www.sciencedirect.com Open in urlscan Pro
162.159.137.70  Public Scan

Form analysis
1 forms found in the DOM

GET /search#submit

<form class="QuickSearch" action="/search#submit" method="get" aria-label="form">
  <div class="search-input">
    <div class="search-input-container search-input-container-no-label"><label class="search-input-label u-hide-visually" for="article-quick-search">Search ScienceDirect</label><input type="search" id="article-quick-search" name="qs" value=""
        class="search-input-field" aria-label="Search ScienceDirect" aria-describedby="article-quick-search-description-message" placeholder="Search ScienceDirect"></div>
    <div class="search-input-message-container">
      <div aria-live="polite" class="search-input-validation-error"></div>
      <div id="article-quick-search-description-message"></div>
    </div>
  </div><button type="submit" class="button small u-margin-xs-left button-primary button-icon-only" aria-disabled="false" aria-label="Submit search"><svg focusable="false" viewBox="0 0 100 128" height="20" width="18.75" class="icon icon-search">
      <path
        d="m19.22 76.91c-5.84-5.84-9.05-13.6-9.05-21.85s3.21-16.01 9.05-21.85c5.84-5.83 13.59-9.05 21.85-9.05 8.25 0 16.01 3.22 21.84 9.05 5.84 5.84 9.05 13.6 9.05 21.85s-3.21 16.01-9.05 21.85c-5.83 5.83-13.59 9.05-21.84 9.05-8.26 0-16.01-3.22-21.85-9.05zm80.33 29.6l-26.32-26.32c5.61-7.15 8.68-15.9 8.68-25.13 0-10.91-4.25-21.17-11.96-28.88-7.72-7.71-17.97-11.96-28.88-11.96s-21.17 4.25-28.88 11.96c-7.72 7.71-11.97 17.97-11.97 28.88s4.25 21.17 11.97 28.88c7.71 7.71 17.97 11.96 28.88 11.96 9.23 0 17.98-3.07 25.13-8.68l26.32 26.32 7.03-7.03">
      </path>
    </svg></button><input type="hidden" name="origin" value="article"><input type="hidden" name="zone" value="qSearch">
</form>

Text Content

JavaScript is disabled on your browser. Please enable JavaScript to use all the
features on this page. Skip to main contentSkip to article
ScienceDirect
 * Journals & Books

 * 
 * Search

RegisterSign in

 * Access through your institution
 * Purchase PDF

Search ScienceDirect



ARTICLE PREVIEW

 * Abstract
 * Introduction
 * Section snippets
 * References (39)
 * Recommended articles (5)


ADVANCED ENGINEERING INFORMATICS

Volume 58, October 2023, 102183


FULL LENGTH ARTICLE
KNOWLEDGE GRAPH MODELING METHOD FOR PRODUCT MANUFACTURING PROCESS BASED ON
HUMAN–CYBER–PHYSICAL FUSION

Author links open overlay panelChen Ding, Fei Qiao, Juan Liu, Dongyuan Wang
Show more
Add to Mendeley
Share
Cite
https://doi.org/10.1016/j.aei.2023.102183Get rights and content




ABSTRACT

The data generated in the product manufacturing process are usually distributed
in different formats, triggering fragmented knowledge and disconnected
information. To address this problem, we present a knowledge graph modeling
method for the product manufacturing process. First, the concepts of
human–cyber–physical (HCP) elements are analyzed in detail. The HCP-related
classes, attributes, and relations are defined in a formalized manner in the
ontology modeling process. Second, a knowledge graph model for the product
manufacturing process (KGM/PMP) is constructed by three steps, including
knowledge extraction, knowledge fusion, and knowledge reasoning. When
constructing the KGM/PMP model, a deep learning method called BERT-D’BiGRU-CRF
is presented to automatically extract knowledge from the manufacturing data.
Moreover, a set of reasoning rules are designed to infer new knowledge. Finally,
a case study is carried out to validate the effectiveness of the proposed
method. The validity of the BERT-D’BiGRU-CRF method on knowledge extraction is
verified by comparing performance with four other methods. The applicability of
the knowledge graph model is demonstrated through developing a prototype system.
With this system, manufacturing knowledge can be provided for the demanders
rapidly and accurately.


