Weimin Zhong*,Shuming Liu,Feng Wan,Zhi Li

Key Laboratory of Advanced Control and Optimization for Chemical Processed of Ministry of Education,East China University of Science and Technology,Shanghai 200237,China

Keywords:Equipment selection Ontology technology Knowledge base system Styrene process

ABSTRACT Equipment selection for industrial process usually requires the extensive participation of industrial experts and technologists,which causes a serious waste of resources.This work presents an equipment selection knowledge base system for industrial styrene process(S-ESKBS)based on the ontology technology.This structure includes a low-level knowledge base and a top-level interactive application.As the core part of the S-ESKBS,the low-level knowledge base consists of the equipment selection ontology library,equipment selection rule set and Pellet inference engine.The top-level interactive application is implemented using S-ESKBS,including the parsing storage layer,inference query layer and client application layer.Case studies for the industrial styrene process equipment selection of an analytical column and an alkylation reactor are demonstrated to show the characteristics and implementability of the S-ESKBS.

1.Introduction

With the continuous development of intelligent information technology,the smart factory has received extensive attention.Smart factory uses the Internet of things technologies for highly integrated manufacturing,and employs the intelligent systems to simulate the experts to analyze,infer,judge,think and make decisions for the practical problems of industrial process[1].Purpose of Smart factory is to change the way of human-centered problem-solving in traditional industrial processes and transforming human intelligent activities to information systems' intelligent activities[2].

Knowledge base system is an important part of smart factory,also known as artificial intelligence database system[3].As a combination of artificial intelligence technology and traditional database technology,the knowledge-based system uses knowledge representation to save,infer and reuse the domain-related theoretical knowledge,factual data,and heuristic knowledge effectively[4].

Ontology is widely used to build domain models in the development of knowledge base system,since it provides the basic vocabularies needed for modeling and explains the relationship among them[5–7].As ontology model can provide powerful knowledge inference,good knowledge expression and convenient knowledge acquisition ability[8],it has become the research focus of knowledge base system that is based on knowledge discovery[9–11].

Styrene is an important organic chemical product and the raw material of petrochemical industry[12].However,the process design for manufacturing styrene product is very complex,especially in the equipment selection stage.To choose appropriate equipment for styrene process,participation or consultation of the industrial experts and technologists is generally required.It is important to store and reuse the knowledge,otherwise amount of human,material and financial resources will be wasted.

In this work,we develop an equipment selection knowledge base system to assist the design of styrene process for quick and accurate equipment selection.The equipment selection ontology library is built using the ontology description language Ontology Web Language(OWL)with the help of the ontology construction tool Protégé.The equipment selection rule set is established using the language Semantic Web Rule Language(SWRL)based on the equipment selection ontology library.

2.Knowledge-based System and Structure of S-ESKBS

2.1.The knowledge base system

Knowledge base system is developed for solving a specific problem,which uses the specific knowledge representation to express the information in the field as knowledge,and conducts the consistency verification and implicit knowledge inference according to the corresponding rules[2,13].Therefore,the knowledge base system consists of two parts:knowledge base and inference application mechanism[14].

Knowledge base is the basis of knowledge expression and knowledge inference,which consists of knowledge and rules.Usually the ontology description language OWL is utilized to implement the knowledge expression[15,16],and the rule description language SWRL is used to implement the rule description[17],and the Pellet inference engine is employed to implement the knowledge inference and the consistency test[3].

Inference application mechanism deals with the utilization of knowledge base.Firstly,it carries out corresponding semantic parsing of client's query instructions and transforms them into machine instructions.Then,it combines with the machine instructions to call the relevant inference query engine to implement inference query for knowledge and rules in the knowledge base.Finally,the inference results are returned to the client interface.

Knowledge base system which consists of knowledge base and inference application mechanism can effectively realize the semantic parsing and implicit knowledge inference.On this basis,the equipment selection knowledge base system for styrene process is constructed in this paper.

2.2.The structure of S-ESKBS

The S-ESKBS is established according to the actual requirements of industrial styrene process,and the conventional client/server software system architecture is adopted.The structure of S-ESKBS is shown in Fig.1.

