B3.0 c2 role of semantic models in smarter industrial ops mom c 3 reprinted with permission 20130325

ISA-95 Best Practices Book 3.0Chapter 2: The Role of Semantic Models in Smarter Industrial Operations

“Source: The MOM Chronicles: ISA-95 Best Practices Book 3.0. Copyright © 2013 ISA. Reprinted by permission. All rights reserved.”

Copyright© 2012 IBM Corp. All rights reserved. Bill Bosler’s reprintCopyright© 2012 International Society of Automation. All rights reserved. Page 1 of 35

I was pleased to have been a “Contributing Editor” to Chapter 2 of the recently published The MOM Chron-icles: ISA-95 Best Practices Book 3.0. I spoke with ISA’s Susan Colwell, who spoke with the Editor, CharlieGifford, who authorized me to make this reprint with the following reference:

“Source: The MOM Chronicles: ISA-95 Best Practices Book 3.0. Copyright © 2013 ISA.Reprinted by permission. All rights reserved.”

For the Reader’s information, I am working with the new Russian equivalent of the American PetroleumInstitute, which they call National Institute of Nefte (oil) & Gaz, (NING) to help the Russian O&G IndustryLeapfrog their performance using ISO 19526 for data management during the Construction Phases andthen use that data for Hand-Over and population of an Engineering Management of Change System to con-figure the MES and Control Systems during the O&M Phase. Part of that Configuration is the use of B2MML(ISA 95) for Integration.

ISO 15926 is based on Semantic Computing Concepts. Hence, my interest in the distribution of this Chap-ter 2 to the Russian Community. If there is an interest in having this document or the entire book translat-ed to Russian for distribution via ISA, please let me know, [email protected] or +7 967 053 27 54.

Here is a version of Charlie’s announcement that the book is published:

Springing from the international success of Books 1.0 and 2.0, the ISA-chartered ISA-95 Best PracticesWorking Group after 2 years of work have completed its third collection of Manufacturing OperationsManagement (MOM) methodology white papers as The MOM Chronicles: ISA-95 Best Practices Book 3.0 .The Book 3.0 collection focuses on MOM system engineering methods to organize the complex 21st-century manufacturing plant and to optimize its role in a global supply chain.

ISA Publishing charges ~$109 for the book. They do NOT sell an electronic version. (BB Note: ISA is plan-ning one for the iPad, Kindle, and a CD-ROM later this year.)

Place your order through our customer service representative, Brittany Lynn ([email protected]), di-rectly at 919-549-8411.

Publicly order Book 3.0 from ISA Publishing at www.isa.org/momchronicles (link).

The paper collection in Book 3.0 is titled “The MOM Chronicles”, which is derived from the legendary workof The Martian Chronicles by Ray Bradbury. Our working group has dedicated the Book 3.0 collection toRay Bradbury, who passed away on June 5, 2012. Ray Bradbury’s futuristic works were (are) a major influ-ence in lives of many of today’s technologist and scientists who look for positive directions and ways toinfluence the world for the better.

The Book 3.0 collection addresses the need for true manufacturing system engineering methods to organ-ize the complexities of the manufacturing operations.

The following are the white papers as book chapters:

Chapter 1: Applying Global MOM Systems in a Manufacturing 2.0 Approach

Chapter 2: The Role of Semantic Models in Smarter Industrial Operations

Chapter 3: Applying Manufacturing Operations Models in a Discrete Hybrid Manufacturing Envi-ronment

Chapter 4: Defining an Operations Systems Architecture

Chapter 5: A Workflow-driven Approach to MOM

Chapter 6: Scheduling Integration Using an ISA-95 Application in a Steel Plant

Chapter 7: Intelligent Integration Interface: I3, A Real-world Application of ISA-95

The ISA-95 Best Practices Working Group discusses the latest methods and thoughts on implementingMOM, EMI and related manufacturing systems.

Charlie Gifford




President & Chief Manufacturing Consultant21st Century Manufacturing Solutions LLCO: 208-788-5434, M: 208-309-0990, F: 208-788-5690E: [email protected]

”Doing your best is not good enough. You have to know what to do. Then do your best.”W.E Deming

2THE ROLE OF SEMANTIC MODELS

IN SMARTER INDUSTRIAL OPERATIONS




The Role of Semantic Models in Smarter Industrial Operations

Table of Contents

Introduction ............................................................................................................................................................ 1

The Manufacturing Complexity Problem ................................................................................................................ 3

Why Semantic Models?........................................................................................................................................... 6

Evolution towards Semantic Model-Based Operations Integration ....................................................................... 8

1. Centralized Application Ownership of Data.................................................................................................... 8

2. Data-Centered Architecture............................................................................................................................ 9

3. Distributed Ownership of Data ....................................................................................................................... 9

4. Message-Oriented Architectures (MOA) ...................................................................................................... 10

5. Service-Oriented Architecture (SOA) ............................................................................................................ 10

6. Information-Oriented Architecture (IOA) ..................................................................................................... 10

7. Model-Driven Architecture ........................................................................................................................... 11

The Information Model – The Heart of the Semantic Model ............................................................................... 11

Operations Lineage ........................................................................................................................................... 13

Operations Data Complexities .......................................................................................................................... 13

Semantic Models................................................................................................................................................... 15

Model Servers ................................................................................................................................................... 16

Semantic Models and Model Management Middleware ................................................................................. 17

Example: Manufacturing Operations and Business Intelligence within Level 3 Workflows Mapped to Level4......................................................................................................................................................................... 20

Putting It All Together ....................................................................................................................................... 21

Conclusion............................................................................................................................................................. 26

Bibliography .......................................................................................................................................................... 27

Glossary, Acronyms and Abbreviations ................................................................................................................ 28

Authors.................................................................................................................................................................. 31

Contributing Editors .............................................................................................................................................. 31

Reviewers .............................................................................................................................................................. 32




The Role of Semantic Models in Smarter Industrial Operations

Introduction

This article looks at the application of semantic model design and technology in industrial operations inte-gration and at the evolving role of Semantic Computing. Semantic (data) modeling is compared as a corecomponent of application architecture, to more familiar architectural integration patterns. The functionalcapability describes the operations basis of semantic models. The value of semantic models is illustratedthrough a series of examples that should be familiar to the reader.

Best practices only match the competition; so to compete by differentiating its performance, each compa-ny must judiciously apply leading practices. Semantic Computing now has sufficient industrial applicationto now be considered a leading practice. As such, various ways in which Semantic Computing supplementlegacy technologies, possible longer term enhancements are suggested to facility safety and integrity.

Three critical elements are often described in discussions around smarter plant solutions, of which seman-tic concepts are a key element. These three critical elements, the three “I”s as they are sometimes labeled,are “Instrumented,” “Intelligent” and “Interconnected.” These elements support the idea that much dataare collected from the world around us and if we use the three “I”s to federate the data, operations intelli-gence is derived and with that we drive timely communication, response and optimization around criticalbusiness tasks. What is important in this approach is the ability to interpret data for timely analysis and toderive understanding from a wide variety of sources in a wide range of formats and context. With SemanticComputing, the calculations and analysis be generalized to permit application to all instances of similar ob-jects.

Data in the real world are subject to constant change. Therefore, structures need to be self-adapting andnot rigidly pre-defined. This difference is referred to as the "Open World" versus the "Closed World." Se-mantic modeling and its technology identifies changes in underlying data and the potential interactions ofthose changes. However, for the most part, the role of semantics is to alert a human in a timely mannerand in the appropriate context so that responses to those changes can be identified and appropriately act-ed upon.

Implemented semantic models have the ability to federate data from any connected data store into an ag-ile, adaptive, fit-for-purpose model that leverages and extends industry standards and ontologies. Whensemantic models are coupled with applications that perform analysis, logic, reports, views etc. that are eas-ily and consistently applied and adapted, manufacturers globally are truly evolving to an environment inwhich business and operations personnel are directly in control of their data, business and operationsrules, and business and operations processes. Some refer to this evolution as the “evolving ubiquitouscomputing model,” since computing power has become highly distributed and pervasive.