INTRODUCTION

Product manufacturing process (PMP) is “an industrial production process that
transforms raw materials into a completed product with certain functions,
shapes, and characteristics” [1]. As a key step in creating product value, PMP’s
fundamental goal is to reduce costs, increase efficiency, and improve quality
through continuous effort [2]. The PMP is mainly divided into two stages, part
manufacturing and assembly process. In these two stages, most processes are
completed in the part manufacturing stage because a product is composed of
multiple parts. PMP contains a series of production activities involved in part
manufacturing, including machining, casting, welding, drilling, etc.
Multifarious data are generated during production activities [3]. Most
manufacturing knowledge resides in the data, including the selection of
processing methods, worker arrangements, and equipment selection. However, data
are dispersed in manufacturing systems with different patterns. Data
heterogeneity leads to the fragmentation of knowledge, making it difficult to
accumulate knowledge in the manufacturing process. Thus, developing a unified
model that can consolidate fragmented knowledge from data is essential.

Generally, most data exist as text, such as engineering documents, process
manuals, and operating logs. This type of data is characterized by multi-source,
unstructured, and low-semantic levels [4]. A wealth of semantic knowledge can be
mined from textual data using semantic mining technology. Semantic knowledge is
the combination of word meanings and contextual information that semantically
represents the relationships between different concepts [5]–[6]. Ontology is one
of the most popular methods for modeling semantic knowledge because of its clear
and explicit explanations of domain concepts [7]. It defines the knowledge in a
machine-understandable way, which is realized by applying the formal terms.
Based on the structural form of knowledge, the reasoning process can be
conducted by the semantic web rule language (SWRL) rules to infer new knowledge
[8]. Because an ontology can characterize structured and fine-grained semantic
information in the manufacturing field, it has been broadly applied to multiple
aspects of the manufacturing lifecycle [9], [10], [11].

Based on this ontology, the knowledge graph presented by Google is the most
promising. It uses an ontology as a framework to formalize classes, attributes,
and relations. The semantic knowledge is organized as “entity-relation-entity”
in the knowledge graph [12]. As a semantic network, knowledge graph can model
knowledge in a semantically directed graph structure to describe concepts,
entities, attributes, and relations symbolically. Furthermore, the dependency,
causality, and physical connection strength of different knowledge are also
contained in the knowledge graph, which gives the knowledge graph powerful
modeling flexibility and expression capabilities [13]. Because of these
advantages, knowledge graphs show great potential for knowledge visualization
[14], knowledge recommendation [15], question answering [16], and knowledge
retrieval [17]. Currently, knowledge graphs are increasingly applied in various
fields, such as hazardous chemical management [18], mineral resource management
[19], railway risk assessment [20], etc. In the manufacturing field, scholars
have extended knowledge graph to assembly process [21], resource allocation
[22], product quality assurance [23], surface defect recognition [24], and
product design [25]. The application of knowledge graph in manufacturing is
still ongoing. According to [26], the application of knowledge graphs in
manufacturing is a promising research direction. To the best of our knowledge,
only a few studies have focused on the use of knowledge graph in PMP.

Because text data in PMP contain a large amount of interconnected information,
the challenge of applying a knowledge graph in PMP lies in two aspects. One is
the design of a proper ontology framework to formalize the important concepts
involved in the PMP. The other is how to effectively mine semantic knowledge in
relevant data, that is, knowledge extraction. When designing an ontology
framework, common concepts must be identified, and then the classes, relations,
and attributes can be defined based on the identified concepts. In the PMP,
production elements are classified into three types: human, cyber, and physical
from the perspective of human-cyber-physical system [27]. Humans are the master
of system to perform decision-making, control, and operation. Cyber system is
the core of the information flow of manufacturing activities. Information
storage, exchange, and sharing is an important part of cyber system. Physical
system is the execution body to complete the working tasks. Each role is
generally recorded as text data. Thus, the important concepts include human,
cyber, and physical-related concepts. However, the concept of humans is
typically ignored or modeled in a coarse-grained manner in conventional
ontology-based modeling methods in the manufacturing field. Therefore, this
study attempts to construct a knowledge graph model for the product
manufacturing process (KGM/PMP) to establish connections between human, cyber,
and physical elements at a semantic level.

