Sage Journals: Discover world-class research

Abstract

A new generation of industrial revolution represented by intelligent manufacturing has come. Multi-product and small batch have become trend in market demands nowadays. Cyber-physical production system is a useful tool to meet this trend by performing tasks intelligently. Since cyber-physical production system performs tasks intelligently and autonomously, tasks and resources should be defined in standardized form in order to be identified and analyzed. Thus, how to define and analyze tasks and resources become the prerequisite for the operation of cyber-physical production system. In this article, a tasks and resources matching and analysis method is proposed for cyber-physical production system. First, the form of process knowledge base is designed, and the method for tasks and resources definition and matching is discussed. Next, the model of system operation logical based on colored Petri net and processing route generation algorithm is built. The parameters forecast method for non-standard processes is designed based on backpropagation neural network. Finally, implementation mechanism and effectiveness of the method are verified by a prototype system.

Keywords

Cyber-physical production system task-resource matching colored Petri net backpropagation neural network intelligent manufacturing

Introduction

With the increased requirements of personalized products, product quality, and service quality, the manufacturing systems need to meet the task of multi-variety, small batch, and personalized features. At the same time, with the continuous development of science and technology, advanced manufacturing technology and ideas are introduced and applied to manufacturing industry such as cloud computing, Internet of Things, big data, and cyber-physical systems (CPS).^1–4 Therefore, the traditional manufacturing model is facing great challenges. In the context of the new market demand characteristics and advanced technology development, the German “Industrial 4.0,” as the representative of the reform, opened the intelligent manufacturing-led fourth industrial revolution prelude. Industry4.0 evolved from embedded system to the CPS.⁵ CPS are engineered systems that are built from and depend on the seamless integration of computational algorithms and physical components.⁶ CPS inherits all the complexities of modern large-scale distributed systems: they have to handle uncertainty and change during operation, control their emergent behavior, and be scalable and tolerant to threats.⁷ CPS has now been applied to many fields. Combined with service-oriented architectures (SOA) and cloud techniques, P Leitão et al.⁸ presented an industrial intelligent CPS architecture scheme based on holons and Multi-Agent Systems (MASs) and studied the MAS-based service strategy. The European IMC-AESOP project is a visionary undertaking which investigates the applicability of cloud-based CPS and SOA in industrial systems, such as monitoring of paper production machines by Fluid House (Finland) and lubrication monitoring and control systems in mining industry by LKAB (Sweden).⁹ H Muccini et al.¹⁰ point out that a number of studies have been using self-adaptation features in the engineering of CPS, but there are various challenges for future research, both in the field of CPS and self-adaptation, including: how to map concerns to layers and adaptation mechanisms, how to coordinate adaptation mechanisms within and across layers, and how to ensure system-wide consistency of adaptation.

With the promotion and application of CPS, a CPS-enabled decentralized, enhanced manufacturing model is formed, which is cyber-physical production system (CPPS). According to Laszlo¹¹ and Uhlemann et al.,¹² CPPS is the combination of CPS and advanced manufacturing technology in depth, which uses CPS-related technologies and relies on different embedded devices. It forms a concurrent network through continuous calculation and communication, which can increase industrial production system flexibility and adaptability in complex and dynamic production environment. CPPS will establish a highly flexible personalized, digital, and intelligent production mode, matching tasks and devices intelligently.

Given the characteristics of CPPS, it contains a large number of different tasks, many of which are personalized and non-standard tasks. The intelligent devices also have different processing and manufacturing capabilities. However, the manufacturing system needs to analyze the tasks to match the devices and execute them autonomously and automatically. If there is no standard definition and matching methods of tasks and devices (resources), the production system will not work. Therefore, definition and matching methods in the system should be established, so as to provide the basis of System operation, which is an important research content in the CPPS field. Tao et al.¹³ studied the method of matching supply and demand in cloud manufacturing environment and designed a simulator. Y Li et al.¹⁴ divided product design into three elements—tasks, personnel and resources, analyzed their attributes, and designed a matching algorithm. Wang¹⁵ designed an attribute classification and formatting method of task in joint operations. D Strang and R Anderl⁹ chose Unified Modeling Language (UML) to model the communications and material flows in CPPS. In the model, knowledge about the individual components can be stored and used to make decisions in the assembly process. The above-mentioned related methods have some guiding significance to this research, but they are dedicated methods for its own application area, lacking of universal. Therefore, the expression method and corresponding matching method cannot be directly applied to the research.