Industrial architectures must be designed to handle ever-changing, disparate data and implied, actual rela-tionships between the data. Data sources include structured and unstructured data, sensor data (currentvalue and historical), images, audio and video. In addition, interactions of proposed data changes must beidentified so that coordinated change is the rule and discovery is minimized. Not only does current datahandling not fit well into standard relational persistence structures, the challenge is to make sense of thisdata in context and adapt to additions, deletions and changes with validation but without undue complexi-ty.

Another major use of Semantic Computing is the monitoring of the overall mechanical integrity of the myr-iad of disparate systems and services deployed in an industrial facility. It is one thing to develop a commonontology, which is a prerequisite to developing a semantic model, but it is a totally different thing to peri-odically verify the complete integrity of the compliance plan for regulations, standards, policies, proce-dures, etc. that apply to an operating industrial facility. Verification requires confirmation that the compli-ance plan is being followed, that system configurations are as approved and that the needed criticalequipment is operating or, in the event that such confirmation cannot be obtained, that special efforts are




being applied to monitor the increased risk in an appropriate manner. If an event continues, the systemshould escalate the alert in such a way that a human cannot suppress the message being sent to higherauthorities.

Interdependence and interactions are becoming too interconnected in our world to not use technology tohelp us verify compliance to stated intent. The United States EPA created Title V to ensure that every plantmanager has a total environmental compliance program and reports verifiable, auditable monitoring re-sults every six months. This requirement has significantly improved environmental compliance. Perhaps itis time to use semantic technology to monitor and verify safety and general compliance that includesstrong, agile governance.

Consider a smarter city traffic management application. Real-time traffic data is provided through trafficlight sensors, Department of Transportation (DOT) speed sensors and video cameras. New roads and trafficsignals are added and some fail. Additional data that are critical to accurate traffic flow prediction cancome from a wide variety of sources including weather reports, accident reports, tweets, transit interrup-tions, calendar events like holidays, seasonal trends like beach traffic, special events like parades, festivalsor major sporting events, emergency dispatch events, and significant news events. This multitude of datasources describes events in their own terminology, which is likely to be somewhat different from that ofthe other sources. All these data must be assimilated in useable form and understood in context, and therelationships between events must be interpreted to infer intelligence.

The roads, signals, sensors, etc. represent the fundamental model of the transportation network. The cur-rent readings, condition of the sensors, declarations about events, vehicles etc. represent the conditions ofthe road system and the level of activity on each road. This is the more transient data to be interpreted.The model changes less frequently, is independently verified in background, and thereby is subjected to anappropriate governance model and usage patterns developed. In fact, changes to the model can even beanticipated by monitoring budgets, construction schedules etc. The current readings and news feeds(whatever their origin) are much more dynamic and are more challenging to verify within a short timeframe. By separating these two classes of information (model and more transient data), additional logiccan be derived to enhance the verification of the more transient data and thereby decrease response time.

To communicate effectively, a common understanding of the events and their context in operations workprocesses must be referenced from these various sources. For example, basic terms such as “vehicle” maybecome ambiguous between data sources and providers, whereas distinctions such as cars, light trucks,semis, buses or motorcycles may be significantly less so. Some characteristics, such as number of axles ornumber of occupants, may become important distinctions. The ontology can relate “vehicles” according tothe situation. Of course, the relevant data being gather are continually changing. Fortunately, the modeldata change much less frequently than the real-time data feeds. Semantic Computing inference enginesfacilitate understanding of both the model and the transient data changes.

Semantic modeling defines the data and the relationships between these data entities. An industrial or op-erations information model provides the ability to abstract different kinds of data and provides an under-standing of how the data elements relate. A semantic model is a type of information model that supportsfixed and ad hoc modeling of data entities and their relationships. The total set of entities in our semanticmodel comprises the taxonomy of classes that can be used in our model to represent the real world. To-gether these ideas are represented by ontology, the vocabulary of the semantic model that provides thebasis on which user-defined model queries are formed. The model supports the representation of entitiesand their relationships, supports the constraints on those relationships and entities and aggregates termi-nologies as appropriate for the query; for example, the definition of “vehicle.” This provides the semanticmakeup of the information model.

In the World Wide Web Consortium (W3C) version of Semantic Computing, access to data by an individualelement is separated from the data store by the ontologies-aligned rules being applied to the data for vali-dation, calculation etc. and derived inferences. This is much more than just a fundamental concept forproductivity and mechanical integrity of the system or overall architecture.




Current Semantic Computing technology enables:

Federating data from multiple data stores into a normalized form without MOVING the data fromits Master Data Store or system of record (SOR)

Creating an agnostic data model that is dynamically structured by the engineer to focus on thethen current need

Delivering data to consumers in any form or schema and to any required tool

Integrating business and operations rules within an information model that is independent of anysingle application

Developing algorithms, etc. operating across all existing and new instances of desired objects with-out special efforts on the part of the user

These are major transitions from the typical methods of information management and business processes.As a result, they do require reorientation and restructuring of the IT team to produce the information fromthe traditional manner for the business and operations teams.

The long term goal of Semantic Computing is summarized as achieving an agile, adaptive environment inwhich:

Data + Model = Information

Information + Rules = Knowledge

Knowledge + Action = Results

In our traffic management example, a semantic model permits users to understand relationships such as 1)those between the traffic light sensors and the intersections being monitored, 2) the association of anygiven traffic light sensor with other sensors on the same road, or 3) the relationship of the roads to otherintersecting roads and to major highways. The semantic model may also yield similar information aboutbus lines or subway lines to further describe the types of services available within the locations serviced.The extrapolated relationships between stations and street addresses, service lines and surface roadroutes additionally provide the basis for understanding the implications of specific disruptions in masstransit service on road traffic.

As an additional complication, a single application must interact with multiple domain models or domainontologies. One means to achieve that interaction is to merge existing ontologies into a new ontology. It isnot necessary to merge all the information from each of the original ontologies because that integrationmay not be able to be satisfied logically. In addition, the new ontology may introduce new terms and rela-tionships that serve to link related items from the source ontologies. In an example in a later section, thispaper closely investigates how semantic models best fit.

The Manufacturing Complexity Problem

In today's business climate, the challenge of effective decision making is exacerbated by an influx of dataand by increased urgency for actionable information. Line-of-business managers are looking for the infor-mation infrastructure to alert them to evolving undesirable situations early enough that they can mitigatethe undesirable situations with contextual visibility and decision support in a business intelligence (BI) andoperations intelligence (OI) solutions. In a manufacturing enterprise, information exists in a variety of is-lands that must interface and communicate with each other to orchestrate operations work process execu-tion. Each island is a System of Record (SOR) in an operation department or area where the informationsystem is the authoritative data source for a given data element or piece of information. SORs include ERP,MES, LIMS, QMS, WMS, SCADA and IO devices, etc. The structure of information within each SOR changesas changes occur in the product SKU (stock keeping units) counts, production mix, line scaling for through-put rate and work processes due to continuous improvement initiatives and process technology improve-ments. SORs use a very specific data model that is designed for specific application(s) functional require-ment(s) in supporting real-time work processes. In today's rapidly changing world, the non-static nature of




MOM and process control applications and the long adoption time frames are major inhibitors to continu-ous improvement and the competitiveness of the enterprise. These are major cost factors in implementingand maintaining integrated MOM and process control solutions.

SORs normally have a limited ability to adapt to process and data changes at a reasonable cost and within ashort time frame. This is the main reason that 60-90% of IT budgets are for maintenance and not for en-hancements. By exposing the SOR’s data to Semantic Computing applications (see “Semantic Models” for adefinition of Semantic Computing), additional functionality is more easily integrated, but more importantly,formal compliance with all regulations, standards, policies, procedures etc. is able to be verified withineach SOR and between SORs, and independent of each and every SOR.

Operations within the manufacturing enterprise are driven by ever-changing structured business and oper-ations processes (activities) utilizing the information and capabilities of multiple SORs. Each SOR partici-pates in work process execution with equipment, materials and people, which creates dynamic, ad-hocoperations processes and ontologies of information. Since 80–90% of the information used by knowledgeworkers already exists, Semantic modeling technologies are effective at focusing the actionable data frommultiple SORs, permitting knowledge workers to use each the SOR effectively to execute accurate andtimely actions. In truth, Semantic Technologies can update the underlying data store, but for the near fu-ture it is probably best for knowledge workers to use the appropriate SOR to update the data.