Another critical part of constructing a knowledge graph is knowledge extraction,
that is, the identification of entities and relations from documents. Mainstream
methods for recognizing entities and relations can be classified into pipeline-
and joint-based methods [28]. The former treats entity recognition and relation
extraction as independent subtasks. That is, relation extraction was performed
after the entity recognition task. The latter refers to entity recognition and
relation extraction tasks that are performed simultaneously using data.
Pipeline-based methods isolate the interaction and interconnection between
entity recognition and relation extraction and lack information sharing, thus
causing error accumulation. However, joint-based methods can overcome the
problems in pipeline-based methods because entity recognition and relation
extraction are processed in the same model. This enables information sharing and
interactions between the two phases of entity recognition and relation
extraction. Moreover, owing to the superior performance of deep learning in
natural language processing, most joint-based methods have been developed using
deep learning to complete knowledge extraction tasks. Thus, this study presents
a deep-learning-powered method to perform entity recognition and relation
extraction tasks jointly.

The motivations for this study are as follows. In PMP, the use of knowledge is
necessary to perform production tasks effectively. However, some manufacturing
processes depend heavily on the skills of experts, inevitably resulting in an
intellectual burden [29]. To address this problem, developing a knowledge
modeling method that fully uses manufacturing knowledge from text data is
essential. Thus, knowledge can be rapidly applied to meet real-world demands. In
addition, from the perspective of model construction, a knowledge model must be
established without significant time or cost. The proposed knowledge graph model
is built in this manner because the deep-learning-powered method can
automatically extract knowledge from data. In the knowledge graph model,
knowledge is represented in a unified and intuitive manner and reasoning rule
can be designed to infer new knowledge to enhance the model’s usability.

Compared with existing studies, the main contributions of this study are
summarized below.

 * 1)
   
   A knowledge graph is applied in the PMP to link the corresponding knowledge
   existed in the multi-source and unstructured text data. A knowledge network
   full of connections is built to realize human–cyber–physical (HCP) fusion at
   a semantic level.

 * 2)
   
   A deep-learning-powered method called BERT-D’BiGRU-CRF was developed to
   automatically extract the manufacturing knowledge of the PMP, thus reducing
   the cost and effort required to build the knowledge model.

 * 3)
   
   A prototype system for the knowledge graph model was developed based on the
   browser/server (B/S) pattern, which can dynamically exhibit knowledge in a
   visualized interface. Thus, knowledge can be rapidly provided to demanders by
   browsing the webpage of the prototype system.



The remainder of this paper is organized as follows. In Section 2, the proposed
knowledge graph modeling method for the PMP is briefly described. In Section 3,
an ontology knowledge modeling process based on HCP fusion is presented. Section
4 presents the knowledge graph model of the product manufacturing process.
Section 5 presents a case study to verify the effectiveness of the proposed
method. Finally, Section 6 concludes the study.


SECTION SNIPPETS


THREE DIMENSIONS OF THE KGM/PMP MODEL

To adapt the proposed knowledge graph model to the different production
activities of PMP, it is necessary to analyze the intrinsic interconnections
between different production elements from manufacturing text data. Common
concepts in the data should be extracted and organized as a tree-like structure
to generate a knowledge graph model. Hence, this study dissects the model in
three dimensions, forming a three-dimensional knowledge space. This was
conducive to improving the model’s accuracy,


MODELING PREPARATION

As no available ontology can be reused, a new ontology framework for PMP must be
designed. The first step was to determine the scope of the model. The modeling
scope was limited to the manufacturing process of discrete manufacturing
workshops to satisfy actual demands. Subsequently, text data with different
patterns, such as operating logs, process manuals, and engineering documents,
were collected from practical manufacturing scenarios. Thus, the common concepts
associated with the HCP


CONSTRUCTION OF KNOWLEDGE GRAPH MODEL FOR PMP

The main steps for constructing the KGM/PMP are shown in Fig. 5. The first step
is to transform the data with different patterns into standard-format data that
can be used as input for the deep-learning-powered method. Manufacturing
knowledge is then automatically extracted from the text by the BERT-D’BiGRU-CRF.
After knowledge extraction, redundant information and ambiguous entities are
deleted, and pieces of triples are consolidated to realize knowledge fusion.
Because implicit relations


