Special Issue Article

Advances in Mechanical Engineering

2017, Vol. 9(2) 1–16
Ó The Author(s) 2017
Development of a collaborative DOI: 10.1177/1687814017691437
journals.sagepub.com/home/ade
material management system for
offshore platform projects using agent
technology

Duanfeng Han, Qinghua Zhou, Jinghua Li, Jiaxuan Wang, Boxin Yang and
Miaomiao Sun

Abstract
Accuracy management of production materials plays an important role in offshore platform projects. To improve both
the efficiency and accuracy of material management in offshore platform projects, a novel material management system
characterized by the notion of a material information unit was proposed in light of a review of the industrial Internet of
Things and multi-agent technology. The material information unit is used to carry data on production materials and serve
as a basic information unit in the system. Furthermore, agents of different functions are employed to tackle assorted
tasks. Using this system, material information can be precisely recorded, tracked, and traced during the manufacture of
offshore platforms. In addition to this system, an algorithm targeting at multi-agent negotiation is designed. Finally, a pro-
totype system is developed to verify the feasibility and efficiency of this material management process.

Keywords
Offshore platform project, material management, material information unit, multi-agent system, agent collaboration

Date received: 31 August 2016; accepted: 10 January 2017

Academic Editor: Murat Uzam

Introduction Since the 1990s, information technology has been

As the ocean resources such as marine oil and gas have
been explored, these years have witnessed the sharp
demand for offshore engineering structures such as off-
shore platforms and pipe ships, and it is expected to
continue to grow. An offshore platform project (OPP)
usually includes the processes of shipbuilding and the
installation of offshore equipment,1,2 which differs
from traditional shipbuilding processes. Furthermore,
the manufacturing process used in OPPs involves more
outsourcing of tasks, mechanical components, and
equipment operation. Nowadays in China, most marine
structures are manufactured by shipyards whose man-
agement methods are undeveloped, leading to numer-
ous problems in material management such as waste
and additional costs.
applied to the control of production materials using dif-
ferent methods to achieve higher production efficiency
and more accurate material management in the OPP
manufacturing process. With the development of the
Internet of Things (IoT) and its supporting technology,
material information management that applies to every
single object has been technically feasible since 2010,3
but it needs to be equipped with an enterprise data sys-
tem that has strong information processing capability.
Harbin Engineering University, Harbin, China
Corresponding author:
Jinghua Li, Harbin Engineering University, Chuanhai Building, 145 Nantong
Street, Nangang District, Harbin 150001, Heilongjiang, China.
Email: likewalls@163.com

