Research on Ontology Modeling of Steel Manufacturing Process Based on Big Data Analysis

As an important method that steel industries ride the Indutrie 4.0 wave, knowledge management is expected to be versatile, effective and intelligent. Mechanism modeling difficulties, numerous influencing factors and complex industrial chains hinder the development of knowledge and information integration. Using data potentials, big data analysis can be an effective way to deal with knowledge acquisition as it solves the inaccuracy and imperfection mechanism modeling may lead to. This paper proposes a big data knowledge management system(BDAKMS) adhering to data driven, intelligent analysis, service publication, dynamic update princip le which can effectively extracts knowledge from mass data. Then, ontology modeling gives the knowledge unified descriptions as well as inference details combined with semantic web techniques.


Introduction
Indutrie 4.0 was put forward in Hannover Messe and brings prospects to hasten the strategic transformation of steel industries as well.One of the aspects that Industrie 4.0 propels manufacturing revolution is co mprehensive, unified and do main-oriented knowledge collect ion to instruct business operation.Ref. [1] proposes fast matrixbased approaches to knowledge acquisition in covering informat ion systems and reduce the computational t ime compared with classical rough set, especially large-scale data sets.However, this method is limited to static informat ion and discretization is not universal.Ref. [2] builds a generic EUEDE contextual architecture for knowledge acquisition, co mpatible with current DSS.Ontology is used for knowledge formalization, so that the relationship between the knowledge items contribute to the rules for decision making, but EUEDE excludes external factors which are important to decision knowledge extraction.Ref. [3] co mbines knowledge graph with case analysis to min imize the gap among distinct levels of expert ise in access to knowledge.More than the difference or diversity as it has addressed, the inconsistency should not be neglected in mult i-source tacit knowledge.Ontology makes it possible to knowledge interchange and reusability wh ich is essential to knowledge modeling.Manufacturing taxonomy helps ontology modeling and results in ontology based knowledge hierarchy [4].Topdown development approach can satisfy diversified levels of ontology granularity.Lifecycle [5] management is proposed and brings an impetus to ontology based knowledge update.Ontology has not only been applied in the manufacturing process [6] but also product design and cost budgeting [7].Ontology is the core of semantic web.While ontology enriches semantic description, knowledge is expressed more co mprehensively and integrated from distributed systems.Evolved fro m RDF Schema, ontology web language(OWL) has a richer vocabulary to define relat ions for inference [8] and is developed to describe ontologies.Big data are produced in every segment in manufacturing: For enterprise performance, big data help identify the product demands, productivity and performance via diverse business objectives.For production, big data make it possible to detect the right facility causing the product flaw.Currently, steel making industries have applied big data analysis to quality management, ore blending solution optimizat ion, supply chain management, equipment fault diagnosis and financial, etc.For examp le, BAO Steel finds better ore blending schemes with analysis of historical data.Anyang steel industry employ big data analysis to find out the key factors that cause the steel cracks [9].Focusing on the difficulties in knowledge acquisition of steel industries, this paper presents a big data analysis based knowledge management structure (BDAKMS).Ontology is used to model the domain knowledge and finally enhance the usability and interoperability.Th is paper also gives an ontology modeling case based on big data analytical results and achieves accurate inference through SWRL ru les and semantic web technologies.This paper focuses on discussions about ontology modeling methodology and semantic deduction.A converter steel making case is given to explain and verify the techniques' validity.incorporates the valued knowledge with ontology description language.The new knowledge entities combined with the existing have to be inspected by knowledge testing.Knowledge evaluation module assesses the knowledge items periodically to eliminate the outdated or inaccurate knowledge and make a request to big data analysis for update if possible.Moreover, a completed knowledge management system also needs inference engine and workflow engine to circu late the knowledge resources.

Knowledge Service Layer.
In addition that this layer has knowledge retrieval and inference modules like tradit ional do main knowledge base, it enables new knowledge publication interface, service composition according to certain business needs and user analysis requests that act as big data analysis objectives.

Knowledge Acquisi tion Ba sed on Big Data Analysis
Based on big data analysis, knowledge acquisition is the prerequisite for BDA KMS.As knowledge is organized in lifecycle management which indicates that the process fro m data to knowledge is iterative and repetitive.Dynamic external and internal environ ment of enterprises also contribute to the iteration.In general, b ig data analysis layer keeps on extract ing data fro m M ES, ERP, SCA DA, sensors and real-time database to do massive analytical works.The knowledge evaluation module assesses the knowledge quality o f both newfound and existing knowledge items, and rep laces the outdated informat ion with the most appropriate.The knowledge acquisition flo w is shown in Fig 2 and it is summarized as three phases.

Phase1 Data preparation.
Mass data source can be divided into SQL data and NoSQL data(log, web, video and sensor data, etc.) that have been stored in distributed memo ry system or cloud storage infrastructure.The storage system is equipped with fau lt tolerance, load balancing and cluster mechanis m, and it is designed to provide the mult i-source and heterogeneous data with un iform environment.Data ETL (Extract ion, Transform, Load) extracts data across diverse platforms in preparation fo r data integration.For SQL data, ETL tool accesses data easily in SQL query.For NoSQL, ETL is imp lemented in Hadoop for data search.Furthermore, data cleaning improves the data consistency, integrity, usability and accuracy.For instance, tuple neglect or default fill deals with data missing.Clustering or regression method handles noisy data [10].To provide the heterogeneous data access interfaces, data integration unifies the specification of data definition and constraint.