In this article, the definition of tasks and devices (resources) in CPPS and the matching mechanism between them are studied, to form a universally applicable and complete theory. The main contents are summarized as follows: (1) the establishment of the process knowledge base; (2) the definition of the task: through the definition, the users can obtain the parameters required by the process, the task of the process line. The users can design tasks to automatically capture the process of the task line; (3) the definition of devices: through which the user can get the achievable process type and parameter standard for each device. (4) The establishment of matching mechanism: based on the definition of tasks and devices, extract and match the relevant parameters of tasks and devices, so as to form a mapping between tasks and devices. The tasks-resources mapping is shown in Figure 1. (5) A management information system containing all the studies to show the feasibility of the method.

Figure 1.

Tasks-resources mapping diagram.

Method of building process knowledge base

Standard definition in the process knowledge base is useful and necessary to describe a task or resource in a system. It allows the system to accurately identify the contents of the tasks and resources. The process knowledge base contains the following sections:

Parameter set, that is, the parameters needed to create a system or resource descriptions. The system needs to establish the parameter system according to the technical characteristics of different fields. The parameter set is denoted as $Paras = (Para_ID, Para_Name, Para_Unit)$ .

Process set, that is, all processes that the manufacturing system may perform. When the system accepts a new task, it can be identified and processed by the device for each process to correspond to the specific process in the set. The definition of the process includes the process number and process name. The process set is denoted as $Processes = (Process_ID, Process_Name)$ .

Process-parameter mapping relationship set. As different processes involved in different parameter requirements, in order to describe the tasks and resources, the mapping between processes and parameters is needed. The mapping relationship is represented by the process number and the parameter number set. Process-parameter mapping relationship set is denoted as $P a r a s_P r o c e s s e s = (P r o c e s s_I D; P a r a_I D_{1}, \dots, P a r a_I D_{n})$ .

Device definition and machining parameter acquisition method

Device definition

The definition of a device includes the definition of the device function, the definition of the capability, and the definition of the state. There are three attributes that need to be defined specifically, descriptive, technical, and performance attributes.

The descriptive attribute is defined as $(Machine_ID, Machine_Name)$ , that is, the device number and device name, used to distinguish between different devices.

The technical attribute is defined as $(Machine_Type, Machine_Para, Machine_Sta)$ , which is the set of process parameters that can be executed by the device, the corresponding set of execution parameters, and the current device status.

The performance attribute is defined as $(Machin e_{Time}, Machin e_{Qua}, Machin e_{C})$ , that is, the processing time, processing quality, and processing cost of the device for different processes.

Based on the above attribute definition, assuming the existence of device $p$ , for the process $j$ , the definition of the device rules is

$\begin{matrix} Machin e_{p} = (Machine_I D_{p}, Machine_Nam e_{p}, \\ Machine_St a_{p}, Machine_Typ e_{pj}, Machine_{Para}_{pj}^{k}, \\ Machine_{Time}_{pj}^{K}, Machine_{Qua}_{pj}^{K}, \\ Machine_C_{pj}^{K}) \end{matrix}$ (1)

where $k$ is the parameter index, and $K$ is the set of $k$ . The correspondence of $(Machine_Typ e_{pj}, Machine_{Para}_{pj}^{k})$ is obtained by the $Paras_Processes$ . Time, quality, and cost are important scheduling basis in the implementation of the task and varied among different processes. Equation (1) shows that for the implementation of process $j$ on device $p$ , its execution time, quality, and cost need to be determined by the parameter set $K$ .