Creative Commons CC-BY: This article is distributed under the terms of the Creative Commons Attribution 3.0 License
(http://www.creativecommons.org/licenses/by/3.0/) which permits any use, reproduction and distribution of the work without
further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/
open-access-at-sage).
2 Advances in Mechanical Engineering
However, in China, information management is not
sufficiently developed for large-scale projects. So far,
no material information management or material qual-
ity tracing system has been applied to OPPs. In accor-
dance with the manufacturing features of OPPs, the
organizational structures of production enterprises are
relatively independent, but work in close collaboration.
This is similar to the multi-agent system (MAS) model,
and thus, the application of MAS technology is pre-
sumed to be compatible with the existing digital man-
agement systems.
Considering the above observations, an agent-based
offshore platform project material management system
(OPPMMS) is presented. The OPPMMS is an intra-
enterprise system in view of the enterprise production
network of an advanced offshore platform manufactur-
ing enterprise. The management cycle in the OPPMMS
extends from the purchase that has arrived at factory to
the completion of product delivery. The management
functions for OPPMMS are to collect and process the
massive amount of material information acquired by
the IoT during the complex OPP production process,
which facilitates the further refinement of quality and
process management.
The rest of this article is organized as follows. Section
‘‘Literature review’’ outlines some related work regard-
ing OPP material information management, agent tech-
nology, and the industrial IoT in view of the literature
reviews. Section ‘‘Information processing mechanism in
the OPPMMS’’ establishes the information processing
mechanism for the OPPMMS by defining the notion of
the material information unit (MIU) and the model that
is built on it. Section ‘‘OPPMMS architecture’’ mainly
discusses the OPPMMS architecture, including the func-
tional definition and internal structure of the individual
agent. Section ‘‘OPPMMS working mechanism’’ focuses
on working mechanisms that handle a lot of materials
and complicated processes involved in OPP manufactur-
ing processes and a large volume of MIU data generated
at different locations on the enterprise IoT. Section
‘‘OPPMMS validation’’ validates the effectiveness of
MIU and related management methods and the feasibil-
ity of system using examples of MIU data processing
and a prototype system performance.
Literature review
The existing studies can be categorized into three
domains, namely, OPP material information manage-
ment, agent technology, and the industrial IoT. This
section provides a brief review of several influential
studies in these domains.
Material information management
Material information management, as part of the man-
ufacturing process, operates in two directions. One is
Figure 1. Material information tracing and tracking during the
manufacturing process.
material record tracking and the other is material infor-
mation tracing. The relationship between OPP material
management and the manufacturing process is shown
in Figure 1.
Tracking and tracing are two fundamental aspects of
material information management.4 The tracking pro-
cess has the same flow direction as the production pro-
cess, during which material transformation data are
recorded in real time from raw material usage to the
composition of the finished product. On the contrary,
the tracing process flows in the opposite direction and
displays historical data relating to material transforma-
tion, starting from the end products and/or by-products
and working back to the raw materials used. Currently,
material tracing and tracking management are attract-
ing considerable attention in various industrial sectors.
A standardized data format that can reflect the stability
of the information system is urgently needed.
When it comes to files recording material informa-
tion management, modern processors generally view
the bill of materials (BOMs)5 as an important means of
tracking the production process. As a kind of static
document regarding the production process in an enter-
prise resource planning (ERP) system, the BOM defines
the OPP product structure, constituent parts, and
installation order. The BOM can be used to either
retrieve the assembly relationship among materials or
define the attributes of materials. However, material
management in an ERP system focuses on logistics and
cost management, the main function of which is to
record the quantity of materials used in the production
process. Furthermore, OPP manufacturing is such a
dynamic and complex process6 that the process of mate-
rial information recording and management in original
OPP-ERP system which uses BOM alone is insufficient
to meet the requirements of management regarding
material information traceability and trackability.
Therefore, more items are needed to provide the infor-
mation that is required. Apart from the buying bill of
material (B-BOM) and cost bill of material (C-BOM) in
the original ERP system, the OPP enterprise production
management process has developed a wide range of
BOM forms,7 such as the engineering bill of material
(E-BOM) from the product data management (PDM)
system, the process bill of material (P-BOM) from the
computer-aided process planning (CAPP) system, and
Han et al.
the manufacturing bill of material (M-BOM) from the
computer-aided manufacturing (CAM) system to
achieve a more detailed record of system data. These
additional types of BOM use different data processing
methods and formats for the same materials in the pro-
cess of assembly, production, and logistics, and frag-
mented data sources, which adds an additional data
conversion process for material information retrieval
and restricts the effectiveness and efficiency of material
information tracking and tracing. Therefore, a unified
and standardized information unit that enables suffi-
cient accuracy in terms of the trackability and traceabil-
ity of production materials to meet information
management requirements is required. In relation to
material information management during OPP manu-
facturing, Li and Liu7 designed a new material manage-
ment file called the bill of lots (BOL), which combines
diverse aspects of information with the BOM, and then
proposed the concept of a traceable entity unit (TEU)
for material information management. However, the
detailed design and application of the TEU were not
presented. Even though the composition of material
codes and its relationship with manufacturing have
been widely researched8,9 in shipbuilding projects, tra-
ceability properties have not yet been considered. In
this regard, the material coding system that is used in
the petrochemical industry10 is of great reference value.
Agent technology
As a new branch of artificial intelligence, agent
technology11 has been studied since the 1970s. The
MAS,12 a form of distributed artificial intelligence, has
been studied and applied in various areas. In the
domain of large-scale engineering, applications aided
by the MAS have been developed for quality manage-
ment,13,14 product design,15,16 task scheduling and
planning,17–19 supply chain systems,20,21 and workflow
systems.22–24 There is a scarcity of research on the
application of MAS technology to the development of
product manufacturing systems, such as manufacturing
resource management,25,26 control of manufacturing
systems,27 manufacturing task assignments,28 and man-
ufacturing system integration.29 The distribution and
collaboration features of the MAS make it suitable for
assisting complex multi-sector manufacturing systems.
In this sense, the majority of the existing system mod-
ules can be simulated by the agent or at least connected
with the MAS by the data interface agent (IA) to
improve OPP material management performance.
Mobile agent technology is another effective way to
complete distributed tasks and has attracted extensive
interest.30 In agent-based systems, a mobile agent has
the capability of working regardless of the agent run-
time environment deployed in enterprise nodes. From
this viewpoint, it seems ideally suited to the execution
3
of bulk and varied tasks. Indeed, combined with the
MAS, the mobile agent is generally utilized to develop
intelligence production data systems, such as data ware-
housing,31 sensor networks,32,33 and production data
collection.34 Thus, mobile agent technology is capable
of being applied to management processes dealing with
a huge amount of OPP material information collected
by sensors.
Industrial IoT
To obtain accurate material information, it is necessary
to implement an industrial IoT.35 This achieves direct
communication between the materials and the enter-
prise’s network. Using such a network, material infor-
mation and the associated instructions can be traced
over time. An important element of the IoT is identifi-
cation tag technology.36 Tags are attached to the mate-
rial entity and data are transited directly through the
IoT, enabling the information to be processed more
smoothly and accurately. Thanks to the development
of identification tags and wireless technology, a series
of tags such as radio frequency identification
(RFID)37–40 and quick response (QR) codes40,41 have
been used in large-scale manufacturing processes dur-
ing the last few decades. Due to the generic RFID,
Kang and Lee42 proposed and developed a novel set of
traceability services for manufacturing management.
Velandia et al.43 used RFID to trace and track material
information for the production of camshafts.
Chongwatpol and Sharda44 presented an RFID-based
traceability approach to improve production schedul-
ing performance.
IoT information tags need to be pasted or etched
onto the material surface. For the large steel structures
in OPPs, surface tags will be more likely to abrade or be
lost during the manufacturing process. Moreover, steel
structures possess a strong magnetic field, which is not
conducive to the accurate transmission of RFID infor-
mation. As a result, the reading and writing of IoT tags
are more likely to be wrong or become lost, therefore
the data recording system of the IoT should be avail-
able for data analysis.
Discussion
The following conclusions can be drawn after analyzing
the situation regarding OPP material management:
1. Although the TEU was proposed to facilitate
material management during the OPP manufac-
turing process, it is only a conceptual model,
lacking coding rules and content specifications,
which is far from satisfactory for efficient OPP
material information management.
4 Advances in Mechanical Engineering