The ISA-95 Part 3 standard, Activity Models for Manufacturing Operations Management (MOM), identifiesfour models of activities that involve various SORs for MOM that must exchange data and events:

Production Operation Management

Quality Operation Management

Maintenance Operation Management

Inventory Operation Management

ISA-95 Part 3 states that MOM activities are those activities of a manufacturing facility that coordinate per-sonnel, equipment, material and energy in the conversion of raw materials and/or parts into products.Each MOM activity model is made up of tasks within functions and data exchanges between tasks and afunction. The function and tasks may be performed by physical equipment, human effort and/or infor-mation systems, per Purdue Enterprise Reference Architecture (PERA) (www.pera.net) concepts (the origi-nal reference model for ISA-95). For each operation in an operations route for work order, some subset ofthe four MOM activities, functions, tasks, and exchanges is invoked to execute the work processes of thatoperation to complete the work order using various pieces of equipment, SORs and people. Following thisparadigm, the operations resources (materials [raw, intermediate, consumed, and finished goods], equip-ment and tooling, MOM and process control systems and people) are treated as participants equivalent toSORs in the execution of operations processes. Operations resources are sources of data and are targets offunctionality invocation that change in state, similar to a database. In other words, a given task and or dataexchange is only performed by people, equipment or systems of record (SORs) – an information systemwhich is the authoritative data source for a given data element or piece of information. The application ofoperations resources at each operation of a route is executed as a set of tasks for each activity as definedin the operations definition and rules. For a given product and production line, the operations definitionsand rules to schedule and execute an operations schedule are typically parts of various SORs. The methodof orchestration for all the operations of a production route must validate the changes to the operationsdefinitions during order execution and must be responsive to change while maintaining the consistency ofinformation between SORs to produce the desired result of each activity’s tasks and data exchanges.

As manufacturing Level 4 business and Level 3 operations processes execute, data from various SORs areused and/or changed. Events are triggered from various sources to advance process execution and/or startnew business and operations processes. Traditionally (that is, pre-Semantic Computing), some these pro-cesses are defined and executed outside of SORs, leaving some transient history of the process executionoutside of these systems. In other words, a large number of these systems don't store the information




change history that is needed to understand the lineage of causes and effects that is required, in turn, tounderstand “what, when and why.” This information is essential for analysis to drive continuous processimprovement within the manufacturing enterprise. With Semantic Computing, this information is easilycaptured and made part of the formal record.

Applications such as process historians, if used to capture information, typically only include time contextand are not designed to handle complex and transactional data from SORs such as ERP and the MOM ap-plications. Likewise, ERP systems typically lack detailed execution information or high speed transactionsfunctionality; for example, to identify which production request was first processed after equipmentmaintenance was performed on a specific piece of equipment. To get this information from traditional sys-tems, someone has to know how to query records related to the equipment, operations order, work ordersand operations segment from Operations Execution Management Systems (Production, Quality, Inventoryand Maintenance). The queried records must be collocated and orchestrated from the time-stamp of therecords, production orders and work orders to find the complete genealogy of the production request thatwas processed around the time that the equipment was last returned to service. Time, being the consistentdata element across SORs, is the relevant context available to link information between SORs. Of course,this presumes that time was properly synchronized between the systems. With Semantic Computing, thesequeries can be structured and reused in an easily maintainable fashion, and with audit trails.

Manufacturing enterprises are rich with Standard Operating Procedures (SOPs) that define how operationsprocesses should be performed. Building on the example above, the manufacturing enterprise likely hasdetailed SOPs for equipment maintenance procedures, qualification of equipment into production and therouting and dispatch of products through the manufacturing process.

Currently and historically, SOPs exist in paper form, referenced on an as-needed basis. Since the SOPs exe-cute work processes manually (outside of an SOR), the genealogy and understanding of a detailed accountof what happened, when it happened, and why typically involves (at best) collecting and correlating paperrecords; at worst, meetings where people assemble the records by discussing what they recall as havinghappened. With Semantic Computing and automated business and operations processes, this changes, butthere is much resistance to this valuable change. Today’s manufacturing companies have strong, inherentresistance to change and to the reality that one has to understand the risks in and benefits of achieving thereward.

Compounding the problem, information within SORs typically exists in a form specific to the task or workprocess of a specific operation and not in a form that makes sense to domain experts looking at a broaderchallenge. In most cases, modifying, replacing or extending SORs to meet new needs is cost prohibitive, ifan option at all. The pragmatic solution is a less intrusive approach that meets the requirements of the dy-namic manufacturing environment and provides access to relevant, timely information in many forms fordepartments in the business and plant. In fact, domain experts throughout business and industry need astructured information environment to alert them of their puritan events for their contextual views foranalysis to derive their decision making. Via Semantic Computing, semantic models (with their separationof data from the data store and their ability to act on that data) play a key role in this pragmatic form ofMOM architecture.

Whether for city traffic management or management of an oil refinery, the inferences and understandingprovided by the application of semantic models via Semantic Computing are critical to properly derivingthe correct insights from the monitored SORs and process instrumentation. This near real-time analysisultimately leads to optimized, agile business and operations processes through timely, responsive decisionmaking. Semantic Computing greatly enhances the federation of data from various SORs by using its infer-ence capabilities to alert the correct person in the right business process to take the correct action in atimely manner and then to escalate if timely resolution does not occur.




Why Semantic Models?

What, exactly, are semantic models and how are they helpful for this type of operations systems integra-tion? First, for clarity, the distinction between models in Unified Modeling Language (UML) vs. Web Ontol-ogy Language (OWL) is explained:

UML is a modeling language that is used in software engineering to design artifacts largely around object-oriented systems. When operations systems integration based on information-oriented architecture (IOA)is explained in this UML context, the semantic models are leveraged as the functional core of an applica-tion in order to provide a navigable model of data and associated relationships that together representknowledge in our target domain.

The Web Ontology Language (OWL) is a family of knowledge representation languages for authoring ontol-ogies. The languages are characterized by formal semantics and Resource Definition Framework (RDF) /RDFSchema (RDFS) /XML-based serializations for the Semantic Web. OWL is endorsed by the World Wide WebConsortium (W3C) and has attracted academic, medical and commercial interest. The data described by anontology in the OWL family is interpreted as a set of "individuals" and a set of "property assertions" whichrelate these individuals to each other. An ontology consists of a set of axioms which place constraints onsets of individuals (called "classes") and the types of relationships permitted between them. These axiomsprovide semantics by allowing systems to infer additional information based on the data explicitly provid-ed. A full introduction to the expressive power of the OWL is provided in the W3C's OWL Guide.

An ontology defines the terms used to describe and represent an area of knowledge. Ontologies are usedby people, databases, and applications that need to share domain information (a domain is just a specificsubject area or area of knowledge, like medicine, tool manufacturing, real estate, automobile repair, finan-cial management, etc.). Ontologies include computer-usable definitions of basic concepts in the domainand the relationships among them (note that here and throughout this document, definition is not used inthe technical sense understood by logicians). They encode knowledge in a domain and also knowledge thatspans domains. In this way, they make that knowledge reusable.

The word ontology has been used to describe artifacts with different degrees of structure. These rangefrom simple taxonomies (such as the Yahoo hierarchy), to metadata schemes (such as the Dublin Core), tological theories. The Semantic Web needs ontologies with a significant degree of structure. These need tospecify descriptions for the following kinds of concepts:

•Classes (general things) in the many domains of interest

•The relationships that can exist among things

•The properties (or attributes) those things may have

Ontologies are usually expressed in a logic-based language, so that detailed, accurate, consistent, sound,and meaningful distinctions can be made among the classes, properties, and relations. Some ontology toolscan perform automated reasoning using the ontologies, and thus provide advanced services to intelligentapplications such as: conceptual/semantic search and retrieval, software agents, decision support, speechand natural language understanding, knowledge management, intelligent databases, and electronic com-merce.

Ontologies figure prominently in the emerging Semantic Web as a way of representing the semantics ofdocuments and enabling the semantics to be used by web applications and intelligent agents. Ontologiescan prove very useful for a community as a way of structuring and defining the meaning of the metadataterms that are currently being collected and standardized. Using ontologies, tomorrow's applications canbe "intelligent", in the sense that they can more accurately work at the human conceptual level.