This system consists of the low-level knowledge base and the toplevel interactive application.The low-level knowledge base is the core part of the S-ESKBS,which is used to define and form the equipment selection ontology model as well as the equipment selection rule set.The Pellet inference engine implements the knowledge inference and the consistency test for the equipment selection ontology library and the equipment selection rule set.The top-level interactive application is the implementation of the S-ESKBS,which is used to parse the low level knowledge base and carry out inference query of parsed data.The results of query are returned to the client interface finally.Thus,the top-level interaction application includes the parsing storage layer,the inference query layer,and the client application layer.

3.Low-Level Knowledge Base of S-ESKBS

3.1.The equipment selection ontology library

The equipment selection ontology library whose objective is to build the knowledge base system for styrene process is constructed based on seven-step domain ontology method[18,19].The design stage of the ontology can be divided into three steps:consulting the experts and technologists of industrial styrene process,obtaining the core concepts of the styrene process,and determining the logical relationship among these concepts.

The core concepts used to establish the equipment selection ontology library are obtained by the preparedness and from consulting industrial experts.Then,they are sorted to build the core concept term set.After obtaining the core concepts of styrene process,the class hierarchy,attributes,attribute facets and instances between different classes of equipment selection ontology library are created.The ontology library of S-ESKBS can be divided into three main categories:the styrene equipment,the styrene production process and the industrial styrene process.Each main class can be subdivided into several sub-categories.The schematic is shown in Fig.2.

Another important step in building ontologies is to define the different object attributes and data attributes.The object attributes are used to represent the relationship between two classes,and the data attributes are used to describe the specific values of the instance data.There are total 18 object attributes and 32 data attributes in the equipment selection ontology library in this work,including reaction,production process,temperature,device type and so on.

An instance can be created by generating a concrete object of the ontology class that describes the detailed implementation of the ontology class.It is based on the classes and attributes in the equipment selection ontology library.Table 1 shows the example of Alkylation reactor R101 which can illustrate the case implementation.

Fig.1.Structure of the equipment selection knowledge base system.

Fig.2.Classes and class hierarchies of S-ESKBS's ontology library.

Table 1 Device instance R101

3.2.The equipment selection rule set

The role of the equipment selection rule set is to expand and refine the equipment selection ontology library.The Pellet Inference Engine is utilized to compensate for the shortcomings of the reasoning ability caused by the description logic.SWRL language[20]is used as the rule description language in this work and it combines the OWL Lite,the OWL DL and the Rule ML language to compensate for deficiencies of OWL.

SWRL is mainly composed of Imp,Atom,Variable and Built-in.Among them,Imp constitutes the rules of SWRL;Atom is the body(reasoning premise)and head(reasoning results)of the basic components;Variable is used to record the variables in Atom;Build-in is a module for logical operations in SWRL.As the class hierarchy of equipment selection ontology library has defined Tower,Packed Tower,Plate Tower,Yes,No,Weight,etc.The attributes and attribute facets has defined has Blister Mixture,has Corrosive Mediu,has BigLiquid Load, has Big Viscosity Mixture, has Suspended Materials, has Liquid-gas Ratio Fluctuation Adaptability,,etc.The instances have defined Tower,Packed Tower,Plate Tower,Yes,No,Weight,etc.The SWRL description of equipment selection rules can apply the above classes,attributes,and instances.For example,the SWRL rule of packed tower selection is as follows:

Tower(?d),hasBigDiameter(?d,False),hasCorrosiveMaterial(?d,True),hasFoulLiquid(?d,False),hasVacuumOperation(?d,True),hasTwoCo_existingLiquidPhases(?d,True),hasOperatingFlexibility(?d,False)-＞PackedColumn(?d)

The body of this SWRL rule means:device d is a tower equipment,with a small tower diameter,containing corrosive media,with turbid liquid,with operational flexibility,including gas–liquid two phases,with vacuum operation.The head of this SWRL rule is:device d is packing tower.

