An agent-based service-oriented integration architecture for chemical process automation☆

Na Luo,Weimin Zhong,Feng Wan,Zhencheng Ye,Feng Qian*

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

Keywords:Automation system Integration Web services Service-oriented architecture Agent

ABSTRACT In reality,traditional process control system built upon centralized and hierarchical structures presents a weak response to change and is easy to shut down by single failure.Aiming at these problems,a new agent-based service-oriented integration architecture was proposed for chemical process automation system.Web services were dynamically orchestrated on the internet and agent behaviors were built in them.Data analysis,model,optimization,control,fault diagnosis and so on were capsuled into different web services.Agents were used for service compositions by negotiation.A prototype system of poly(ethylene terephthalate)process automation was used as the case study to demonstrate the validation of the integration.

1.Introduction

Automation system is essential to run chemical processes safely and efficiently in the large scale plant.It can gather information automatically from various sensors,analyze operational situations and take control actions.Extended functions are integrated into automation system,such as exploring more profitable opportunities,optimizing process,scheduling,and planning.Nowadays,automation system is integrated with business and supplier databases[1].The increasing automation system is dependent on many applications,which are connotation packaged,stand-alone and distributed.

Due to the distributed nature and the hierarchy architecture of chemical process automation[2], flexible framework is required to support information sharing and exchanging,as well as software interoperation.Some feasible frameworks have been developed to meet this need.Multi-agent systems(MAS)[3]were firstly applied in control systems to make them more adaptable and manage their complexity[4].Many applications[5]have proved the success of MAS in the context of discrete manufacturing.In terms of MAS in continuous process,some work has alzso been done.Teppo Pirttioja[6]and Antti Pakonen[7]developed agent based process automation systems separately.Gao[8]presented an agent-based intelligent system to support coordinate manufacturing execution and decision-making in chemical process industry.Though agent-based solutions for enterprise integration have been an active area in the past ten years,industrial support is lacking on the development and deployment of practical agent applications.Besides MAS,holonic manufacturing is being developed as a possible solution in the next generation manufacturing systems because of its consistence with the nature of chemical process automation.McFarlane[9]studied the development of automation systems in continuous manufacturing and discussed holonic technologies in steel rod rolling mill.Chokshi[10]applied holonic principles in chemical process industries which supported flexible unit operations to dynamically integrate and collaborate with others when the production conditions changed.Though holonic architecture is appropriate for hierarchy system,the concrete technology supportis missing.Compared with holonic manufacturing,agent technology is a suitable approach for the implementation of holonic and reconfigurable manufacturing control applications[11].

Different from MAS and holonic manufacturing,Service Oriented Architecture(SOA)is a new framework which is often implemented using web service technology.SOA provides a communication platform between distributed and heterogeneous systems and applications.Web service-driven industrial systems are defined as next-generation systems[12].For industrial automation,a service oriented paradigms named SIRENA project[13]has been used as a case study.It is easy to reconstruct the automation system referring SIRENA.However,SIRENA only considered intelligent devices.In fact,more intelligence in process automation should be embedded in automation system.It is obvious that only SOA cannot solve all the problems.

In order to address the numerous facets of huge process automation,combination among heterogeneous technology was raised.Among these,the combination of SOA and agents received more attention recently.Poggi[14]discussed the agent-based service oriented structure for system integration.Shen[15]integrated collaborative intelligent manufacture based on agent-based service-oriented integration architecture,in which every agent core was built into a web service.Rishi[16]combined service oriented architecture with agent technology which is used for business service mappings.Liu[17]proposed a multi-agent-based service-oriented architecture for inter-enterprise cooperation system and established the basis for transforming the interenterprise cooperation business models into multi-agent-based SOA components.Considering good prospects of MAS and SOA,Huhns[18]even presented an agenda for the deployment of such solutions.

Combination of agents and web service technologies into a cohesive solution avoids the weaknesses of each individual technology and reinforces their individual strengths[19].Though the combination method has been used in many manufacturing systems,no related system is reported in the domain of chemical process automation,as far as our information goes.In reality,traditional control systems builtupon centralized and hierarchical control structures present a weak response to change and are easy to shut down by single failure.Aiming at these problems,this paper proposed a new agent-based service-oriented integration architecture for chemical process automation,wherein web services were dynamically orchestrated on the internet using agent behaviors built in them.A prototype system was illustrated as the case study to demonstrate the validation of the method.