2. The existing material management systems for material management process. Then, using industrial
OPPs do not support management of single- IoT tags and agent technology, an agent-based proto-
material records and the traceability and track- type framework is set up for OPP enterprises to manage
ability of all material data. These insufficiencies material information. In the following sections, we ela-
are magnified when a large amount of material borate on the standard MIU design for material man-
data need to be managed in OPP manufactur- agement, an analysis model and algorithm based on the
ing. Furthermore, information that is merely MIU, agent framework construction, and the agent
identified by a material tag without analysis is cooperation mechanism.
prone to data errors. Therefore, these systems
are not suitable for material information man-
agement for OPPs. Information processing mechanism in the
3. To effectively manage material data during the OPPMMS
construction of OPPs, it is necessary to connect
MIU
the material entity with the industrial IoT. In
spite of the fact that IoT tags enable communi- Standardization of information files should make mate-
cation with the material entity, the extremely rial management more efficient. In this section, the
harsh manufacturing conditions encountered in MIU is defined as a standard of the OPPMMS and an
constructing OPPs make it easy to lose tag analysis model is then developed.
information.
Structure of the MIU. In accordance with the BOL7 and
In response to the above analysis, this article pre-
the summary of the characteristics of material informa-
sents the following technical solution:
tion in the production process, the material information
1. Define the MIU. The MIU serves as the base unit to be included in the MIU for the OPP manufacturing
for the proposed system and is used to handle process is as follows: (1) material identification infor-
material information. In this article, the MIU is mation, (2) material logistics information, (3) material
defined for material management, and a reason- processing information, (4) material assembly informa-
able MIU data format is designed, ensuring that tion, and (5) other correlated and remark information.
the requirements for traceability and trackability Figure 2 shows the information that should be carried
of material management are satisfied. by the MIU.
2. Design the MIU analysis function for material According to the above analysis, material informa-
information management. Given the characteris- tion obtained during the OPP manufacturing process
tics of OPPs and the nature of the MIU, an can be divided into two parts: material attribute
MIU analytical model is presented for material information and material correlation information.
information management. Mobile agents that Thus, two categories of MIU are designed, namely,
are to be adopted in the OPPMMS are used to attribute-material information unit (a-MIU) and
rectify the MIU in a timely manner so as to correlation-material information unit (c-MIU). The
minimize the negative influences of poor pro- former represents the corresponding material attribute
duction conditions. The analysis function is
therefore proposed after considering features of
the MIU model and agent technology.
3. Develop an OPPMMS framework based on
agent technology. As a branch of computational
intelligence, the agent-based system has the
capacity to perform multiple distributed and
parallel tasks. Therefore, developing an agent-
based prototype of the OPPMMS framework is
believed to be advantageous. The capability to
process huge amounts of data will be improved,
and the ability to collect and analyze informa-
tion will be enhanced. In addition, other systems
and approaches are able to connect with the
OPPMMS framework through data-compatible
interfaces.
In this study, a standard data format for MIU is
designed after analyzing the requirements of the OPP Figure 2. Material information carried by the MIU.
Han et al.
information, while the latter represents the correlation
information. This arrangement enables the OPPMMS
system to meet the traceability and trackability
demands of material information management systems.
The MIU is formatted with data codes for use by
OPP manufacturers. The MIU encoding rules consist of
sequential a-MIU and c-MIU sub-rules, each of which
has a corresponding data code fragment. These encod-
ing rules are aligned with enterprise characteristics and
actual production requirements. The encoding rules of
a-MIU and c-MIU are discussed in the next section.
a-MIU. The a-MIU contains unique descriptions to
enable different materials to be distinguished in the
OPPMMS and is derived from a modification of assem-
bly BOM information, reflecting the assembly structure
of a certain offshore platform. In fact, different materi-
als vary in terms of assembly levels in OPP manufactur-
ing processes. For example, raw materials, work in
progress (WIP) products, and finished products are
typically located in three separate assembly levels. The
layered encoding structure in the OPP assembly process
and the a-MIU encoding rules are separately exhibited
in Figures 3 and 4, respectively.
In the a-MIU code, the material assembly level is
defined by the material assembly sequence within the
BOM. Moreover, the material sequence number speci-
fies the unique material in this level. To avoid duplicat-
ing material codes in different projects or different
structures, the upper level assembly code and project
code need to be listed for each material, which ensures
a unique a-MIU code and satisfies the identity coding
demands. Compared with identity coding using other
encoding rules, this has the advantage that the assem-
bly sequence of the product is encoded by the a-MIU,
and on this basis, a related data management function
can be developed.
Figure 3. OPP assembly levels.
5
c-MIU. Correlation refers to relationships between the
materials and its production activities. During the pro-
duction process, the materials are processed into fin-
ished products through a series of processing steps. In
each step, the materials are correlated with processing
information. The c-MIU consists of material produc-
tion, logistics, and other aspects that involve opera-
tional records for the materials. This c-MIU
information is acquired in a timely manner through
IoT sensors and formalized using specific rules. Thus,
the material’s status can be obtained and recorded. In
this sense, precise operational information relating to
materials recorded by the MIU in the various produc-
tion stages can be retrieved. An example of an MIU
explaining the encoding rules in relation to both
a-MIU and c-MIU is designed in Figure 4.
The MIU is encoded with two code fragments. In
this way, the corresponding material records can be
obtained accurately through the MIU encoding
method, which achieves the traceability and trackabil-
ity necessary for material information management.
MIU data model. From the above analysis, the MIU data
model illustrating the relationship between a-MIU and
c-MIU can be demonstrated via a tree structure. The
MIU data model is demonstrated in Figure 5.
Three components are selected to illustrate this
model, and their a-MIUs are designated as F1, F2, and
E1. In Figure 5(a), the assembly process of F1 and F2
into E1 and the relationships between the transporta-
tion and warehousing processes are recorded.
Correspondingly, the MIU including the middle part of
the encoding rules and both sides of F1, F2, and E1 is
manifested in Figure 5(b). In particular, the abbre-
viated form of the c-MIU is expressed in brackets. The
a-MIU and c-MIU are stored in the form shown in
Figure 5(c). Furthermore, Figure 5(d) demonstrates the
data model after considering the relationship between
the a-MIU and c-MIU.
The encoding rules and data model used for the
MIU provide a data standard for processing the huge
amount of data emerging from the Intranets of OPP
enterprises. The production structure information and
construction procedure information in the BOM are
able to correspond effectively through the MIU, which
not only enables bidirectional searching but also pro-
vides fundamental knowledge for material information
analysis.
Analysis based on the MIU
MIU-relevant model. The relevant model using both a-
MIU and c-MIU represents the relationship between
OPP production and the materials used. In fact, it
reflects the relationship between OPP construction data
from the industrial IoT and BOM assembly
6 Advances in Mechanical Engineering

Figure 4. Encoding rules for a-MIU and c-MIU.