Semantic models allow users to ask questions about what is happening in a modeled system in a more nat-ural way, in the form of structured queries, transactions, interfaces and reports. As an example, an oil pro-duction enterprise might consist of five geographic regions, with each region containing three to five drill-ing platforms and with each drilling platform being monitored by several control systems, each with a dif-




ferent purpose. One of those control systems might monitor the temperature of extracted oil, while an-other might monitor vibration on a pump. A semantic model allows a user to ask a question like “What isthe temperature of the oil being extracted on Platform 3?” without having to understand details such aswhich specific control system monitors that information or which physical sensor (typically represented byan OPC Tag) is reporting the oil temperature on that platform.

Therefore, semantic models are used to relate the physical world as it is known to control systems engi-neers (in this example) to the real world as it is known to line-of-business leaders and decision makers. Inthe physical world, a control point such a valve or temperature sensor is known by its identifier in a par-ticular control system, possibly through a tag name such as 14-WW13. This could be one of several thou-sand identifiers within any given control system, and there could be many similar control systems across anenterprise. To further complicate the problem of information referencing and aggregation, other datapoints of interest could be managed through databases, files, applications, or component services, witheach having its own interface method and naming conventions for data accessing.

A key value of the semantic model, then, is to provide access of information in the context of the realworld in a consistent way. Within a semantic model implementation, this information is identified using“triples” of the form “subject-predicate-object.” For example:

Tank 1 <has temperature> Sensor 7

Tank 1 <is part of> Platform 4

Platform 4 <is part of> Region 1

These triples, taken together, make up the ontology for Region 1 and can be stored in a model server, as isdescribed in more detail later in this article. This information is easily traversed using the model query lan-guage to answer questions such as “What is the temperature of Tank 1 on Platform 4?” much more easilythan was the case without a semantic model relating engineering information to the real world.

Another advantage of semantic models for this type of application is maintenance. Consider Figure 1.

Figure 1: Information Model Structural Approaches

The real-world model described here can be implemented with any of the types of models shown in Figure1. The Relational Model has relations between entities established via explicit keys (primary, foreign) enti-ties and for many-to-many relationships of associative entities. Changing relationships in this case is cum-bersome because it requires changes to the base model structure itself, which can be difficult for a popu-lated database. Querying for the kind of data based on a Relational Model can also be cumbersome since itresults in very complicated where clauses or significant table joins.

The Hierarchical Model (Figure 1) has similar limitations when it comes to real-world updates and is notvery flexible when it comes to trying to traverse the model “horizontally.”




The Graph Model, (right side of Figure 1) is how semantic models are implemented to make it much easierto both query and maintain the model once deployed. For example, a new relationship must be represent-ed that had not been anticipated during design. With a triples store representation, additional representa-tion is easily maintained. A new triple is simply added to the data store. This is a critical point: The relationsare part of the data, not part of the database structure nor part of a specific SOR.

Likewise, the model is traversed from many different perspectives to answer questions that were notthought of at design time. In contrast, other types of database design might require structural changes toanswer new questions that arise after initial implementation.

Semantic models (based on graphs) allow inferences to be made easily in a nonlinear way. For example: anonline service is considered for purchasing books or music. Such an application should be very good atmaking additional purchase suggestions based on your buying patterns. This is very common for e-retailsites which provide recommendations such as “since you liked this movie, you might also like these mov-ies”, or “because you liked this music, you would probably also like the following.”

One way to accomplish this is to use a semantic model and to add relations such as the following:

Enya <is similar to> Celtic Women

In addition, an ontology is established in which both Enya and Celtic Women are part of the music genrecalled “New Age.” These relations, once established in the model, make it simple to offer those types ofsuggestions when needed.

Evolution towards Semantic Model-Based Operations Integration

Over last 50+ years, a number of architectural approaches have been defined for the integration of systemsand for the representation of their information and processes. These have included data-oriented, mes-sage-oriented, service-oriented, and information-oriented approaches. The questions to be explored are:

How do these various approaches differ and relate?

Where, exactly, do semantic models fit from a real-time operations integration architecture per-spective?

What value do semantic models provide as a key component of real-time operations integrationarchitecture?

Figure 2: Operations Systems Integration Evolution

1. Centralized Application Ownership of Data

From 1950 to 1980, monolithic applications required that the application own all the data. All access todata was local; all definitions, relations and knowledge were contained in the application. Departmental-ized users across a business had direct domain knowledge so there was no need for integration.

Over the last 30 years, as computers became more affordable and proliferated throughout industrial oper-ations and included programmable logic controllers (PLCs) and embedded computing functionality in enddevices, the complexity of composite applications and systems has steadily increased. Between 1990 and




the present, databases and client-server architectures were and still are used to allow multiple concurrentclients to interact with a centralized data store.

With a centralized data store, a centralized application owns its portion of the data, and other applicationsmake direct calls to obtain information or to request that the called application perform some action. His-torically, this has involved direct invocation of another application through application program interfaces(APIs) or remote function calls (RFCs). APIs and RFCs are contained in a client library, to which the callingapplication is linked. The calling application, in this case, is responsible for understanding the semantics ofthe called application and is responsible for all data transformations, etc. Although fast from a perfor-mance perspective, this approach has proven costly from a maintenance perspective (and brittle, becausea failure in one application has a ripple effect through all applications directly connected to each other). Inaddition, in most cases the methods of data validation are poorly documented. Consequently, new re-quirements by a new consumer of the basic data result in parallel systems being deployed to capture theadditional data, with the introduction of alternative data validation rules. This practice results in multiple"versions of the truth" so all users of the data spend 50–75% of their time aligning and verifying data be-fore the data is trusted enough to be analyzed. This preparation time has been increasing as the complexityof the systems and point-to-point integrations has increased. An additional benefit of Semantic Computingis to reduce this wasted time.

2. Data-Centered Architecture

An alternative to a centralized application owning the data is a data-centered architecture. This is clearly astep forward from the direct connectivity approach because applications do not directly connect to eachother to exchange information. Data-centered architecture is anchored through the single definition of rel-evant business and operations data, around which systems are integrated and applications are developed.Put simply, data-centered architectures establish a common data model for a centralized data store and forclient applications that interoperate using the rules and requirements of this centralized data store. Unfor-tunately, data is copied from its point of creation at various SORs, where unique validation rules are ap-plied. This also results in multiple “versions of the truth.”

One early example of such an approach was SAP’s earlier incarnations; these SAP versions were (or at leastseemed to be) basically a suite of 40+ applications developed to interact around a central data model. Alt-hough SAP supported other integration approaches for external applications, SAP had taken a data-centered approach to intra-application communication, with the resultant duplication of data.

The pre-XML approach to integration, although inherently simple, was implemented with fixed data struc-tures, which resulted in a fairly tightly coupled system in which all system components are affected bychanges in the data and in the data structures. This architectural approach often led to a single point offailure in the shared data store; historically, it was not time-critical in business systems, but in real-timework process application in operations, this approach corrupted operations systems and caused large effi-ciency and quality losses and cost.

3. Distributed Ownership of Data

With advances in networks from 1990 to the present, application complexities grew into distributed sys-tems built out of multiple communicating (integrated) applications. While early implementations weremostly an attempt to break monolithic architectures apart, they were still based on the single applicationmodel and designed for a specific domain.

Eventually, this led to a new class of systems in which multiple applications are integrated to create newfunctionality, and use data contained in each application to implement functionality not initially intendedby the original application designers. In this case, each application is a System of Record (SOR) – an infor-mation system that is the authoritative data source for a given data element or piece of information.

Problems arising from the distributed nature of these systems quickly led to the realization that distributedarchitectures are hard and brittle. A better way to communicate between systems was needed to handle




communications failures and issues with application availability. This led to the idea of loosely coupledmessage-oriented architectures.

4. Message-Oriented Architectures (MOA)

Message-Oriented Architecture (MOA) is for exchanging information (documents) where there are no im-plied semantics about what should be done with a received document.

What does this definition mean? It basically means that MOA is for broad-scale information sharing. Anexample is stock ticks, in which a financial services firm has an MOA backbone (e.g. ,TIBCO, MQ, or MSMQ)to distribute changes in stock values to any application that is interested. The MOA doesn't dictate whatsomeone does once they know a stock value has changed; the MOA just informs them that it has hap-pened. By this definition, MOA is primarily used for data synchronization and event notification. As a result,MOA is often publish/subscribe based.