CASE STUDY

To construct the KGM/PMP model, BERT-D’BiGRU-CRF was used to automatically
extract semantic knowledge from manufacturing data. In the knowledge extraction
step, the effectiveness of the BERT-D’BiGRU-CRF method has a direct effect on
the quality of the extracted knowledge. To validate the robustness of
BERT-D’BiGRU-CRF, two datasets were used in this study, i.e., the PMP dataset
and a public resume dataset. In addition to the PMP dataset for constructing the
KGM/PMP model, a public resume


CONCLUSIONS

We proposed a knowledge graph modeling method oriented toward product
manufacturing. In the modeling process, the concepts of the HCP elements were
analyzed in detail, and the classes, attributes, and relations were defined. In
addition, a deep-learning-powered method called BERT-D’BiGRU-CRF was presented
to extract knowledge from the text data of the PMP automatically. A set of
reasoning rules is designed to map the feature information to the manufacturing
knowledge, including the processing


CREDIT AUTHORSHIP CONTRIBUTION STATEMENT

Chen Ding: Methodology, Conceptualization, Software, Writing – original draft,
Formal analysis, Resources, Writing – review & editing. Fei Qiao: Funding
acquisition, Project administration, Formal analysis. Juan Liu: Formal analysis,
Writing – review & editing. Dongyuan Wang: Formal analysis, Writing – review &
editing.


DECLARATION OF COMPETING INTEREST

The authors declare that they have no known competing financial interests or
personal relationships that could have appeared to influence the work reported
in this paper.


ACKNOWLEDGMENTS

This work was supported by the National Natural Science Foundation of China
under Grants 62133011, 61973237, and 62273260.




REFERENCES (39)

 * J. Zhou et al.
   
   
   HUMAN-CYBER-PHYSICAL SYSTEMS (HCPSS) IN THE CONTEXT OF NEW-GENERATION
   INTELLIGENT MANUFACTURING
   
   
   ENGINEERING
   
   (2019)
 * Y. He et al.
   
   
   AN ONTOLOGY-BASED METHOD OF KNOWLEDGE MODELLING FOR REMANUFACTURING PROCESS
   PLANNING
   
   
   J. CLEAN. PROD.
   
   (2020)
 * Z. Wu et al.
   
   
   SEMANTIC HYPER-GRAPH-BASED KNOWLEDGE REPRESENTATION ARCHITECTURE FOR COMPLEX
   PRODUCT DEVELOPMENT
   
   
   COMPUT. IND.
   
   (2018)
 * M. Alkahtani et al.
   
   
   A DECISION SUPPORT SYSTEM BASED ON ONTOLOGY AND DATA MINING TO IMPROVE DESIGN
   USING WARRANTY DATA
   
   
   COMPUT. IND. ENG.
   
   (2019)
 * D. Li et al.
   
   
   A SEMANTIC-LEVEL COMPONENT-BASED SCHEDULING METHOD FOR CUSTOMIZED
   MANUFACTURING
   
   
   ROB. COMPUT. INTEGR. MANUF.
   
   (2021)
 * S. Zhu et al.
   
   
   KNOWLEDGE-BASED QUESTION ANSWERING BY TREE-TO-SEQUENCE LEARNING
   
   
   NEUROCOMPUTING.
   
   (2020)
 * X. Zhang et al.
   
   
   ONTOLOGY-BASED SEMANTIC RETRIEVAL FOR ENGINEERING DOMAIN KNOWLEDGE
   
   
   NEUROCOMPUTING
   
   (2013)
 * X. Zheng et al.
   
   
   A KNOWLEDGE GRAPH METHOD FOR HAZARDOUS CHEMICAL MANAGEMENT: ONTOLOGY DESIGN
   AND ENTITY IDENTIFICATION
   
   
   NEUROCOMPUTING.
   
   (2021)
 * Y. Yu et al.
   
   
   CHINESE MINERAL NAMED ENTITY RECOGNITION BASED ON BERT MODEL
   
   
   EXPERT SYST. APPL.
   