Figure 5. MIU data model: (a) original material information, (b) material information unit (MIU), (c) MIU data storage format, and
(d) data model of MIU.

documentation from the CAM system. If we set the rel-
evant factor u for a pair of c-MIU and a-MIU, the
MIU-relevant model will be created in the form of a
matrix, as illustrated in Figure 6.
Figure 6(a) manifests the data model of the MIU,
which illustrates the correspondence between a-MIU
and c-MIU. Figure 6(b) exhibits the value of relevant
factor u. Actually, u is an integer that refers to the
recording frequency of the relational records of a-MIU
and c-MIU shown in the database. In the OPP
manufacturing process, the values of u are defined in
three distinct situations as follows:
1. u = 0 denotes that the corresponding materials
c-MIU and a-MIU are not relevant. For
instance, in Figure 6(b), line 1, column 3,
c-MIU(W1) is not relevant to a-MIU(F2), thus
the relevant factor u13 = 0.
2. u = 1 means that the corresponding materials
c-MIU and a-MIU are relevant. In Figure 6(b),
Han et al. 7

Figure 6. MIU-relevant model: (a) data model of MIU, (b) value of relevant factor u, and (c) relevant model of MIU.

line 1, column 1, c-MIU(W1) is relevant to

a-MIU(E1), thus the relevant factor u11 = 1.
3. u . 1 indicates that the corresponding MIUs
contain coding errors and need further process-
ing. For example, if u = 2, this reveals that the
relationship record of a pair of a-MIU and
c-MIU is stored twice, that is, this is an MIU
encoding error, possibly as a result of a problem
with the corresponding IoT data sensor or data
nodes.

The horizontal coordinate axis is created by sorting

the a-MIU in accordance with the assembly order in
the BOM, and the longitudinal coordinate axis is cre-
ated by sorting the c-MIU based on the starting time of
the operation; thus, an MIU-relevant model with a rec-
tangular coordinate system is generated. Then, the rele-
vant MIU model can be simplified as a matrix:
U = (uij)a 3 c, which is designed in Figure 6(c). In this
model, uij means the relevant factor of a-MIUi and c-
MIUj; the parameters i and j represent the vertical and
horizontal coordinates of the MIU in the coordinate
system, respectively; and a and c are the amounts of a-
MIU and c-MIU, respectively, in the relevant model.
MIU analysis function. In light of the previous definition
of the MIU, in the MIU-relevant model, the required
records for the corresponding materials can easily be
obtained using the MIU. However, because of the tech-
nical characteristics of the IoT tag and the complexity
of the OPP production network, data errors in the
MIU are unavoidable. Therefore, an automatic analysis
function for MIU data is designed for the OPPMMS
system to identify possible erroneous MIU data and
contact the relevant data source for further processing.
According to the MIU-relevant model, this problem
can be described as a dualistic function with index (i, j),
where i and j are described as above, as demonstrated in
Figure 7. Generally, the OPP production process is con-
tinuous, and so the relevant model embodies a
Figure 7. Various cases in the MIU-relevant model.
continuous area composed of the points where u = 1. If
the value of u at point (i, j) is not continuous, as pre-
sented in case 2 and case 3 in Figure 7, then various
points in the section from (i 2 k, j 2 k) to (i + k, j + k)
need to be selected to acquire the corresponding MIU
for processing. The value k defines the size of the selec-
tion space, and k = 1 is common. In addition, for incor-
rect records where u . 1, the adjacent points are more
likely to arise from MIU data error; thus, the corre-
sponding MIUs of the adjacent points need to be
selected for processing, as displayed in case 1 in Figure 7.
In the agent-based OPPMMS, the above formula
and criteria are used to MIU analyze algorithm design
for intelligence agent. If we define that uij can be
expressed as u(i, j), an algorithm of the MIU analysis
function is described as pseudo-code, as manifested in
Table 1.
8 Advances in Mechanical Engineering
Table 1. Algorithm for MIU analysis.
Index Content
1 Specify the analysis interval as a matrix: U = (uij)a3c,
where a and c,
respectively, represent the number of records of a-MIU
and c-MIU in analysis interval
2 Enter the search radius k for uij
3 Traverse the search space, marking the conditions
u-point:
for i = 1 to a
for j = 1 to c
if uij . 1, mark the uij and nearly point from
u(i 2 k, j 2 k) to u(i + k, j + k)
else for i1 = i 2 k to i + k
for j1 = j 2 k to j + k
if u(i1, j1) 2 u(i, j)  1,
mark point (i, j) and nearly point from
(i 2 k, j 2 k) to (i + k, j + k)
4 Record all marked data for further operation
a-MIU: attribute-material information unit; c-MIU: correlation-material
information unit.
OPPMMS architecture
System framework
The OPPMMS that we have developed is based on a
multi-agent structure and consists of a manager agent
(MA), an encoder agent (EA), an analyzer agent (AA),
an IA, and a resource agent (RA). These agents are
classified into three categories: monitor agents, function
agents, and data agents. Monitor agents mainly include
MAs and backup MAs acting as mediators for the
other agents to ensure that the OPPMMS runs prop-
erly. Function agents include EAs and AAs, which exe-
cute the main functions of the OPPMMS regarding
external instructions, predetermined mission planning,
and consultation with various other agents. Data agents
include IAs and RAs, which provide data services in
response to the demand from other agents and imple-
ment data transmission services between the OPPMMS
and external systems.
The relationship among these agents is illustrated in
Figure 8. These agents should be run at many different
digital terminals and are distributed across different
enterprise departments that are compatible with the
material management process. Additionally, the proto-
type system is equipped with one or more control cen-
ters and distributed task architecture.
In an OPPMMS based on multi-agents, EAs are
deployed on production nodes as required, generating
corresponding MIUs as encoder rules in the knowledge
library and then sending the MIUs to the RAs. The
a-MIU generation process reflects the hierarchical
product structure in the BOM, and the c-MIU genera-
tion process reflects the results following execution of
Figure 8. Agents and agent categories in the proposed
OPPMMS.
the product construction plan. This process manifests
the characteristics of the MIU architecture. The proto-
type OPPMMS framework, which combines business
flow with data flow, is exhibited in Figure 9.
Agent communication
The OPPMMS communication network plays an
important role in an agent-based system. Regarding the
communication protocol, many kinds of agent commu-
nication languages (ACLs) have been proposed for
MASs, such as the knowledge interchange format
(KIF), the knowledge query manipulation language
(KQML), and the ACL by the Foundation for
Intelligent Physical Agents (FIPA-ACL).45 In view of
the FIPA-ACL standards that have been widely and
successfully employed, the FIPA-ACL is selected to
facilitate communication among the agents. Figure 10
illustrates a typical ACL-based agent communication
message. In terms of the system’s component develop-
ment process, extensible markup language (XML) is
used to format the ACL content.
The agent communication mechanism is designed in
Figure 11 and is divided into four layers: communica-
tion object, communication type, communication con-
tent, and agent ontology. In the OPPMMS, agent
communication has been divided into two types: node-
to-node communication for operational instructions
and blackboard communication for task assignment. In
accordance with the data requirements of the agent and
the task description, the agents themselves select the
corresponding communication type. Each immobile
agent in the OPPMMS receives and sends related mes-
sages or mobile agents to a communication object via
its communication interface.
Detailed design of agents
Functional definition of agents. Each agent in the system
has its own specific functionality. All the agents are
organized toward a common goal through the indus-
trial IoT. The functionalities of each agent are as
follows:
Han et al.
Figure 9. OPPMMS framework based on the MAS.
2.
3.
Figure 10. Agent communication message.
1. MA. The MA is at the center of the multiple
agents in this system and is responsible for
interaction management and conflict resolution
among all other agents. The knowledge library
of the MA stores the rules and logic for the
MAS management and conflict resolution
9
schema. In addition, the MA sends and receives
the necessary messages to and from the other
agents to coordinate their operations.
MIU EA. The function of the EA is to perform
MIU encoding work. The MIU will be gener-
ated by the EA in accordance with the rules of
the knowledge library and additional informa-
tion. After MIU generation, the EA will send
the MIU to other agents for further processing.
MIU AA. Generated by an immobile agent, the
AA is designed as a mobile agent in accordance
with the agent transfer protocol. For instance,
when MIU data need to be analyzed, an AA
will be generated and sent to a mobile agent
context for task execution. Once the AA has
been activated in a mobile agent context, it will
analyze the MIU using the MIU analysis algo-
rithm shown in Table 1 and send the results to
the corresponding EAs for further processing.
10
Figure 11. OPPMMS agent communication mechanism.
4. IA. The IA plays an important role in this sys-
tem, being used to connect the MAS with exter-
nal systems. Crucial information, such as the
BOM and OPP construction plan, is sent to the
knowledge library of other agents through the
IA. Furthermore, the IA provides an interactive
channel for system users such as enterprise man-
agers, who can monitor system processes and
intervene as required.
5. RA. The function agents and the IA are con-
nected to all the enterprise databases by the RA,
which stores all the information on materials
during the OPP manufacturing process. After
receiving the MIU data from the EA, the RA
will classify and store them. Moreover, when
MIUs need to be analyzed, the RA generates
AAs and provides the runtime environment for
the AAs.
Agent structure. In view of the belief–desire–intention
(BDI) model, an agent comprises a collection of com-
ponents. Each collection contains a goal (intention), a
controller (desire), and some knowledge (belief) about
its external environment.
Agents in an OPPMMS are divided into two cate-
gories: mobile agents and immobile agents. Immobile
agents are deployed on the production nodes or data
nodes in networks, execute the crucial tasks in an
OPPMMS, and provide the runtime environment for
mobile agents. Mobile agents include AAs, which are
generated by immobile agents to analyze MIUs. The
Advances in Mechanical Engineering
agent structure in an OPPMMS is developed in
Figure 12.
OPPMMS working mechanism
The OPPMMS uses centralized data storage, which is
integrated with the enterprise database system, to inte-
grate with other management systems in an enterprise.
However, a lot of materials and complicated processes
are involved in OPP manufacturing processes, and a
large volume of MIU data are generated at different
locations on the enterprise IoT. The calculation load
involved in MIU analysis is relatively large and needs
to be performed by multiple agents. Therefore, a multi-
agent cooperation mechanism for MIU analysis has
been designed in line with the MAS architecture.
Main operational scope of the OPPMMS
The main operational scope of the OPPMMS is divided
into two parts: MIU data formation and MIU data
analysis. MIU data formation involves the acquisition
of material information from IoT sensors. MIU data
analysis involves the analysis of the MIU data and for-
warding the results to the corresponding EAs for fur-
ther processing.
MIU data formation. As Figure 13 shows, the MIU data
formation process involves the acquisition of material
attributes and the correlation of material information
from the IoT sensors, which is then encoded into
Han et al. 11