There are two ways to build an equipment selection rule set by SWRL.One is to write SWRL rules directly into the OWL file;the other is to write SWRL rules into the Rule Tab of Protégé.In this work,the second approach is adopted.The rule set of this example contains a total of 43 rules,parts of which are described in Table 2.

Table 2 S-ESKBS rule set

3.3.The inference engine

The inference for the low-level knowledge base requires Inference Engines to improve and expand the proposed S-ESKBS ontology library.It has two main functions:on one hand,the equipment selection ontology library is tested for consistency to make sure there is no logical error between two classes,or between class and instancein the S-ESKBS ontology library;on the other hand,it is to find hidden information from the S-ESKBS through the reasoning of the implied knowledge.

Table 3 Rule 23,24

Pellet is an inference engine based on the Tableau description logic algorithm developed for the OWL-DL language,which is written in Java language.Pellet supports all the features of the OWL-DL including XML data types and enumeration types.At the same time,Pellet also supports the DIG interface protocol.With the above advantages,the Pellet Inference Engine is used for the inference of the ontology library of the low-level knowledge base.

The Protégé5.0 built-in Pellet Inference Engine is adopted here to verify the low-level knowledge base.It combines the equipment selection ontology library and the equipment selection rule set.The result of the inference shows that there is no logical error in the low-level knowledge base,and the consistency test is passed in this work.For example,the rules 23 and 24 of the equipment selection rule set are shown in Table 3.The rules and the Pellet Inference Engine are combined to derive the low-level knowledge base and the reasoning results are displayed in Fig.3.The Pellet Inference Engine is applied to complete the implied knowledge discovery,and further improve and expand the equipment selection ontology library.Finally,the low-level knowledge base is constructed.

4.Top-level Interactive Application of S-ESKBS

4.1.The parsing storage layer

This layer is the fundamental part of top-level interactive application with the function of parsing and storage of the low-level knowledge base.Since the ontology library and the rule set which are built in Protégé are not conducive to query and inference at the application side,they need to be stored in the database for inquiry and inference.

There are many development tools that can help ontology developers to parse the ontology.The most widely used one is Jena API[21],which provides many classes and methods for parsing the data contained in the ontology.It also has a good reliability and stability for database support.In this work,Jean API is employed in the parsing storage layer to parse the low-level knowledge base,and the parsing results are stored in the data storage module which uses MySQL database.

Two Jena API toolkits called com.hp.hpl.jena.rdf.model toolkit and com.hp.hpl.jena.ontology toolkit are used.The first toolkit provides an abstract class and method for parsing the resource description framework language,which can obtain the semantic information contained in the building ontology.The second toolkit provides an abstract class and method for accessing and manipulating ontology semantic files such as OWL,RDF,DAML+OIL for the analysis of the low-level knowledge base in the Eclipse environment through the Java programming,and making the analysis results which are stored in the database.

4.2.The inference query layer

The inference query layer is the core of the top-level interactive application,and it is used to inference the query of the low-level knowledge base.This layer consists of the SparQL query module and the Jena inference module.

The results of parsing are stored in the form of an RDF triplet.The data of this format can only be manipulated by RDF triad format query instruction.The implementation process of SparQL query module is shown in Fig.4[22].Firstly,the SparQL query module calls the parser to parse the query instructions which are entered by the client into RDF triplet format.Secondly,the Jena inference module is called by the query engine to infer the query.Finally,the module returns the inference results to the client.

Fig.3.Reasoning results.

Fig.4.The flow chart of SparQL query module.

When the query instruction is executing,the query engine of the SparQL query module calls the Jena inference module to carry out the inference.The process of the Jena inference module is shown in Fig.5[23].Firstly,the query instructions that are parsed as RDF triples are read.The Inference Engine Registrar creates an inference engine based on the information contained in the ontology described by this RDF triplet,and the associated rule base stored in the database.Secondly,the inference engine is bound to the ontology that needs to query the reasoning to generate the model object for retrieval.Finally,the generated model object is manipulated by the ontology and model API,and the implicit relation between the concepts is deduced,and then the query results are returned to the SparQL query module to complete the reasoning query.