   (2022)
 * C. Liu et al.
   
   
   USING TEXT MINING TO ESTABLISH KNOWLEDGE GRAPH FROM ACCIDENT/INCIDENT REPORTS
   IN RISK ASSESSMENT
   
   
   EXPERT SYST. APPL.
   
   (2022)

B. Zhou et al.


A NOVEL KNOWLEDGE GRAPH-BASED OPTIMIZATION APPROACH FOR RESOURCE ALLOCATION IN
DISCRETE MANUFACTURING WORKSHOPS


ROB. COMPUT. INTEGR. MANUF.

(2021)
S. Kwon et al.


ENRICHING STANDARDS-BASED DIGITAL THREAD BY FUSING AS-DESIGNED AND AS-INSPECTED
DATA USING KNOWLEDGE GRAPHS


ADV. ENG. INF.

(2020)
Z. Li et al.


ZERO-SHOT SURFACE DEFECT RECOGNITION WITH CLASS KNOWLEDGE GRAPH


ADV. ENG. INF.

(2022)
A.G. Bharadwaj et al.


KNOWLEDGE GRAPH CONSTRUCTION FOR PRODUCT DESIGNS FROM LARGE CAD MODEL
REPOSITORIES


ADV. ENG. INF.

(2022)
B. Wang et al.


TOWARD HUMAN-CENTRIC SMART MANUFACTURING: A HUMAN-CYBER-PHYSICAL SYSTEMS (HCPS)
PERSPECTIVE


J. MANUF. SYST.

(2022)
H. Han et al.


CONSTRUCTION AND EVOLUTION OF FAULT DIAGNOSIS KNOWLEDGE GRAPH IN INDUSTRIAL
PROCESS


IEEE TRANS INSTRUM MEAS.

(2022)
L. Guo et al.


AN AUTOMATIC METHOD FOR CONSTRUCTING MACHINING PROCESS KNOWLEDGE BASE FROM
KNOWLEDGE GRAPH


ROB. COMPUT. INTEGR. MANUF.

(2022)
E. Behzad et al.


THE EVOLUTION AND FUTURE OF MANUFACTURING: A REVIEW


J. MANUF. SYST.

(2016)
F. Ameri et al.


INDUSTRIAL ONTOLOGIES FOR INTEROPERABILITY IN AGILE AND RESILIENT MANUFACTURING


INT. J. PROD. RES.

(2022)
View more references


CITED BY (0)




RECOMMENDED ARTICLES (5)

 * Research article
   
   
   MODULAR FAULT DIAGNOSIS IN DISCRETE-EVENT SYSTEMS WITH A CPN DIAGNOSER
   
   IFAC-PapersOnLine, Volume 48, Issue 21, 2015, pp. 470-475
   Show abstract
   
   This paper addresses the problem of fault diagnosis in discrete-event system.
   The system under investigation is modelled as a labelled Petri net. We first
   propose an equivalent encoding of the classical diagnoser with the help of a
   CPN diagnoser. We then apply this encoding to define a modular CPN diagnoser.

 * Research article
   
   
   A CROWDSOURCED CO-MODALITY TRANSPORTATION SYSTEM INTEGRATING PASSENGER AND
   FREIGHT
   
   Advanced Engineering Informatics, Volume 58, 2023, Article 102142
   Show abstract
   
   Co-modality transportation aims to integrate passenger and freight flows into
   a single system by sharing public transport between passengers and freight.
   This paper proposes a crowdsourced co-modality transportation system (CCTS)
   for enhancing the practical implementation of co-modality transportation in
   urban areas. Firstly, empirical investigations are conducted, including
   semi-structured interviews with industry practitioners and online surveys
   with passengers, to collect system design considerations. Secondly, an
   overview of the crowdsourced co-modality management platform (CCMP) is
   presented by highlighting three key components: co-modality services,
   information infrastructure, and decision support system. Thirdly, we
   formulate the parcel assignment problem in the CCTS and examine the factors
   that influence the performance of the CCTS. The empirical investigations
   suggest that crowdsourced passengers care about the potential risks and the
   degree of operation difficulty in crowdsourced activities but do not care
   about privacy issues. In addition, the experimental studies show that
   crowdsourced co-modality is more feasible and promising if more passengers
   are willing to provide crowdsourcing services.