Figure 12. OPPMMS agent structure.

Table 2. MIU data formation process.

Step Content

1 The EA senses the sensor information from IoT

2 EA obtains the required BOM file through the IA
3 In view of the sensor information and BOM file, EA
generates MIUs
4 EA sends MIUs to RA
5 According to the MIU type (a-MIU or c-MIU), RA
stores the MIU and forms the MIU data model
6 For the a-MIU, the EA sends a-MIU to the external
system through the IA.
The a-MIU data will be written to the IoT tag of the
material entity
Figure 13. MIU data formation sequence.
EA: encoder agent; IoT: Internet of Things; BOM: bill of materials; IA:
interface agent; MIU: material information unit; RA: resource agent;
a-MIU: attribute-material information unit; c-MIU: correlation-material
information unit.

a-MIU and c-MIU by the EA and sent to the RA for

data storage. This process is presented in Figure 13 and
Table 2 and mainly reflects the sequential cooperation
of the EA, RA, and IA.

MIU data analysis. MIU data analysis is mainly per-

formed by the AAs that are generated by the RAs.
When analysis of MIU data is required, first the MA
specifies the analysis content, and then the RA provides Figure 14. MIU data analysis sequence.
the corresponding MIU data and generates AAs to
analyze the MIU. Considering the results, the corre-
sponding EA will be notified for further processing
(e.g. finding the cause of the data error). This process is Multi-agent cooperation mechanism
designed in Figure 14 and Table 3. The MIU analysis task is executed by the AAs that are
In the MIU analysis process, multi-agents run a generated by the RAs. The MIU acquisition capacity
cooperation mechanism to allocate the tasks in Step 2. and AA running space are limited for a single RA.
This mechanism is described as follows. Therefore, an effective and efficient multi-agent
12 Advances in Mechanical Engineering

Table 3. MIU data analysis process.