Processing quality is the evaluation of equipment service quality when finishing the tasks, including processing accuracy, product qualification rate, and so on. For the $Machine_{Qua}_{pj}^{K}$ , we have

$Machine_{Qua}_{pj}^{K} = \sum_{h = 1}^{H} ω_{h} Q_{h}$ (2)

where $Q_{h}$ is the normalized metric of the service quality index of the device $h$ , $ω_{h}$ is the corresponding weight ( $0 \leq ω_{h} \leq 1$ and $\sum_{h = 1}^{H} ω_{h} = 1$ ), and $H$ is the number of devices service quality indicators.¹⁶ Based on the processing costs, the operating characteristics of each device, and historical data statistics, the average cost per unit of time $c_{p}$ can be calculated, which is

$Machine_C_{pj}^{K} = Machine_{Time}_{pq}^{K} \cdot c_{p}$ (3)

Due to the variation of coefficient, the processing time of different processes are not the same. It is difficult to fully enumerate them in the definition of the device. Therefore, for each device, the common standard time library is established in this study. The format is

$(Machine_Typ e_{pj}, Machine_{Para}_{pj}^{k}, Machine_{Time}_{pj}^{K})$

When the device receives the common standard process, it can directly get the time in the library for subsequent performance evaluation. When the relevant parameters in the received process are not incompatible with the library, backpropagation (BP) neural network is used for time prediction based on the common standard time library. The BP neural network prediction method will be described in detail in the following section. Equation (4) gives a unified way to obtain processing time

$Machine_{Time}_{pj}^{K} = f (Machine_Typ e_{pj}, Machine_{Para}_{pj}^{k})$ (4)

When the process is a standard process, the function $f$ is the retrieval function of the time library. When the process is non-standard, the function $f$ is the trained neural network.

Processing time quota method based on BP neural network

The neural network connects a large number of neurons to each other and forms a network to abstract, simplify, and simulate the human brain, thereby attempting to reflect the basic characteristics of the human brain. As a relatively mature method, it has been applied in many disciplines, including processing time forecast and quota, with its unique distributed storage, parallel processing, adaptive self-organizing self-learning, and highly fault-tolerant.¹⁷ Therefore, BP neural network is chosen to quota processing time, and specific operation will not go into details. The effectiveness of the method is verified by welding process with a workshop welder.

According to the processing characteristics, select the four factors related to processing time: steel plate thickness, electrode diameter, weld thickness, and weld length. Part of the data is shown in Table 1. MATLAB is used to train and test neural network.

Table 1.

Welding data.

Serial number	Steel plate thickness	Electrode diameter	Weld thickness	Weld length	Time
1	2	2.5	2	0.5	5
2	2.5	2.5	2.5	0.9	8.1
3	3	2.5	3	1.4	11.2

The parameters for the BP neural network are as follows:

The function “premnmx” is used to normalize the data.

The bipolar S-shaped function is selected as the transfer function of the first layer, and linear function is selected as the transfer function of the second layer.

The number of nodes in the hidden layer is defined using Kolmogorov theorem and empirical formula, which is 5.

The error target is set to be 10⁻⁶.

The training process is shown in Figure 2. It can be seen that the 570th iteration reached the training target.

Figure 2.

Sample training process chart.

Using another 14 sets of data as testing data, the results are shown in Table 2. The average error is less than 10%, indicating the forecast is good. Before the actual operation of the system, all the BP neural network are well trained and stored in the system to ensure the timeliness of processing.

Table 2.

Predicted value and actual value comparison for processing time.

Actual value	Predictive value	Actual value	Predictive value
9.9	10.84	33	32.36
8.8	10.26	40.7	41.66
9.9	5.34	52.8	52.61
11	5.4	67.1	65.24
13.2	13.14	82.5	78.95
18.7	17.94	118.8	104.56
27.5	24.44	136.4	128.15

Definition and analysis of tasks

Task definition method

Task definition could enable the system to automatically identify tasks and match the appropriate device according to the routing and process requirements. The task definition should include routing, task requirements, machining status, and results. Thus, three attributes are needed, namely, descriptive attributes, technical attributes, and economic attributes:

Descriptive attributes are defined as $(Task_ID, Task_Name, Step_ID, Step_Name)$ , corresponding to task number, task name, process number, and process name.

Technical attributes are defined as $(S t e p_T y p e, S t e p_P a r a, S t e p_T i m e, S t e p_Q u a, S t e p_P r e, S t e p_R e l)$ , which corresponds to the process technology category, the process parameter set, the processing time of the process, the processing quality of the process, the assembly of the immediately preceding process, and the set of related processes. Among them, the set of immediate predecessor process indicates that the process can be carried out only if all processes in the set are completed, and the set of associated processes in a certain process means a set of processes that cannot be carried out concurrently with the process. Through the two attributes $Step_Pre and Step_Rel$ , the processing route can be described.