4.3.The client application layer

The client application layer is the implementation of the top-level interactive application.This layer involves the front-end interaction which uses JavaWeb three-tier architecture and Model-view-controller(MVC)design patterns to achieve a better S-ESKBS.The specific implementation is illustrated in Fig.6.

The JavaWeb three-tier architecture consists of presentation layer,business logic layer,and data access layer[24].The presentation layer is used to receive the client's request and return the response data.The business logic layer is used to complete the corresponding operation of the inference query layerby combining JavaBean.The data access layer is used to store and read the database from the parsing storage layer,and provides data services for the business logic layer and the presentation layer.

MVC is a software design pattern for implementing user interfaces on computers,which consists of a controller,a model,and a view.The software design pattern enables high cohesion and low coupling by separating business logic,data,and interface displays.

The function of the controller is to accept user requests,invoke model responses,and select views to display response results.The objective of the model is to encapsulate the application state,deal with the business process,and notify the view business status updates.The role of the view is to send the user input to the controller,allowing the controller to view the selection and accept the data update request.

When the client application layer runs,it first sends an HTTP request from the client to the servlet in the presentation layer.Servlet has two ways to deal with the accepted request.One is to redirect directly to the client to return to a page,and the other way is to apply the business logic layer of the model and database interaction to complete the data processing,and then return the results to the Jsp which is the response to the client.

5.The Application of S-ESKBS

Industrial styrene process mainly consists of alkylation unit,ethylbenzene distillation unit,ethylbenzene dehydrogenation unit and styrene distillation unit.In this paper,a software of S-ESKBS for the industrial styrene process is developed,and the interface is shown in Fig.7.The alkylation reactor in alkylation unit is the most import equipment of the industrial styrene process,and the analytical column in dehydrogenation unit is another very typical equipment besides the reactor.In this paper,two typical cases of the selection of these two pieces of equipment are shown to verify the accuracy of the S-ESKBS.In the early planning and design stage,it is known that the styrene production capacity is 80,000 tons per year.

The alkylation reactor has the following requirements:a gas–solid reaction in static bed with gas reactant and solid catalyst,with a small flowrate,and without backmixing.The analytical column has the following requirements:the tower is with a small tower diameter,containing corrosive media,without turbid liquid,with gas–liquid two phases,vacuum operated and without operational flexibility.Taking the above equipment selection requirements as the query command,the recommended result of the equipment selection of this reactor is a fixed-bed reactor,and the recommended result of the equipment selection of the tower via the software is a packed column.

6.Conclusions

In this paper,an equipment selection software for industrial styrene process is developed based on S-ESKBS.Case studies are described for the selection procedures of the alkylation reactor and the analytical column,and the specialty and accuracy of the S-ESKBS are verified.The structure of the S-ESKBS is composed of two parts,the low-level knowledge base and the top-level interactive application.The low level knowledge base is the core of the S-ESKBS,including the equipment selection ontology library and the equipment selection rule set.The ontology description language OWL is used to construct the equipment selection ontology library based on the seven-step method.The rule description language SWRL is used to construct the equipment selection rule set based to the design requirements of the S-ESKBS.The low-level knowledge base uses the Pellet inference engine to authenticate the low-level knowledge base and implement knowledge inference.The top-level interactive application of the S-ESKBS includes the parsing storage layer,the reasoning query layer and the client application layer.The parsing storage layer applies Java-based Jena API to parse the low-level knowledge base into RDF triplets and store the results of parsing into the data storage module.The reasoning query layer consists of SparQL query module and Jena inference module,which is used to query the low-level knowledge base.The client application layer is the implementation of the top-level interactive application,using JavaWeb three-tier architecture and MVC design patterns to achieve the front-end interaction.

Fig.6.SSF-ESKBS client application layer.

Fig.7.The software interface of the S-ESKBS for the industrial styrene process.

The domain ontology technology and knowledge-based system now are widely used in many fields,but for the equipment selection of industrial process,there are very few reports.It is a new try in this paper to construct the S-ESKBS and develop the software,which will help to select the equipment of industrial styrene process more convenient and smart.In the future work,some more complex units and industrial processes will be considered,which need more complicated and reliable expert knowledge and inference rules.