Step Content

1 MA gives the target analysis interval of the MIU

2 Considering the target analysis interval of MIU, MA organizes multiple RAs for task allocation
3 and 4 After task determining, RA generates AAs to analyze the MIU data. The analysis method is described in Table 1
5 AA sends analysis results to corresponding EA
6 According to analysis results, EA sends MIU and other required information to the operators through IA
for further process
7 EA terminates corresponding AA

MA: manager agent; MIU: material information unit; RA: resource agent; AA: analyzer agent; EA: encoder agent.

Table 4. Cooperation procedure for agents.

Step Content

1 The MA as the CO releases the CG that is the MIU analysis from (a0, c0) to (a1, c1) via the blackboard
communication type
2 Regarding the response of cooperation agents, several RAs decide whether to participate in task bidding
3 MA collects bid RAs
4 In OPPMMS, the set of evaluation criteria for bid RAs is proposed: C = (c1, c2, c3, c4), where
c1: MIU analysis capability of RA
c2: network response time of RA
c3: task load of RA
c4: reliability of RA
5 MA specifies two backup management agents as evaluation agent, to form a set {e1, e2}
6 MA establishes weighting P coefficients W for the evaluation criteria
W = (w1, w2, w3, w4), and 4i = 1 wi = 1
7 Set rij as the evaluation score of bid agent evaluated by evaluation agent ej, in line with evaluation criterion ci from
C in Step 3. For each bid agent, the first-evaluation matrix R is generated as follows
2 3
r11 r12
6 r21 r22 7
R=6 4 r31 r32 5
7

r41 r42
8 In accordance with W in StepP 6 and R in Step 7,Pthe negotiation evaluation matrix N is generated as follows
N = W
R = (n1, n2), and n1 = 4i = 1 wi ri1 , n2 = 4i = 1 wi ri2
9 In view of the methods of maximum membership degree, for bid agent i, the final estimation FEi is proposed as
follows: FEi = max(n1, n2)
10 According to the final estimation FEi of the bid agent i from high to low, the CO selects RAs to form the CR, until
the sum of RAs analysis capacity is greater than the CG requirements
11 After the identification of CT, the CR will be executed to implement the CG. For MIU analysis task, the execution
process of each agent is described in Table 1

MA: manager agent; CO: cooperation organizer; CG: cooperation goal; MIU: material information unit; RA: resource agent; OPPMMS: offshore
platform project material management system; CR: cooperation result; CT: cooperation team.

collaboration mechanism should be proposed in light of
the evaluation regarding the working criteria of agents
in OPPMMS, aiming to select the targeting agents to
cooperatively make MIU analysis as well as avoid task
conflicts and resource waste. In this regard, fuzzy com-
prehensive evaluation method has the capacity to pro-
vide robust and accurate evaluation corresponding to
multiple criteria, which can be adopted and applied in
MAS to deal with task allocation problem while achieve
a comparatively high efficiency.46 On top of these, to
take advantage of the autonomy of MAS, task alloca-
tion process in OPPMMS can be designed as a bidding
mechanism, guaranteeing an overall optimal and agile
solution. Therefore, the MAS-based cooperation
mechanism is designed as follows to complete the MIU
analysis autonomously and effectively. The cooperation
model is designed as a multi-tuple
Cooperation = (CO, CG, CT, CR)
The cooperation organizer (CO) is an agent that
proposes cooperation objectives and develops coopera-
tion plans. In an OPPMMS, all the immobile agents
can be set as COs. Cooperation goals (CGs), which are
presented by the COs, include cooperation tasks and
task performances. The cooperation team (CT) is set
Han et al. 13

Figure 15. Interface and interaction process of the prototype system.

by the CO in accordance with the requirements of the

cooperation task. The cooperation result (CR) is the
allocation of cooperation tasks in light of the negotia-
tions by the CT.
In accordance with the agents’ cooperation model in
an OPPMMS, the agent cooperation procedure for a
collaborative MIU analysis that is made by the task
characteristics includes an evaluation method and cri-
teria for bid agents that are designed based on fuzzy
comprehensive evaluation.46 Table 4 presents a
hypothetical MIU analysis task example to describe Figure 16. MIU data items.
this process.

Example of MIU data formation

OPPMMS validation
Judging by the characteristics of system user demands
and the above analysis, an OPPMMS prototype that
adopts a browser/server structure has been developed
and implemented using Java Agent Development
Framework (JADE) 4.3.3. This prototype system has
been deployed in an OPP enterprise in China’s
Shandong province to verify the performance. The pro-
totype system interfaces together with data flow and
typical communication messages are illustrated in
Figure 15.
Considering the MIU data formation presented in
Table 2, the OPPMMS implements the functions that
form and store MIUs after processing the material
information acquired by the IoT sensors. Figure 16
shows some items of MIU data that have been exported
from the OPPMMS.
To verify the functions of the MIU data model in
material management, a set of examples is used to simu-
late the application of the MIU:
1. Material information tracing. Suppose that there
are some processing quality problems in material
14
Figure 17. Task execution for MIU data analysis.
with the a-MIU code F03G01. Hence, the related
material records need to be traced. More precisely,
through the OPPMMS, all the MIUs with search
conditions such that the a-MIU code is F03G01 will
be retrieved. In Figure 16, this is represented as
MIU: \MIU = F03G01-F1A150903. in the first
row of the first column. In accordance with the
encoding rules displayed in Figure 4, the decoding
information is obtained as follows:
The material with code F03G01, on date 03/09/15, was
processed through a procedure in which the code of the
worker was F1 and the code of the process was A.
2. Material information tracking. Suppose that there
are operational errors involving material with the c-
MIU code F1A150903. Thus, the material that has
passed through this operation needs to be tracked.
This means that through the OPPMMS, all the
MIUs with search conditions such that the c-MIU
code is F1A150903 will be retrieved. In Figure 16,
this is represented by three MIUs: \MIU =
F03G01-F1A150903., \MIU = F03G02-F1A150
903., and \ MIU = F03G04-F1A150903. in the
Advances in Mechanical Engineering
first row of the first, second, and fourth columns,
respectively. Consequently, material with a-MIU
code F03G01, F03G02, or F03G04 in any product
needs to be located and checked to confirm whether
the operational errors have impacted on other
materials.
Example of MIU data analysis
Here, production data from an OPP enterprise are used
to support the MIU analysis process in our OPPMMS.
The MIU data analysis procedure using cooperation
mechanism parameters is as follows:
1. Obtain the MIU analysis tasks. In this task, the
MIU analysis interval is manifested in Figure 17
and includes 19,960 MIUs.
2. The MA releases the task in the system.
3. The RA learns about the MIU analysis task and
decides whether to participate. There are four
RAs participating in bidding for this task, form-
ing a set of bid agents B: B = {b1, b2, b3, b4}.
4. The MA specifies two backup MAs as evalua-
tion agents {e1, e2}.
Han et al. 15

Table 5. Evaluation process data.