The rest of this paper is organized as follows:Section 2 provides an introduction of SOA and web services.Based on these technologies,Section 3 discusses the hierarchy architecture of chemical process automation and provides an analysis of ontology.The integration structure based on agent and SOA is illustrated.In Section 4,a typical chemical process automation system of poly(ethylene terephthalate)is integrated based on the new architecture as a case study.Section 5 concludes the paper with some perspectives.

2.Service-oriented Architecture and Web Service

MAS have been a research topic in computer science already for a long time which can be found easily in many books.So this paper focuses on reviewing service oriented architecture and web service technology.

Service Oriented Architecture(SOA)is a new framework which utilizes services as underlying elements for developing applications.It is used to support the development of rapid,low-cost,interoperable,evolvable and massively distributed applications.Web services are currently the most promising technology to implement SOA.As an interface,web services are proposed mainly to make the information of the original isolated sites communicate and share with each other.A series of open Internet-based standards are used in web service,including Simple Object Access Protocol(SOAP),Web Services Description Language(WSDL)and Business Process Execution Language for Web Services(BPEL4WS).Web service can be used in any environments(Windows,Linux)which support these standards.

There are several key technologies when building and using web service.The first is how to describe data.XML is chosen as a standard description method.In order to transmit data in any environments,SOAP is used as an information exchange protocol.Web service uses WSDL to be understood by users.UDDIis a protocol of Universal Description,Discovery and Integration,which is a platform-independent,XML-based language for describing business on the Internet.With all of these,web services can go well beyond simply exchanging information to accessing,programming,and integrating application services encapsulated within old and new applications.

Services are autonomous platform independent computational elements and they could be described,published,discovered,orchestrated and programmed using XML for the purpose of developing massively distributed interoperable applications.These standards together provide an open XML-based mechanism for application interoperability,service description and service discovery.

3.Chemical Process Automation Based on Agent and SOA

3.1.Analysis of chemical process automation

In chemical process,automation system is concerned with managing and controlling the physical activities,aiming to execute the plan and monitor the process of the product.With the automation system,the process is required to exhibit a number of desirable characteristics such as efficient, flexible,reliable and safe operations,and their seamless integration with supply and distribution chains.Due to the complexity,multiple levels of control are involved,as illustrated in Fig.1.The higher levels deal with strategic or tactical issues.For the temporal horizon of days and weeks,the planning is concerned.Within a shorter temporal horizon,scheduling is defined to respect a specific criterion.In the lower levels,the real-time operational issues include the process control conventionally.For higher efficiency of process operation,this level is expanded to cover optimization,steady state modeling dynamic modeling and process control.Also,monitoring and diagnosis are essential parts of the whole system.It is obvious that the automation system is suitable to be integrated with hierarchy architecture.In the architecture,the global visibility of control afforded by higher levels in terms of wider time and physical scopes enables the system to behave in a predictable and stable manner when the conditions are planned and stable.

3.2.Integration architecture

It is obvious that the process automation conforms to hierarchy architecture.In another point of view,the nature of the system is distributed.So the automation system can be seen as the combination of parallel services and agents in five levels which is illustrated in Fig.2.At the bottom of the architecture is the physical process,which provides information for other software application.Above are different automation systems which are put into use in different times and different operational systems.Considering the heterogeneous characteristics,these applications are encapsulated into fine-grained services.For the cooperation of these services,agents with different functionalities are constructed above these services.The objective of agents comes from the results of planning and scheduling which are provided by ERP or other applications.All of these are located in same or different physical positions and custom users interact with the automation system by local or remote clients.Each level is discussed in detail as follows.

3.2.1.Physical process level

In the chemical plant,a process can be seen as a composition of units interconnected through piping streams and transferring equipment.The process units use the utilities supplied by utility suppliers to perform their process tasks as specified in the product recipe.In the physical process,materials are exchanged to execute the processing tasks.Additionally,real-time information is shared on implementing the basic control functions.From the automation point of view,the interactions of information maintain the physical process running at the target settings.

Information of the process comes from controllers and instruments running on different units.For seamless connection and interoperability in industrial automation,OPC is applied to exchange information.In the application,process operational data and functionalities can be made available as services in a vendor independent fashion[20].Work is under way to standardize information models for physical device information,analyzer devices,plant operation and maintenance,batch control and PLC programming[21].The operation modules take care of the activities of a process automation service through using the models and updating the run-time data structures.

Fig.1.Chemical process automation systems.Adapted from Ref.[11].