Economic attributes are defined as $(Task_DT, Task_Ext, Step_C, Task_Pro, Step_Pro)$ , which corresponds to mission lead time, deferred cost, process cost, mission profit, and profit for each process.

Based on the above attribute definitions, for a task $i$ with the number of the process $j$ , the definition format of this task is

$\begin{matrix} Tas k_{i} = (Task_I D_{i}, Task_Nam e_{i}, Task_D T_{i}, Task_Ex t_{i}, \\ Task_\Pr o_{i}, Step_I D_{ij}, Step_Nam e_{ij}, Step_Typ e_{ij}, \\ Step_{Para}_{ij}^{k}, Step_Tim e_{ij}, Step_Qu a_{ij}, Step_\Pr e_{ij}, \\ Step_Re l_{ij}, Step_\Pr o_{ij}) \end{matrix}$ (5)

where $Step_{Para}_{ij}^{k}$ represents the relevant parameter of step $Step_I D_{ij},$ and $k$ is the index of the parameter. The association of $(Step_Typ e_{ij}, Step_{Para}_{ij}^{k})$ is obtained from the $Paras_Processes$ as defined above.

Generation method of task processing route based on colored Petri net

It is necessary to design the processing route of the task and the method to generate the logical model, in order to let task be identified and processed by the device. Petri net was first proposed by Dr Carl Adam Petri of Germany¹⁸ and later attracted the attention of academia and industry. After half a century of development, Petri net theory was gradually mature and widely used in the construction of model and simulation of the manufacturing process, such as path optimization, production scheduling, pipeline balance, and other optimization issues.^19,20 Petri nets are an important tool for describing and analyzing complex processes from a process perspective, and they can be used for dynamic simulation and system behavior analysis. The visualized Petri net model can be transformed into a storable language stored in a radio frequency identification (RFID) tag, sensed and parsed by the device terminal, and then makes intelligent decisions about the manufacturing process. Colored Petri net (CPN) is used as a logical model to characterize the processing route of the task in this article. The schematic diagram of the processing route CPN is shown in Figure 3.

Figure 3.

Schematic diagram of the processing route CPN.

In order to express the logical relationship between the order of processing and each process, five modes of transitions are defined, the definition of each transition is shown in Figure 4. In addition, the function $GetSumSta (sid, ss, ssum)$ is used to determine whether all branches of a process are finished. Each $Su m_{i}$ set includes the process number $Step_ID$ and the completion state $Step_Sta$ of all immediate predecessor processes of the branch. $ssum$ was set to 1 if only all of the $Step_Sta$ is 1, indicating the end of the branch.

Figure 4.

Interfaces of knowledge database management.

When receiving a task, the CPPS needs to analyze the task autonomously and generate the machining route. Therefore, an algorithm is needed to generate the above network according to the definition of the task. The specific algorithm is as follows:

Step 1. Traverse all the processes of the task, get the number n, and generate n sets S₁, …S_i, …, S_n, recording the corresponding Step_ID and Step_Sta; set all Step_Sta to be 0.

Step 2. Traverse the $Step_Pre$ of all processes, find the process that occurs only once, record the number $m$ and the $Step_ID$ of its next process, generate $m$ normal execution transitions, and record the process $Step_ID$ which it represents.

Step 3. Generate a $n \times m$ matrix, and marked the position in the matrix with 1 or −1, according to the general implementation of the transition and the relationship between the process.

Step 4. Traverse the $Step_Pre$ of all processes, find the number of processes that occur multiple times, record the times as $p$ . Find the $Step_ID$ of its next process and generate $p$ process sets of $Step_Sum$ . Generate $p$ branches to start the transition and record the corresponding process set $Step_Sum$ .

Step 5. Generate a $n \times p$ matrix on the right side of the original matrix, and label the corresponding position in the matrix as 1 or −1, according to the relationship between the branch transition and relationship of processes.

Step 6. Traverse the $Step_Pre$ of all processes, find the $Step_ID$ of the process that has multiple $Step_Pre$ . Record the number of these processes as $q$ , and find the corresponding step $Step_ID$ , respectively, to form $q$ process sets $Step_Sum$ . Generate $q$ branch-ending transition, and the process set library $Su m_{1}, \dots Su m_{i}, \dots, Su m_{n}$ , and record the corresponding process set $Step_Sum$ .