Bid agent b1 Bid agent b2 Bid agent b3 Bid agent b4

ri1 ri2 ri1 ri2 ri1 ri2 ri1 ri2

r1j 0.20 0.80 0.80 0.20 0.50 0.50 0.70 0.30

r2j 0.40 0.60 0.25 0.75 0.35 0.65 0.55 0.45
r3j 0.45 0.55 0.60 0.40 0.45 0.55 0.75 0.25
r4j 0.30 0.70 0.45 0.55 0.55 0.45 0.50 0.50
N 0.3225 0.6775 0.5775 0.4225 0.4550 0.5450 0.6550 0.3450
FEi 0.6775 0.5775 0.5450 0.6550
RA capability 12,000 MIUs 8000 MIUs 6000 MIUs 8000 MIUs

RA: resource agent; MIU: material information unit.

5. In accordance with enterprise expert guidance,
the weighting factors for the evaluation criteria
set C, which is described in Step 4 in Table 4,
are determined to be W = (0.40, 0.25, 0.25,
0.10).
6. In accordance with the evaluation procedure,
which is explained in Steps 5–10 in Table 4, and
the weighting coefficients for evaluation criteria
W, the bid agents b1–b4 are evaluated. The eva-
luation process data are shown in Table 5.
7. According to the final estimation FEi and analy-
sis capability of bid RAs, bid agents b1 and b4,
with a total of more than 20,000 MIU analysis
capabilities that satisfy the task requirements for
19,960 MIUs, are selected to execute the MIU
analysis.
8. Set the analysis radius k = 1 to execute the anal-
ysis. Figure 17 shows the OPPMMS export of
abnormal MIU data after this MIU analysis.
The OPPMMS explains that it can automati-
cally analyze the MIUs in accordance with the
instructions.
Conclusion and future work
This article proposed an intelligent OPPMMS in light
of the IoT and multi-agent technology. The develop-
ment of this OPPMMS is based on agent technology
and newly designed MIUs. The MIU, which comprises
material attribute information and material correlation
information, was designed to meet material trackability
and traceability management requirements. In view of
the MIU design, the MIU-relevant model was imple-
mented as the working mechanism of an agent-based
OPPMMS. Thus, an OPPMMS prototype including
agent definitions and an IoT-based system framework
was constructed. Additionally, the agent structure and
communication framework were presented, and the
multi-agent cooperation mechanism in accordance with
the fuzzy comprehensive evaluation was elaborated.
Finally, the OPPMMS prototype was developed and
tested, which confirms the practicability and effective-
ness of this research. The advantages of our OPPMMS
can be summarized as working automation, informa-
tion accuracy, and functional modularity.
Future work will focus on the improvement of the
OPPMMS and other integrated systems. Furthermore,
an advanced MIU encoding format, analysis and retrie-
val algorithms for MIU data, intelligent analysis of
larger volumes of MIUs and other OPP data, and opti-
mized cooperation mechanisms for multi-agents will be
investigated.
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 sup-
port for the research, authorship, and/or publication of this
article: This article was financially supported by the National
Natural Science Foundation of China (project no. 51679059).
References
1. Sjogren P, Bellgran M, Fagerstrom B, et al. Manufactur-
ing aspects of offshore fabrication and installation. Int J
Marit Eng 2014; 156: 277–284.
2. Speight JG. Chapter 3. Offshore platforms. In: Speight
JG (eds) Subsea and deepwater oil and gas science and
technology, vol. 39. Oxford: Elsevier, 2015, pp.71–106.
3. Ren Z, Anumba CJ and Tah J. RFID-facilitated con-
struction materials management (RFID-CMM)—a case
study of water-supply project. Adv Eng Inform 2011; 25:
198–207.
4. Mousavi A, Sarhadi M, Fawcett S, et al. Tracking and
traceability solution using a novel material handling sys-
tem. Innov Food Sci Emerg 2005; 6: 91–105.
5. Watts FB. Chapter 5. Bill of material (BOM) process. In:
Watts FB (ed.) Configuration management metrics.
Oxford: William Andrew Publishing, 2010, pp.97–118.
6. Jin-Lou Z, Yi-Fei C and Zhao L. Risk assessment of off-
shore engineering equipment projects—a case study of
fall pipe vessel in China. In: 2011 international conference
on management science and engineering (ICMSE), Rome,
13–15 September 2011, pp.994–1002. New York: IEEE.
16
7. Li Z and Liu J. A BOL based material traceability model
in offshore equipment projects. J Harbin Eng Univ 2011;
32: 525–529.
8. Lee SS, Lee JK, Park BJ, et al. Development of Internet-
based ship technical information management system.
Ocean Eng 2006; 33: 1814–1828.
9. Sun C. The study and application of ship material code
based on lean shipbuilding. Nanjing, China: Jiangsu Uni-
versity of Science and Technology, 2012, pp.32–50.
10. Ma XM. The research on material codes in engineering
design. J Petrol Petrochem 2012; 8: 28–32.
11. Jennings NR and Wooldridge M. Applications of intelli-
gent agents. Agent technology. New York: Springer-Ver-
lag, 1998, pp.27–58.
12. Hãkansson A, Hartung R and Nguyen NT. Agent and
multi-agent technology for Internet and enterprise systems.
New York: Springer, 2012.
13. Leitão P, Rodrigues N, Turrin C, et al. Chapter 16.
Multi-agent system for integrating quality and process
control in a home appliance production line. In: Leitão P
and Karnouskos S (eds) Industrial agents. Amsterdam:
