, University of Bonn, Bonn, 1962.

34.

Jensen

A brief introduction to coloured Petri nets. In: Proceedings of the international workshop on tools and algorithms for the construction and analysis of systems, Enschede, 2–4 April 1997, pp.203–208. Berlin; Heidelberg: Springer.