Fig.2.Integration architecture.

3.2.2.Service level

Service granularity is a key problem for definition between physical process level and agent level.At service level,a lot of legacy sub-systems are capsulated.There are several kinds of services in the automation system.

3.2.2.1.Data services.For the real-time control,the interconnection with physical devices is required,making it able to read data from sensors and to send actions to actuators.Data in industrial process are classified into three categories which include historical process operation data,real time operation data and predictive data.Historical data describes the previous states of process operation,while predictive data reflects expectations of process performance at certain operation condition,which can be used to provide a short-term process operation forecast.Predictive data can be obtained on the basis of process simulations or the modeling of the historical data.These raw process operation data are often with noise which makes the result unavailable by other services.Data service corresponds to information retrieving service specialized for data retrieving or mining methods.Considering the characteristic of the chemical process,data correction,coordination, filtering and reconciliation methods are encapsulated in the services.Of course,new data processing methods can be added into the structure easily.

3.2.2.2.Model services.Models are developed to support process operations[22,23].In practice,the integration of complementary model based technologies into a single application is not possible because these process models will exhibit various characteristics.Models are either black-box or white box,empirical or mechanistic,small-or largescale,and they can encompass a single key unit or multiple units of equipment[24].These process models can also be derived for a single purpose or span multiple applications.They can capture slow and/or fast dynamics,and be computationally demanding or relatively inexpensive to solve.For the process models,it is obvious that they are distributed and heterogeneous.However,their application targets to process operation support and process control is also the part of the framework.In this project,the previous developed heterogeneous models are encapsulated into different services as shown in Fig.2.

3.2.2.3.Optimization services.Optimization of chemical process is organized in multiple levels according to the distributed models.The objective of the optimization is based upon the production scheme.Through optimization,operating conditions will be regulated to achieve optimal performance.For the optimization is in local level,the calculation is out of difficulty,while in some multi-modal problems,conventional optimization methods can hardly get global optimum.Evolutionary optimization methods such as genetic algorithm,particle swarm optimization algorithms are applied for solving the problem.For problems with complicated constraints,cooperation of services compromises the contradiction of optimization results and makes it possible to reach a certain feasible result.In optimization services,different optimization algorithms are encapsulated.

3.2.2.4.Control services.The control services aid the operator in executing and controlling the different steps of a process transition.Among these services,regulatory control services interact with the Distributed Control System and perform actions such as reconfiguring the controller settings based on the current state of the process.The sequential control services coordinate among the discrete steps required for executing sequential operations.Predictive model control services are used for controlling complex system based on explicit models.The alarm management services help reconfigure the alarm management system to the current state,and thus prevent alarm floods and nuisance alarms.

3.2.2.5.Monitoring and fault diagnosis services.This kind of services include basic fault,fault monitoring and diagnosis prediction.The basic fault diagnosis is based on the detection of operation units and online diagnosis.Fault monitoring detects the data online and sends the information to fault center.Also it disposes the fault information,evaluates the process and predicts the fault.

3.2.2.6.Service communication.As a communication protocol,SOAP is independent of platform.It can be used for information exchange between different systems and make communication much easier between sub-systems.Also,SOAP solves the interface between different systems.

3.2.3.Agent negotiation level

In service level,basic services provide simple functionalities which cannot solve complex problems.Dynamic composition of service is necessary for flexible problems.Composition of services is directed by agent that is driven by objectives.There are two types of service compositions in the agents.One is simple composition.The agent constructs simple services in sequence on single level,which is modeled by listing the invocations of the involved atomic services one after the other.The data flow between the web services is modeled through the use of common variable names.The other is a nested composition.In construction,agent develops composite process including another composite process as one of its sub-processes.In nested compositions,there are multiple levels of composite services in contrast to simple compositions.Each contained composite process should also be axiomatized in a recursive fashion until there are only atomic web services to be represented as individual events.In nested composition,the principles of service decision are included.The interaction of reconfiguration service agents is illustrated in Fig.3.