Elsevier, 2015, pp.287–300.
14. Mahdavi I, Shirazi B, Ghorbani N, et al. IMAQCS:
design and implementation of an intelligent multi-agent
system for monitoring and controlling quality of cement
production processes. Comput Ind 2013; 64: 290–298.
15. Ostrosi E and Ferney M. Fuzzy agents for product con-
figuration in collaborative and distributed design process.
Appl Soft Comput 2012; 12: 2091–2105.
16. Hao Q, Shen W, Zhang Z, et al. Agent-based collabora-
tive product design engineering: an industrial case study.
Comput Ind 2006; 57: 26–38.
17. Vázquez JCD, Martı́nez AC, Gómez A, et al. Intelligent
agents technology applied to tasks scheduling and com-
munications management in a critical care telemonitoring
system. Comput Biol Med 2007; 37: 760–773.
18. Chouhan SS and Niyogi R. Plan restructuring in multi
agent planning. Procedia Comput Sci 2015; 46: 396–401.
19. Berdun L, Amandi A and Campo M. An agent specific
planning algorithm. Expert Syst Appl 2012; 39: 4860–4873.
20. Groves W, Collins J, Gini M, et al. Agent-assisted supply
chain management: analysis and lessons learned. Decis
Support Syst 2014; 57: 274–284.
21. Seco A and Vieira C. A multi-agent supply chain simula-
tion analysis through a statistical mixed model. Procedia
Technol 2014; 16: 163–171.
22. Mcgreavy C, Wang XZ, Lu ML, et al. Objects, agents
and work flow modelling for concurrent engineering pro-
cess design. Comput Chem Eng 1996; 20: S1167–S1172.
23. Yang YJ, Sung TW, Wu C, et al. An agent-based work-
flow system for enterprise based on FIPA-OS framework.
Expert Syst Appl 2010; 37: 393–400.
24. Huang GQ, Huang J and Mak KL. Agent-based work-
flow management in collaborative product development
on the Internet. Comput Aided Design 2000; 32: 133–144.
25. Lim MK and Zhang DZ. An integrated agent-based
approach for responsive control of manufacturing
resources. Comput Ind Eng 2004; 46: 221–232.
26. Zhou G and Jiang P. Using mobile agents to encapsulate
manufacturing resources over the Internet. Int J Adv
Manuf Tech 2004; 25: 189–197.
Advances in Mechanical Engineering
27. Barbosa J, Leitão P, Adam E, et al. Dynamic self-
organization in holonic multi-agent manufacturing systems:
the ADACOR evolution. Comput Ind 2014; 66: 99–111.
28. Choi KH, Kim DS and Doh YH. Multi-agent-based task
assignment system for virtual enterprises. Robot Comput
Integrated Manuf 2007; 23: 624–629.
29. Sheremetov LB and Smirnov AV. Component integra-
tion framework for manufacturing systems re-engineer-
ing: agent and object approach. Robot Auton Syst 1999;
27: 77–89.
30. Kranakis E, Krizanc D and Marcou E. The mobile agent
rendezvous problem in the ring, vol. 1. Princeton: Morgan
& Claypool Publishers, 2010.
31. Oueslati W, Hamdi H, Dhouioui Z, et al. A mobile agent
view synchronization system to uphold a trajectory data
warehouse. Procedia Comput Sci 2015; 60: 276–283.
32. Biswas PK, Qi H and Xu Y. Mobile-agent-based colla-
borative sensor fusion. Inform Fusion 2008; 9: 399–411.
33. Xu Y and Qi H. Mobile agent migration modeling and
design for target tracking in wireless sensor networks. Ad
Hoc Netw 2008; 6: 1–16.
34. Dong M, Ota K, Yang LT, et al. Mobile agent-based
energy-aware and user-centric data collection in wireless
sensor networks. Comput Netw 2014; 74: 58–70.
35. Gubbi J, Buyya R, Marusic S, et al. Internet of Things
(IoT): a vision, architectural elements, and future direc-
tions. Future Generat Comput Syst 2013; 29: 1645–1660.
36. Lee I and Lee K. The Internet of Things (IoT): applica-
tions, investments, and challenges for enterprises. Bus
Horizons 2015; 58: 431–440.
37. Attaran M. RFID: an enabler of supply chain operations.
Supply Chain Manag 2007; 12: 249–257.
38. Masciari E. RFID data management for effective objects
tracking. In: Proceedings of the 2007 ACM symposium on
applied computing, Seoul, Korea, 11–15 March 2007,
pp.457–461. New York: Association for Computing
Machinery.
39. Myny K, Steudel S, Smout S, et al. Organic RFID trans-
ponder chip with data rate compatible with electronic
product coding. Org Electron 2010; 11: 1176–1179.
40. Qian JP, Yang XT, Wu XM, et al. A traceability system
incorporating 2D barcode and RFID technology for
wheat flour mills. Comput Electron Agr 2012; 89: 76–85.
41. Tarjan L, Šenk I, Tegeltija S, et al. A readability analysis
for QR code application in a traceability system. Comput
Electron Agr 2014; 109: 1–11.
42. Kang YS and Lee YH. Development of generic RFID
traceability services. Comput Ind 2013; 64: 609–623.
43. Velandia DMS, Kaur N, Whittow WG, et al. Towards
industrial internet of things: crankshaft monitoring, tra-
ceability and tracking using RFID. Robot Comput Inte-
grated Manuf 2016; 41: 66–77.
44. Chongwatpol J and Sharda R. RFID-enabled track and
traceability in job-shop scheduling environment. Eur J
Oper Res 2013; 227: 453–463.
45. Boella G, Governatori G, Hulstijn J, et al. Time and
defeasibility in FIPA ACL semantics. J Appl Logic 2008;
9: 634–637.
46. Liu X, Peng G, Liu X, et al. Development of a collabora-
tive virtual maintenance environment with agent technol-
ogy. J Manuf Syst 2010; 29: 173–181.