By itself, MOA does not address any data model for the exchanged information. It moves data from point Ato point B with the requested quality of service. It does nothing to define the content, nor does it capturethe semantics of the system.

MOA can be extended to include structured data when the MOA expects a response. In that case, the datamodel for information to be exchanged is typically based on industry standards. Examples include EDI(Electronic Data Interchange), B2MML (XML implementation of the ISA-95 standard) and BatchML (XMLimplementation of the ISA-88 standard). The data model used can also be used for the data model that wediscussed in the “Data-Centered Architecture” section above.

5. Service-Oriented Architecture (SOA)

While MOA helped to ease communication complexities between applications in distributed systems, typi-cal systems were built by doing point-to-point integration, with information semantics being buried, hid-den within the implementation and not available at all at runtime.

SOA adds a standardized approach for interoperation between applications or application components.SOA adds a platform-independent description of the interface contract that is discoverable at runtime. Thisenables ad-hoc client access to exposed information, which was sometimes impossible within MOA with-out intimate knowledge of the data being exchanged and the structure of the messaging backbone. Unfor-tunately, not all information was exposed, but the concept did provide a degree of improvement. Withstrict design and governance oversight, the degree of improvement is very large.

Within SOA, the consumer (client) interacts with a provider for a few well-defined purposes (e.g., pro-cessing an order). Information is very task specific and doesn’t change often. When information doeschange, a new service is added to maintain “backward compatibility” with existing systems. An individualservice definition captures the semantics of a service and its data, but fails to deliver any knowledge aboutthe system.

6. Information-Oriented Architecture (IOA)

The preceding architectural approaches arguably complement each other and pragmatically make sense touse together due to the variety in the installed base. Indeed, any evolving architecture must provide a wayto efficiently coexist and/or integrate with existing systems. Evolution is needed. Revolution is simply notaffordable or realistic. IOA extends SOA to include a canonical view of, and access to, the information in thesystem being integrated to serve as the basis for business and operations intelligence and analytics in sup-port process optimization and enhanced decision making. IOA provides the market a foundation for com-posing business and operations processes that collectively create composite applications around a singleinformation model. That model defines the canonical form for data exchange and defines the canonicalside of data mediations. This differs from the Data-Centric Approach above in that the canonical model isthe interchange model and not necessarily the model of any given service. The challenge is that the datamodel is designed for a given set of questions. Reorienting it for other purposes requires governance rulesand processes that are similar to those of SOA.




Semantic Computing, on the other hand, defines data at an elemental level, so models for specific purpos-es are easily created without moving the data from the appropriate data store. Semantic Computing is notfor all applications, but there are many to which it is uniquely suited. A few key examples are 1) strong,agile governance; 2) live HAZOPs; and 3) data management throughout the design, procurement, construc-tion, hand-over, startup and commissioning phases and into the operations and maintenance phases. Giv-en that these example solutions address a relatively small set of people and a focused set of long runningfunctions and tasks, with Semantic Computing the usual scalability challenge is not an issue.

IOA typically includes MDM (Master Data Management) and BI and OI tools as a complement to SOA. In hisData Integration Blog post (http://www.dataintegrationblog.com/robin-bloor/heart-information-oriented-architecture-middleware/) Robin Bloor points out that IOA also may include a semantic data map to pro-vide context to the information being accessed in MDM and the integrated applications. This idea is con-sistent with the basic premise of this paper. However useful the previously described architectural ap-proaches have proven, they lack context for the information being acted on. SOA, combined with stand-ards based messages (e.g., OAGIS, B2MML, and BATCHML), provides the ability to create and integratecomposite processes and applications for services such as order management or production tracking.However, there is still no overlying context for the information that can be requested by client applications.

This context is provided in IOA by an overlying model of the real world that provides a context for infor-mation requests. This way, requests (and associated services, definitions of data etc.) are associated withan object in the model that defines its meaning and provides its context. As an example, a model can becreated for an industrial enterprise based on industry standards, such as ISA-95 and ISA-88, which is usedto define the enterprise hierarchy of an oil drilling platform. The model, at the lowest level of the hierar-chy, contains instances of equipment resources such as pumps or motors to which information requests,actions, and responses are associated. The association then provides the context to support queries suchas “find the available work orders for this pump,” “report the current temperature of this motor,” or “cal-culate the average value of pH in this tank over the last week.” With this addition of context to the IOA, theIOA behaves as Semantic Computing. In fact, ISO 15926, Industrial automation systems and integration—Integration of life-cycle data for process plants including oil and gas production facilities is one such stand-ard definition.

All of this information could have been obtained, one way or another, with any of the previously describedarchitectures. What Semantic Computing does is to introduce provider-independent context, rules andcontinuous governance into the discussion in a way that is meaningful to the industrial enterprise. Seman-tic Computing simplifies the task of accessing data and of associating meaningful actions with events relat-ed to the modeled objects.

7. Model-Driven Architecture

The discussion to this point has focused on the use of semantic models to support operations systems inte-gration and, arguably, the creation of composite/integrated applications via use of SOA, middleware, Se-mantic Computing and (where appropriate) a common information model. In Semantic Computing, multi-ple information models can be built as needed for specific needs to eliminate the need for a fixed commoninformation model. This might sound similar to what was discussed earlier as Model-Driven Architecture,but in reality it is very different. Model-driven architecture, explained in detail in Alan Brown’s excellentpaper “An introduction to Model Driven Architecture” is about using process models in the context of ap-plication design to drive the development of the application, perhaps including generation of the applica-tion code itself. The current paper’s discussion, in contrast, focuses on the ad hoc or focused use of mod-els, in conjunction with SOA and appropriate middleware, to act upon information provided in context,with a focused view of information available in the enterprise and independent of the SOR.

The Information Model – The Heart of the Semantic Model

The term Information Model (IM) is generally used for models of individual things such as facilities, build-ings, process plants, etc. An information model formalizes the description of a problem domain withoutconstraining how that description is mapped to an actual implementation in software. Simply put, an in-




formation model is a way to abstract disparate data. An information model is about describing what datameans (ontology) and where it fits. In the Semantic Computing context, that data can be organized to an-swer any specific question. Today, vendors usually use Semantic Computing concepts to address a specific,predefined set of questions in a limited set of views. An information model allows us to understand andabstract knowledge for its design intent.

ISA-95 is a great example of a domain-specific information model with the design intent of optimizing theoperations of industrial facilities by addressing the four operations activities discussed above (see TheManufacturing Complexity Problem). The ISA-95 standard properly states that its model structure does notreflect a specific business organizational structure within a company, but is a model of operations activi-ties. Different companies assign responsibilities for activities or sub-activities (functions and tasks) to dif-ferent business organizational groups. In other words, ISA-95 defines instances of data/objects and activi-ties that are in the manufacturing operations management (MOM) domain. ISA-95 activities do not assignwhich systems have to be in place or how these systems should be implemented or integrated. Informationmodels help us understand how disparate pieces of information relate to each other, but not how systemsshould be implemented

ISO 15926 Industrial automation systems and integration—Integration of life-cycle data for process plantsincluding oil and gas production facilities is another example of lifecycle information management forequipment, from concept through preliminary engineering, detailed engineering, procurement, construc-tion, handover, and into the operations and maintenance (O&M) phase, where ISA-95 also applies. ISA-95is complementary in the O&M phase of ISO 15926, which serves as the Engineering Management ofChange Database to keep the knowledge about all equipment and system configurations consistent. Suchconsistency management is presumed for ISA-95, but it is explicitly outside the scope of ISA-95.

A semantic model is one specific kind of information model. Semantic models help to identify patterns andtrends in information and to discover relationships between disparate pieces of information. A semanticmodel uses two important constructs:

Vocabulary: A collection of concepts given a well-defined meaning that is consistent across con-texts.

Ontology: A contextual relationship behind a defined vocabulary, the cornerstone of defining aspecific knowledge domain.

Semantic models consist of a network of concepts and of the relationships between those concepts. In theindustrial enterprise, the concept may be “Production Request”, “Batch” or “Equipment.” ISA-95 Part 1 is agreat example of a semantic vocabulary, defining well understood, domain-specific concepts.