Step 7. Increase $q$ rows and $q$ columns on the basis of the original matrix, and label the corresponding position in the matrix as 1 or −1, according to the relationship between the transitions and the process.

Step 8. Traverse all processes $Step_rel$ and $Step_Sum$ . If there is an associative matrix in $Step_Sum$ , mark the corresponding branch-starting transition and branch-ending transition as non-parallel, and the others are marked as parallel. The algorithm ends.

Tasks-devices matching method

The matching mechanism of the tasks and devices needs to be designed after the definition of them. In this article, the parameters matching function library is like this

$FUNCTION_Mat = (Para_I D_{k}, Fun_Ma t_{k} (Step_Par a_{i}, Machine_Par a_{p}))$

For a certain type of process parameter $Para_I D_{k}$ , whose corresponding process parameter and device parameter are $Step_Par a_{i}$ and $Machine_Par a_{p}$ , there is a function $Fun_Ma t_{k}$ to verify whether $Machine_Par a_{p}$ meets the parameters $Step_Par a_{i}$ . The matching procedure of process $Step_I D_{ij}$ can be described as follows:

Step 1. Obtain $Step_Typ e_{ij}$ and $Step_{Para}_{ij}^{k}$ of the process, and define $Machine_Typ e_{p}$ based on $Step_Typ e_{ij}$ .

Step 2. Find all machines that can carry out this process according to $Machine_Typ e_{p}$ and obtain the corresponding $Machine_ID$ and $Machine_{Para}_{pj}^{K}$ .

Step 3. Traversal $Step_{Para}_{ij}^{k}$ to find $Machine_Para$ based on $Step_ID$ , and call the function $FUNCTION_Mat$ to find all the machines satisfying conditions and record their $Machine_ID$ .

Step 4. Confirm the results of all machines according to equations (2)–(4).

System development and verification

In this section, a task analysis and matching prototype system is developed to verify the framework. The system framework includes three parts, comprising knowledge base management, devices management, and task management. Knowledge database management contains the managements of all parameters and processes, the mapping relation among them, and the matching functions. Devices management contains the management of all machineries’ information, such as basic information, related functions, parameters, processing time, and machining quality. Task management includes information definition and logic analysis of all tasks and the information of device matching.

Figure 4 shows the knowledge database management. Figure 4-A is the parameter sets and Figure 4-B is the process information and its parameter sets. Take the process of MF1 as an example, the parameters and their declaration is needed for processes matching, as shown in Figure 4-C

$\begin{array}{l} P a r a s_P r o c e s s e s = (P r o c e s s_I D; P a r a_I D_{1}, \dots, P a r a_I D_{n}) \\ = (M F 1; M P P 01, M P P 02, M P P 07, M P P 10, E P 02, E P 04) \end{array}$

Figure 5 shows the device resources management. Figure 5-A is basic information of device. Take milling machine “X208C” for an example, its functions contain rectangular plane, rectangular groove, boss and its parameters include processing length and processing width. Figure 5-B is the functional information of device. Take welding device whose number is ZF500 for an example, its functional information has three parts. The first part is task time library, the second part is standard time, and the third part is the statistics of machining quality.

Figure 5.

Interfaces of device resources management.

Figure 6 shows the task management. Figure 6-A is the interface of task definition. Take process T01S01 as an example, it has three executable devices according to the results of matching functions and its detailed execute parameters are shown in Figure 6-B. Figure 6-C shows the information of Petri network which is obtained according to the proposed process definition and the analysis method, including the information of transition, place, and incidence matrix.

Figure 6.

Interfaces of task management.

This system implements the proposed method by setting up knowledge database and specifying tasks and devices definition. The executable device of each process, actuating logic of tasks and CPN are obtained by matching and analyzing, which provide effective support for the operation of CPPS.

Conclusion

To meet the intelligent requirements of CPPS, methods of tasks and resources matching and analysis are prerequisite foundation for the production system operation. To achieve this foundation, the definition rules of tasks and resources are set, and the matching method is designed. The logic model based on the color Petri net is designed to express the logic of the task implementation, and the forming algorithm is designed to guarantee the system intelligence. Finally, a prototype system has been developed, and the relevant examples and functions are built to verify the effectiveness of the relevant system and function.