 * Research article
   
   
   DESIGN OF CONTROL SEQUENCES FOR TIMED PETRI NETS BASED ON TREE ENCODING
   
   IFAC-PapersOnLine, Volume 51, Issue 7, 2018, pp. 218-223
   Show abstract
   
   This paper concerns the design of optimal control sequences for timed Petri
   nets under earliest firing policy. Optimality is defined with respect to the
   sequences duration. A method inspired by model predictive control combined
   with a partial exploration of the reachability graphs of the logical net
   system underlying the timed PN model is proposed. The main contribution is to
   provide, under some assumptions, a suboptimal solution based on an
   approximation of the minimal duration of a feasible firing sequence when only
   its firing count vector is assumed to be known. This approximation is given
   as an interval to which the minimal duration necessarily belongs. For that
   purpose, a systematic tree encoding of the net structure is proposed.

 * Research article
   
   
   NEAR-OPTIMAL CONTROL OF NONLINEAR SYSTEMS WITH SIMULTANEOUS CONTROLLED AND
   RANDOM SWITCHES
   
   IFAC-PapersOnLine, Volume 52, Issue 11, 2019, pp. 268-273
   Show abstract
   
   We consider dual switched systems, in which two switching signals act
   simultaneously to select the dynamical mode. The first signal is controlled
   and the second is random, with probabilities that evolve either periodically
   or as a function of the dwell time. We formalize both cases as Markov
   decision processes, which allows them to be solved with a simple approximate
   dynamic programming algorithm. We illustrate the framework in a problem where
   the random signal is a delay on the control channel that is used to send the
   controlled signal to the system.

 * Research article
   
   
   ROBUST NIGHT FLOW ANALYSIS IN WATER DISTRIBUTION NETWORKS: A BILSTM DEEP
   AUTOENCODER APPROACH
   
   Advanced Engineering Informatics, Volume 58, 2023, Article 102135
   Show abstract
   
   Night flow analysis is predominantly used to identify incipient (gradual)
   leakages in real-life Water Distribution Networks (WDNs). However, due to
   extreme stochastic demand especially in residentially dominated district
   metering areas (DMAs) and the emergence of sleepless cities, traditional
   night flow analysis methods have become inefficient. This study proposes a
   semi-supervised sequence-to-sequence Bidirectional Long Short-Term Memory
   (BiLSTM) deep Auto Encoder (AE) method for night flow analysis in WDNs. To
   enhance the generalisation power of the deep AE, a vigorous data augmentation
   procedure based engineering domain knowledge and inherent properties of DMA
   is presented. The proposed method learns the underlying benign patterns in
   night flow (2:00 am to 4:00 am) and identifies significant deviations that
   represent the development of incipient leakages. The method was validated on
   residential, commercial, and industrial DMAs in a real-life WDN in the
   Ålesund Municipality of Norway. The results from the study showed that the
   proposed method is a robust and superior alternative to traditional night
   flow under extreme stochastic consumer demand. The proposed BiLSTM deep AE
   method was able to identify unreported leakages in the DMAs within 1 day or
   at most 4 days compared with at least 15 days identification time by
   traditional minimum night flow (MNF) and average night flow (ANF) analyses.
   Additionally, traditional minimum night flow analysis also produced more
   false positives compared with the proposed method. The results also revealed
   the usage of point estimates to represent night flow as done in MNF and ANF
   analyses fail to capture all salient information regarding night flow
   compared to the BiLSTM deep AE.


View full text
© 2023 Elsevier Ltd. All rights reserved.
 * About ScienceDirect
 * Remote access
 * Shopping cart
 * Advertise
 * Contact and support
 * Terms and conditions
 * Privacy policy

We use cookies to help provide and enhance our service and tailor content and
ads. By continuing you agree to the use of cookies.

All content on this site: Copyright © 2023 Elsevier B.V., its licensors, and
contributors. All rights are reserved, including those for text and data mining,
AI training, and similar technologies. For all open access content, the Creative
Commons licensing terms apply.