The meaning of a concept is defined by its relationship to other concepts. A good example of this relation-ship from ISA-95 is “Equipment Class” and “Equipment.” Because equipment is semantically linked to anequipment class, it is possible, as a result, for systems and people in the domain to find all the equipmentof a specific class using the semantic relationships.

While there is no formal standard for relationships, semantic models often use the following definitions:

INSTANCE: x3456 is an INSTANCE of Batch

IS_A: Reactor IS_A Equipment

HAS_PART: Reactor HAS_PART Heater

As knowledge changes, the semantic model must evolve. For example, as new operations definitions(product) segments are defined, changes to the information model may be required by linking it to a spe-cific operations/process segment. Other qualitative data related to operations segments can be fed into acontinuously updated model, enriching the knowledge of patterns and influences.

In the Semantic Computing concept, such changes are "discoverable" and are automatically embedded intothe infrastructure.




Operations Lineage

The primary goal of information modeling is to provide an agile, adaptable, well understood methodologyand terminology for accessing data and context within SORs to enrich the knowledge and increase theadaptive-ness and responsiveness of each organization.

One important element of the context of information is its lineage—the ordered history of informationchanges. This is particularly critical in many analytical applications. Lineage helps answer questions such as:

What material lot was used in this batch?

When did we learn this?

Why was this decision made?

In the context of the BP disaster in the Gulf of Mexico:

When did BP decide to buy that design of Blow-Out Preventer?

When did the hydraulic system start to malfunction?

Why did BP decide to continue to operate without fixing it?

Lineage adds essential context that relates instances with the timeline of events and activities. Lineage alsoprovides information about the duration of process execution, be it related to equipment, material trans-formations, system performance or people’s responsiveness. For example, an ideal system captures dataat:

1. Creation within any SOR (request to make product is entered)2. Presentation as a task to an operator (told operator to go make product)3. State change from a piece of equipment indicating that the process has started (now making prod-

uct)

In this example, lineage provides in situ metrology for the performance of processes and for all partici-pants.

Because changes occur in multiple distributed systems, no individual SOR contains a complete change his-tory. Each SOR provides its piece of each puzzle in the form of historical data. Because of this, we are oftenleft with an incomplete and incomprehensible view of the past. No individual SOR will provide completelineage coverage and visibility across all SORs participating in manufacturing operations execution. Seman-tic Computing allows these items to be linked without altering any pre-existing SOR.

Operations Data Complexities

Unfortunately, a majority of the data that exist in the variety of SORs within the manufacturing enterprisedo not come in a nice, flat, table-like form. Data typically appear as an ugly mess of complex structures,hierarchies, collections and arrays. While computers can handle it easily, people can’t easily comprehendthis type of information. Even worse, most of the reporting and BI tools on the market today are designedto work with flat, table-like sources of data. Semantic Computing simplifies this problem by suggesting thata wrapper be placed around each SOR to expose the underlying data for generic access. This approachleaves the SOR “as is” while providing broader access to the data for enhancing corporate performance.

The OPC standards, for example, define different specifications for data access, alarms, events, complexdata, historical data etc., each with its own information model to capture and provide context that is im-portant for the intended use. Even a simple analog item, a pressure sensor for example, is an object withattributes and references to multiple other objects (Figure 3) that provide information that may be im-portant, such as the instrument range or engineering units.




Figure 3: OPC UA Analog Item Representation

Using ISA 95-defined objects as another example, we can see that while Equipment Capability (Figure 4) isa complex object that contains a collection of Equipment Capability Properties, only a few of them may beneeded for a query at any time. Furthermore, each Equipment Capability Property itself is complex object,with multiple data elements.

Figure 4: ISA-95 Operations Capability Model




Semantic modeling helps to define these structures by providing information that can be used by users andtools to extract required information. It is easy to imagine a user or system issuing a query for all Equip-ment Capability Properties for “Capacity” by selecting “Equipment Name” and Equipment Capability Prop-erty’s “Value” and “Value Unit of Measure” attributes.

Historically, depending on the underlying implementation technology, all data may be accessed throughsingle or multiple transactions with the SOR. The format of the data varies from XML documents (B2MML,BatchML) and binary objects (OPC, Tibco, MSMQ) to Web URIs (generic version of URL), and anything inbetween. Establishing communication with these sources of data is not enough; application-specific for-mats must be transformed into a common, standards based form that is well understood and can be usedby all clients. Without vocabulary provided by the model and associated industry standards, this knowledgeis locked in the source SOR.

Now, OPC provides semantic access to the data and various operations on the data, for example, semanticdetermination of average, min, max, etc. over a specified time frame. Also, the quality of the values is in-herently managed.

Semantic Models

This section details semantic models in an example model server (see Model Servers below) deploymentapproach.

As defined by the World Wide Web Consortium (W3C), Semantic Computing “provides a common frame-work that allows data to be shared and reused across application, enterprise, and community boundaries.”While the World Wide Web has generally been about the ability to share documents, Semantic Computingprovides the framework so that individual data elements can be shared and more readily interrogated andunderstood by machines. Semantic Computing supports the notion of common formats for data that canbe presented from a variety of different sources. It also provides the structure for understanding the datarelationships. This supports the interrogation of Web based data based on semantic meaning rather thanrelying on explicit (or implicit) links and references.

The Semantic Computing architecture, as defined by Tim Berners-Lee, is a layered structure with an XMLfoundation for namespace and schema definitions to support a common syntax. The next layer above theXML foundation supports a Resource Definition Framework (RDF) and RDF Schema (RDFS). RDF is a frame-work for a graphical representation of resources. Although it was created to represent information aboutWeb resources, it can be used for a variety of other data types as will be discussed later. The core defini-tion of an RDF element is based on triples in subject-predicate-object form. The machine-readable formatfor RDF is XML (RDF/XML).

An RDF model essentially defines a graph as described through triples. An RDF Schema (aka RDF Vocabu-lary Description Language) provides additional knowledge to the RDF such as the terms that can be used,restrictions that apply and what additional relationships exist. An RDF Schema can be created to describethe taxonomy of classes (as opposed to just resources in RDF) and formalized relationships between re-sources (typing and sub-classing) to define simple ontologies. More complex ontologies can be created us-ing Web Ontology Language (OWL). The ontology vocabulary is the next layer in the Semantic Computingarchitecture.

As mentioned earlier, an ontology provides an understanding of concepts (terms and relationships) withina domain through a defined vocabulary and a model taxonomy. Within a specific industry domain, an on-tology can be used to support multiple applications. Additionally, an ontology could be created to supportgenerally applicable terms and relationships that can span multiple domains. Ontologies define entities andrelationships to represent the knowledge that can be shared across industries, domains and applications asappropriate. In order to facilitate this, ontologies support, a well-defined model of property inheritance.OWL produces this more expressive semantics and to specify mechanisms by which the language can pro-vide "more complex relationships between entities including: means to limit the properties of classes withrespect to number and type, means to infer that items with various properties are members of a particularclass, a well-defined model of property inheritance, and similar semantic extensions to the base languages.




Therefore, more generalized knowledge can be captured (referred to as upper ontologies), which can thenbe further refined to support a specific domain (domain ontologies).

Semantic understanding of data depends on a common vocabulary that defines terms and relationships.RDF Schema provides a framework for a vocabulary that supports typing and sub-typing and the ability todefine data types. More detailed ontologies can be created using OWL, which relies on RDF Schemas butprovides additional language terms in its own namespace. OWL is defined through species or profiles.Providing profiles that restrict the use of terms can make implementations simpler by including the infer-ence engines that can be used. (Inferencing and inference engines (reasoners) are discussed later in thisarticle.) OWL Lite (for users primarily needing a classification hierarchy and simple constraint features) canbe used for taxonomies and simple constraints, OWL DL (so named due to its correspondence with descrip-tion logics) can be used for full expressiveness and OWL Full can be used for no expressiveness constraints.

The SQL Protocol and RDF Query Language (SPARQL) is an SQL-like language for querying RDF (includingRDF Schema or OWL). SPARQL is used to query RDF graph patterns and return results from selected sub-graphs. SPARQL can be used for querying ontologies as well as instantiated model data. SPARQL results canbe presented in traditional form for use in Excel, MS SQL Reporting Service etc. SPARQL is the vehicle foraccessing highly dispersed data to leverage traditional analysis tools.