Phase2 Data analysis and result evaluation.
Data analysis methods are not confined to traditional strategies like reports, statistics, and combination models.Instead, advanced analysis methods like predict ion models, data mining, text analysis, page rank, collaborative filtering have been a research hotspot and widely used.Data mining is used to find out predicative, practical and novel results [11], wh ile relation schema and cloud tags visualize the results.However, the analytical results should also match the business process and get merged in knowledge base if they pass the assessment.

Phase3 Knowledge Management.
Qualified knowledge is expressed by concepts, relations, properties, axio ms, rules and instance [12] of ontology.
On the basis of ontology, semantic web not only makes knowledge service-oriented but also offers an excellent management system which improves knowledge inference efficiency and eliminates Once updated.Above all, the goal of knowledge management is to form the KMS and publish knowledge service.

Ontology modeling and application
Results from big data analysis in BDA KMS should be formalized by ontology and enriched with semantic web.A knowledge management system is finally developed to render knowledge services.

Ontology Modeling Example
A case of converter steel making process is presented in this paper.Ontology classes are classified into seven groups: environmental factors, product, power, facility, ingredient, energy and process schedule as Fig 3 shows.isA represents a relat ionship between class and instance.isPartOf represents a parent-child relat ionship among classes.For examp le, limestone is one instance of slag making material, while slag-making material is a subclass of non-metal materials.Lime quality is related to humidity, which is an instance of environmental factor.The ingredients of limestone are CaO, MgO and SiO2, etc. K-means cluster analysis was applied in 500 crude ore blending samples to distinguish the better plans form the invalid.One of the better plans is expressed in isA

Knowledge Inference
SWRL describes the correlat ions of distributed and heterogeneous informat ion so that the interoperability is facilitated [13].Princip le of SWRL inference is shown in   (1) Parameters specification in this formula is explained in Table 2. Fig 6 shows the edited rule and the atoms in the edited SWRL atoms are exp lained as Table 3 shows.Similarly, lime addition, dolo mite demand and dolomite addition can be calculated in other formu las.Pellet, as inference, is adopted and initializes the following instance shown in Table 3.This paper also develops a prototype knowledge base system in B/S pattern to the inference interface and the results of this case are shown in Fig 7.

Summary
This paper proposes a big data analysis adopting the data driven, intelligent analysis, service publication, dynamic update patterns to transform data to knowledge: integrative service.This paper focuses on ontology modeling in BDAKM S and SW RL rule establishment.Finally, this paper gives a converter steel making case for validity verification.However, befo re this structure is utilized, three aspects are in need of improvements: 1.More accurate and effective analytical algorith ms should be developed catering for steel industries' changing needs.2. Big data analysis and should be integrated in the whole BDAKMS, but how to translate the user analysis to big data analysis objective is an issue to be considered.

DOI: 10
.1051/ C Owned by the authors, published by EDP Sciences, 201

Figure 2 .
Figure 2. Knowledge acquisition flow based on big data analysis

Figure 3 .
Figure 3. Ontology structure of converter steel making process

Figure 6 .
Figure 6.SWRL rule editorThen, the inference engine(Jess, Pellet, Racer or Jena, etc.) combines ontologies with ru les, reasoning out related knowledge.This paper takes a converter steel-making as an example to illustrate the SWRL rules establishment.The formulation is:

Figure 7 .
Figure 7. Inference results in knowledge management system

EMS Data Knowledge Evaluation Security Architecture Big Data Physical Architecture Data Integration Tool Data Quality Evaluation Data Definition Specification ETL Tool Mass Data Source Visualized Result Basic Support Layer Data Cleaning Engine Text Data Sensor Data Data Cleaning Rule Network CommunicationProtocal Intelligent Manufacturing Management Theory Workflow Engine Figure
1. Construction of BDAKM S2.1 Construction of BDAKMSBDAKMS adopts data driven, intelligent analysis, service publication, dynamic update pattern to transform data to knowledge: Start ing with primary data, knowledge is obtained by intelligent analysis methods.Knowledge is accessed with calling service encapsulation by users, accomplishing the interchange between the knowledge level and data level.The construction of BDA KMS is shown inFig 1.This structure consists of five layers: Basic Support Layer.The physical platform adopts Hadoop distributed environment with master slaver structure.NameNode is set to task scheduling while DataNode task execution.Traceability is introduced to security architecture to record the data transmission histories and guarantee the data credibility and privacy.the data that DAS reads should match the mapping mode which is a set of mapping rules between meta data and data to be loaded.In fact, data informat ion structure, data mapping relations and customized filter are designed in HBase for data unification.For o xygen gas consumption in EMS, the information storage structure is shown in Table1.Data source IP: port is saved in Row key while column description in Column family.

Table 1 .
HBase storage structure exampleOntology base, rule base, methodology base and model base express formalized knowledge resulted fro m big data analysis.Knowledge testing algorithm will verify the consistency and integrity of the knowledge system.The knowledge is managed in lifecycle method: The knowledge system only The core of big data analysis is to select the appropriate analytical algorithm through programmed Map and Reduce functions.With support tools and middleware like d istributed storage, cloud computing, Linu x cluster, analytical task scheduling, resource dispatch and data fusion, the goal of this layer is to achieve environ mental factor analysis, product quality forecast, manufacturing 04005-p.2ICMM2016 process optimization, machine vulnerability, cost saving and energy optimization, etc.2.1.2Knowledge Management Layer.

Table 2 .
Parameter and its description