The main contribution of this article is setting some definition rules and combining with some existing methods, and creatively designing some new methods such as the Petri net forming algorithm. All the methods combined together to guide the system to identify the implementation process of tasks, and implement them autonomously, which is achieved by a prototype system. This article provides foundations for autonomous CPPS operation and gives the necessary basis for further intelligent decision making of the system. The next major task is to research CPPS optimization operation mechanism such as scheduling rules.

Footnotes

The authors thank the referees for their helpful comments and suggestions,which contributed to a significant improvement in the original version of this article.

Handling Editor: Peter Nielsen

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: This work was supported by National Natural Science Foundation of China under Grant No. 51775033 and China National Science-technology Support Plan Project under Grant No. 2015BAF08B02.

ORCID iD

Zengqiang Jiang

References

Rosenthal

Mork

et al . Cloud computing: a new business paradigm for biomedical information sharing. J Biomed Inform 2010; 43: 342–353.

Zhao

. The internet of things: a survey. Inform Syst Front 2015; 17: 243–259.

Naimi

Westreich

. Big data: a revolution that will transform how we live, work, and think. Math Comput Edu 2013; 47: 181–183.

Mosterman

Zander

. Industry 4.0 as a cyber-physical system study. Software Syst Model 2016; 15: 17–29.

Cheng

Liu

Qiang

et al . Industry 4.0 development and application of intelligent manufacturing. In: Proceedings of the international conference on information system and artificial intelligence, Hong Kong, China, 24–26 June 2017, pp.407–410. New York: IEEE.

Chang

Gao

Lei

et al . Cyber-physical systems in manufacturing and service systems. Math Prob Eng 2015; 2015: 704213.

Bures

Weyns

Tovar

et al . Software engineering for smart cyber-physical systems: challenges and promising solutions. ACM SIGSOFT Soft Eng Note 2017; 42: 19–24.

Leitão

Colombo

Karnouskos

. Industrial automation based on cyber-physical systems technologies: prototype implementations and challenges. Comput Indus 2016; 81: 11–25.

Strang

Anderl

. Assembly process driven component data model in cyber-physical production systems. Lecture Notes in Engineering and Computer Science 2014; 2214: 947–952.

10.

Muccini

Sharaf

Weyns

. Self-adaptation for cyber-physical systems: a systematic literature review. In: Proceedings of the IEEE/ACM, international symposium on software engineering for adaptive and self-managing systems, Austin, TX, 16–17 May 2017, pp.75–81. New York: IEEE.

11.

Laszlo

. Cyber-physical production systems: roots, expectations and R&D challenges. Proced CIRP 2014; 17: 9–13.

12.

Uhlemann

THJ

Lehmann

Steinhilper

. The digital twin: realizing the cyber-physical production system for industry 4.0. Proced CIRP 2017; 61: 335–340.

13.

Tao

Cheng

et al . SDMSim: a manufacturing service supply–demand matching simulator under cloud environment. Robot Comput: Integrat Manuf 2016; 45: 34–46.

14.

Zhang

et al . Static scheduling of product design projects based on personnel—task—resource matching degree. Indus Eng Manag 2012; 17: 74–80.

15.

Wang

. Modeling for joint operation requirements based on task-capability matching. Command Inform Syst Technol 2011; 2: 5–9.

16.

Yin

Luo

et al . Optimized selection of machine tool equipment based on multi-agent. Comput Integrat Manuf Syst 2016; 22: 1474–1484.

17.

Guo

. Research on the calculation method of man-hour quota based on genetic neural network. Beijing, China: China Institute of Engineering Physics, 2009 (in Chinese).

18.

Petri

. Communication with automata, 1966, http://edoc.sub.uni-hamburg.de/informatik/volltexte/2010/155/

19.

Panescu

Pascal

. Holonic coordination obtained by joining the contract net protocol with constraint satisfaction. Comput Indus 2016; 81: 36–46.

20.

Yang

et al . Modeling and balance of aircraft assembly line based on expanded Petri net. Comput Integrat Manuf Syst 2015; 21: 2596–2603 (in Chinese).

Method of tasks and resources matching and analysis for cyber-physical production system