In Fig.3,an agent reconfigures related services and communicates through XML.Data service collects data such as temperature,pressure,liquid location and other operation state parameters through their OPC interface.As to data which cannot be obtained using the physical instruments directly,data service can use soft sensors to calculate them from recurrent data and return them to the data service.Data service has the function of filtering the data with fixed time interval.Decision service integrates many process experiences.The agent gives the decision of the objective so that this service can work aiming at it.When the agent receives a message,it judges the importance of the information and whether it is needed.Then the agent gives it to decision service.For example,the agent obtains a message that the quality sensor data is out of the limit from the data service,and then the agent gives it to the decision service to dispose of.The decision service first confirms which unit mostly influences the specific product quality.If unit A is the one,decision service returns message to agent and gives it the strategy to dispose of this kind of problem.According to the strategy,the agent activates model service to request calculated results and sends message to data service to affirm the sensor data.When model service returns the calculation,the agent returns the information to decision service for the next step action.When the condition is about fault,the agent sends message to diagnosis service and activates it to solve the problem.If the energy is over consumed,the agentasks the decision service and obtains the objective and constraint conditions.Then the agent activates optimization service to do it.

3.2.4.Strategic level

In the higher level,process planning involves multi-stage decision making based on a variety of manufacturing knowledge and logic.Under the level of planning is scheduling,which represents a significant proportion of all the works undertaken in the production planning.Recently,most industries developed ERP systems,which can be integrated into the distributed environment to provide objective for agent negotiation level.

3.2.5.Representation level

Fig.3.Interaction of services in agents.

In the integration architecture,users only need a browser to obtain the information of the automation system.The representation places the client application in a state.The result of the client traversing a hyper link is accessed.The new representation places the client application into yet another state.Thus,the client application changes(transfers)state with each resource representation.

4.Applications

Poly(ethylene terephthalate)(PET)is an important material for the production of synthetic fibers, films and beverage bottles.The common production method of poly(ethylene terephthalate)(PET)is continuous direct esterification of terephthalic acid(TPA)with ethylene glycol.A PET production usually consists of five sections[25]:(1)primary esterification;(2)secondary esterification,(SE);(3)low polymerization,(LP);(4)intermediate polymerization and(5)high polymerization,(HP).A typical process is shown in Fig.4.

In the automation system ofPET process,distribution control system is often used.Process History Database(PHD)collects the process data and stores them.Also PHD provides OPC interfaces for other application systems.In order to get data through the internet,the data can be obtained in the database server.Description of data services is illustrated in Fig.5.

Integrated automation system for PET was developed based on the agent-based service-oriented architecture.SOAP was used to communicate between web services.This paper mainly focused on five web services which were data service connecting with PHD interface,optimization service based on Matlab,fault diagnosis service based on C++,model service based on Aspen plus,Aspen Dynamics,Hysys and control service based on predictive model control.As an example,optimization service described by XML is illustrated in Fig.6.Data service requested data from PHD.The agent judged if it was needed to simulate the process again.If yes,model service gave simulation results.If some rules were triggered,optimization service gave the best operation parameters.When there were abnormal data,the agent activated fault diagnosis service.Control service provided some control strategies and connected with optimization service.The prototype has been implemented and the interface is shown in Fig.7.

Fig.4.Process of PET.

Fig.5.Description of data service.

For the process optimization,agent nested composition were used.The agent selected the models based on aspen plus.Also,it could use the surrogate model.When there was a rigorous model,the agent could use it.Then,the agent selected optimization service.For there were so many kinds of optimization services,the agent selected the simple algorithm prior.In the system,the traditional algorithm Newton method was the simplest algorithm.The algorithm selected it.But the algorithm was apt to trap into local optimization.Thus,another algorithm GA service was selected too.When the two algorithms obtained different solutions,other optimization services were selected.If the difference of optimization solution was large than the set value,more optimization services were selected until the difference was less than the set value.The optimal operations were sent to service control as the set point.The control services adjusted the process condition.The result of the optimization and on line optimization is illustrated in Fig.8.

In Fig.8,it can be seen that the flow of heat medium decreases with the time.With less flow of heat medium,the energy cost of the process drops approximately 10%.

5.Conclusions

Considering the characteristics of chemical process industries,this paper proposed a new integration architecture based on service oriented architecture and agent technology.With the goal of stable production,high profits and low cost coming from the planning and scheduling,the architecture integrated data analysis,rigorous mechanism and intelligent modeling,optimization algorithm,control strategies,etc.as web services.Agents were used for service compositions by negotiation.This architecture provided a unified framework for system integration of chemical process industries.The integration of poly(ethylene terephthalate)automation system demonstrated that it worked well in reality.Due to its generic nature,the architecture can also be applied to other process industries applications.

Fig.6.Optimization service described by XML.

Fig.7.Interface of system.

Fig.8.Optimization result of the reactor in PET process.