ISO 15926 is an example of a specific implementation of OWL, etc. for the oil & gas industry’s Life CycleData Management.

Model Servers

This section explains the role of the model server as a run-time host for the semantic model.

The model server (or model manager) provides the run-time framework on which the model is deployed. Amodel server must support a number of key functional services to persist and manage the model (ontolo-gy) and also the model instance data. It must also provide tooling and application interfaces for model andinstance data queries and updates. Next, the role of the model server provides the execution capability forthe semantic models discussed above. This capability is provided through open source projects such as Je-na, Joseki, Sesame and Pellet.

Model servers support a number of different persistence layers that include database and file (typically inRDF/XML format, although N3 and Turtle are two other popular notations). Although relational databasescould be used to support RDF data persistence, querying of RDF data (graph based data) stored in a rela-tional database (RDB) is often inefficient and the ability to change the data model without changing thedatabase schema can be lost. A triple store is a special purpose database designed specifically for storageand querying of RDF data. The data structure is optimized for data stored in a triples structure, which cor-responds to the RDF subject-predicate-object form. Both Jena and Sesame provide triple stores.

When we think about model servers at this level, there is not yet a requirement to understand the struc-ture of the persisted data. However, as additional model server functionality is considered, an understand-ing of the data becomes relevant. Jena and Sesame both provide good examples.

First, we should note that Jena provides “a Java framework for building Semantic Computing applications”rather than providing a complete model server. Specifically, Joseki, an open source sub-project to Jena,provides server capability providing an HTTP interface to the RDF data as well as an interface for SPARQLquerying and updating. In addition, Jena provides a programming interface to the RDF data as well as aninference engine. With this additional capability, Jena does need to “understand” the RDF ontology. Rea-soning or inferencing means being able to derive facts that are not directly expressed by the ontology.

Jena provides an inference engine to support reasoning in RDF, RDFS, and OWL but in some instances areincomplete. Jena provides a pluggable interface so that additional inference engines can be integrated. Forexample, Pellet is an open source Java reasoner that fully supports OWL DL and can be plugged into Jena.With this type of extensibility, Jena supports languages such as RDFS and OWL and supports data inferencefrom instance data and class descriptions.




Like Jena, Sesame provides a Java framework that supports persistence, an interface API, and inferencing.However, the inferencing capability within Sesame supports RDF and RDFS but not OWL. For a set of RDFor RDFS data, you can query Sesame and find the implicit information. Since anything that can be inferredcan also be asserted, one approach to supporting inferencing is to explicitly add the implicit information tothe repository as the data is initially created. This is the Sesame approach.

Semantic Models and Model Management Middleware

This section addresses the semantic model services provided through model management middleware.

The purpose of model management middleware is to provide a framework that makes it much simpler tocreate applications that are centered on a semantic model of the real world and that support integration ofreal-time operational data and related enterprise applications. A key component of a model managementmiddleware framework is the base set of services to deploy, run and manage the semantic model based onthe types of model services previously described. In addition, the middleware framework should support aclass of model ontologies and their instantiations. For example, the ontology could be based on industrystandards (such as ISA-95, ISA-88 and ISO 15926) and support the definition of an enterprise model downto assets, every materials transformation and the associated measurements.

Another key component of the framework architecture is support for model-aware adapters that supportintegration of various types of endpoints (OPC, databases, and Web services accessible applications such asCAD), and the mapping of the information flowing between those endpoints and elements of the model.

Therefore, there are two views provided by the middleware framework:

1. The reference model (i.e., the ontology) defines the classes that exist in the model and the rela-tions between them but does not correspond to any particular enterprise or asset.

2. The instantiated model defines instances of the classes that have a direct reference to real-worldentities. They are populated with a set of properties (e.g., s/n, location, temperature) and with re-lationships with other instantiated entities in the model.

As an example of how industry standards (ISA-88/95, OAGIS, MIMOSA, etc.) can be used to model the realworld, consider the following example, which is based on a project for manufacturer of paint.

In Figure 5, classes from ISA-95: Enterprise, Site, Area and Production Unit are instantiated. These, alongwith an additional Work Equipment class, are used to define a physical model starting from an enterpriselevel down to the level of specific pieces of work equipment.

At the work equipment level, typically, measurement classes are attached and mapped to endpoint dataadapters and specific data sources.

Once the model has been instantiated and mapped to endpoints through the adapter layer, the model cannow be used in a number of ways to achieve the business benefits previously described.

In the example paint manufacturing enterprise, applications that need to obtain information about an as-set, such as a tank, now go to a single location—the model server hosting the instantiated model—to ac-cess that information. This includes real-time information on the tank (e.g., temperature), historical infor-mation (e.g., average temperature this week), or more complex types of information (e.g., open work or-ders for this tank, or tanks of this type). The model server provides:

The queries made by applications to get operations information about the tank are made using aconsistent interface method (e.g., SPARQL) regardless of the true source of the information such asSCADA systems, an operations database, or an application such as SAP or Maximo.

The representation of the tank and the enterprise hierarchy around the tank is consistent and isbased on industry standards (ISA-95, etc.). The canonical form is maintained regardless of the un-derlying format being used in the endpoint systems.

The tank information is easily extended now to introduce new information that is deemed to beuseful in the future. For example, a new requirement to relate to equipment failures in an external




asset management system is easily tied to equipment in the model so that the failure informationis queried through the same model context. The model also provides a “canvas,” based on real-world context, to simplify configuration for aspects of production control such as calculation ofKPIs (Key Performance Indicators), definition of actions needed for operations events and genera-tion of alerts for detected problems. The type of information is now able to be associated with anobject in the model and then easily made sensitive to context in the model.

Likewise, the relations in the semantic model now make it much easier for applications to look atthis information across the model laterally to answer questions that were not anticipated in the in-itial creation of the model. As an example, a paint enterprise contains similar types of motors thatcan serve the same function, but which come from different suppliers. Through relations in themodel such as “Motor type A <is equivalent to> Motor type B,” a report is easily produced showingperformance characteristics of all the similar types of motors currently being used in production(across locations, if need be) so better supplier decisions are made in the future. In doing so, thisanalysis concludes that a maintenance action is required to replace one type of motor because an-other type is performing much better. Note in this example that the relations showing equivalencyneed not have been in the originally implemented and deployed model. These could be added laterbased on new knowledge.




Figure 5: Enterprise Hierarchy Based on Industry Standards

In summary, the framework for extending the capability of application integration based on a semanticmodel includes:




Model operations entities: Model operations entities (e.g., Tanks, Pumps) and their relationships tosupport data queries, which may be contained in a number of different systems in a real-worldcontext. This is a powerful concept and it allows us to establish intelligence across the entities (andthe underlying systems) to support analytics and optimization aimed at things such as failure pre-diction, detection of abnormal behavior and detection of and prevention of product quality prob-lems.

Establish global namespace: Establish a common naming definition and information access methodso that an application references entities such as assets that may be named and identified differ-ently by multiple enterprise subsystems in a way that protects the application from knowing thedetails of those subsystems (for example, SCADA/DCS Systems, OPC Servers, SAP or Maximo).

Define canonical form: Define a canonical form to reference information associated with opera-tions entities in the enterprise. For example, a tank being used for the mixing of paint might havetemperature information that can be obtained from lower level OPC servers, or work orders thatcan be obtained from SAP or Maximo. As was previously mentioned, industry standards can beused to supply definitions for that canonical form, which has the advantage of building on accepteddefinitions and vocabulary for common entities such as equipment, locations, personnel etc.

Provide enterprise application interface: Provide a global interface for applications to query andupdate operations entities and their associated data so that the application does not need to knowwhich subsystem owns any given entity or associated data (e.g., OPC servers, SAP or Maximo aswas previously mentioned). The application is provided with a full enterprise view of the data,based on the model of the real world that corresponds to the data. This makes addition of new un-derlying systems much simpler, since the specifics of that are hidden behind the model.

Provide the means to coordinate change across all SORs and verify the configuration consistencyacross all SORs by federating the configurations to the approved set of configurations.

Calculate the current risk level in the facility based on the current values for the HAZOP definition.

Example: Manufacturing Operations and Business Intelligence within Level 3 WorkflowsMapped to Level 4

Operations processes that exist in the manufacturing enterprise clearly define specific ontologies that arerelevant to the domain.

For example, an asset management SOP that is changing the state of the asset links asset data, changereasons and many other information nuggets together. For operations personnel using this SOP, the infor-mation that it is operating on is well understood. Analysis of the process definition quickly produces enti-ties and relations as defined above, but in localized, domain-specific semantic context.

Not only does the process interact with multiple systems, it also has a definite timeline and sequence ofevents and invocations of activities. Every touch point produces an entry for our lineage of information,while workflow itself defines the ontology of the domain it is operating in.

As semantic modeling and lineage are analyzed, the processes that already exist in the manufacturing en-terprise provide a rich source of the metadata necessary to define domain-specific semantic models. Withthese models captured in a computer-executable form and connected to SORs as required, a self-extendingsemantic information model is able to be applied. Every new process definition extends the model withnew definitions and relations.

An important side effect of this approach is that data from a single SOR is contextualized by every workflowthat is utilizing it. Often different ontology added to the same data from a system, based on the usage inthe operations process.

Similarly, if every workflow intelligently captures the transient data of its execution, rich lineage infor-mation is created.




By combining these two sources of metadata and by layering querying capabilities on top of the underlyingsystem, a powerful source of manufacturing BI and OI is created. Metadata from process definitions de-scribes entities and relations and can be presented to users in such way as to help them with query build-ing activities utilizing implied context from business and operations processes.

Putting It All Together

The following example is based on a batch management system built on a model-based platform and utiliz-ing a number of modeling techniques to provide the ability to easily add and adapt functionality by peoplewith various levels of expertise and needs.

In this example, the platform’s model is a combination of two different modeling techniques:

1. Data modeling: Defines data that represents abstract, domain-specific concepts. In short: vocabu-lary is defined that forms a foundation of the data modeling.

Data modeling allows for industry standards as ISA-95 to coexist with localized, business-specific conceptsor even system and application-specific definitions.

2. Service modeling: Defines services or activities that use data from the data model to interact withSORs and people.

Service modeling provides strongly-typed definitions that are consumed by service implementers who in-tegrate with various SORs.

All activities defined in a model translate directly into the domain-specific building blocks used by graphicaltools built for domain experts.

Once defined, a model (Figure 6) forms the foundation of the system, providing building blocks for devel-opers who implement services, process designers who build workflows, and information modelers whocreate queries for extracting relevant information about manufacturing operations performance.

Model definitions are presented as high level, domain-specific building blocks to the process designers,providing the benefit of implicit context definition that forms the foundation of the platform semanticmodel used to access operations information.

As new processes are defined and deployed to the system, the information model is updated to include thecontext implied from the process definition, combined with concepts from the core information model ofthe system.




Figure 6: Model Designer with ISA-95 Concepts

Figure 7 represents a simple SOP that runs every time a lab sample of a batch is taken. This SOP is triggeredby schedule (every one hour after Setpoint 1 is reached) or by the event from the equipment when Set-point 2 is reached. Following the process, the system locates the current equipment used by the batchfrom the Batch Management System. It then loads specifications from the product lifecycle management




(PLM) system to use as instructions for the next step, which is performed by an operator. The next stepissues the work instruction to an operator with information from Batch and PLM to physically take a sam-ple. Because this step is performed by a person it may take a long time, but eventually the operator indi-cates that the sample is ready and has been delivered to the lab. In the succeeding step, the system issuesa work instruction to a lab technician to check and validate that the sample was received. The last steprecords the sample in the laboratory information management system (LIMS), receiving a Sample ID as-signed by the LIMS to the sample. In the succession of this handful of steps, this process interfaced with:

Batch Management System

ERP/PLM

LIMS

PLCs

People

Figure 7: Lab Sample Acquisition SOP

This simple process captures a number of elements of the ontology and the very important lineage of theinformation. Figure 8 represents some of the concepts and the relationships that could be extracted fromthis process. Regardless of where data comes from, queries can instantly be built for the system, such as:

What equipment was used to produce a batch?

Who took the sample?

How the sample was taken?




What specification was used for the batch?

Figure 8: Lab Sample Acquisition SOP Ontology

Let’s now look at the lineage information that is produced. Every time the process runs, the execution ofevery step is recorded as well as all of the information that is passed in and out of activities, including tran-sient information that doesn’t exist in SORs. For example, a “Sampling Procedure” instruction is generatedand presented to the operator, based on the data from PLM. It is not stored in any system of record in-volved with the process. Over time, the PLM data may change, thus hampering our ability to understandwhat exactly was presented to the operator. With lineage data in place, users can ask the system the fol-lowing questions:

When was the sample scheduled to be taken?

When did the operator actually take the sample?

What sampling procedures were used in the last month of operation?

How long, on average, does it take for operators to respond to a sample request?

Figure 9 shows data returned by a query called “Lab Testing Time.” This example query is extracting datausing both semantic ontology and lineage of data to prepare this result. It is using concepts and the rela-tionships from two different SOPs; one was reviewed above and another one runs when lab results areready. This query is using vocabulary knowledge to extract required fields from complex internal objectsand to present them as a simple table that is easy to understand.

In the example in Figure 9, the Open Data Protocol (OData) standard is used to expose complex, unstruc-tured execution data to people and BI applications. The Open Data Protocol is a Web protocol for queryingand updating data that provides a way to free data from the silos that exist in applications today. ODatadoes this by applying and building upon Web technologies such as HTTP, Atom Publishing Protocol (Atom-Pub) and JSON to provide access to information from a variety of applications, services and stores. The pro-tocol emerged from experiences implementing AtomPub clients and servers in a variety of products overthe past several years and is supported by multiple vendors.




Figure 9: Lab Sample SOP – Lab Testing Time Data Source

Once the OData Service is defined, it acts like any other data source that can be queried, supporting com-plex Where statements, ordering and other SQL features. It supports a variety of clients to extract andwork with the information that doesn’t exist in any of the individual SORs.

This example demonstrates how traditional SOA modeling is combined with semantic models inferred fromthe executable business processes to unlock and contextualize unique information, normally not availablefrom the individual systems of record. Use of the information model simplifies end user interaction withthe system and creates a data source for BI tools.




Conclusion

The value of semantic models and Semantic Computing as key components of integration frameworks wasexplained in building solutions. This architecture was discussed in the context of a number of widely usedand well known solution architectures that center on data, messaging and services. Semantic models weredescribed in general terms and then illustrated, through a detailed example, to show how a semantic mod-el based integration framework acts as the foundation on which to build solutions that drive business in-sights and efficiencies.

Semantic models play a key role in the next step in the evolution of solution architectures that support thebusiness goals of obtaining a more complete view of what is happening within operations and in derivingbusiness insights from that view. Semantic models and Semantic Computing based applications based onindustry standards such as ISO 15926 and ISA-95 and all its related standards, take that one step further,especially as application vendors adopt those standards (which, as always, will happen more rapidlythrough pressure from the user community or as a result of competitive pressures).




Bibliography

1. IBM Integrated Information Corehttp://www-01.ibm.com/software/info/integrated/

2. IBM Integrated Information Core - Information Centerhttp://publib.boulder.ibm.com/infocenter/iicdoc/v1r4m0/index.jsp

3. IBM Service Oriented Architecturehttp://www-01.ibm.com/software/solutions/soa/

4. W3C Web Services Architecturehttp://www.w3.org/TR/ws-arch/

5. Open Applications Grouphttp://www.oagi.org/dnn2/

6. RTC – Data-Oriented Architecturehttp://rtcmagazine.com/articles/view/100926

7. United Nations Centre for Trade Facilitation and Electronic Businesshttp://www.unece.org/cefact/ebxml/CCTS_V2-01_Final.pdf

8. Ontologies and Semantic Computinghttp://www.obitko.com/tutorials/ontologies-semantic-web/ontologies.html

9. Jena – A Semantic Computing Framework for Javahttp://jena.sourceforge.net/

10. W3C - RDF Vocabulary Description Language 1.0: RDF Schemahttp://www.w3.org/TR/rdf-schema/

11. W3C – RDF Semanticshttp://www.w3.org/TR/rdf-mt/

12. Fern Halper on Semantic Modelshttp://fbhalper.wordpress.com/2007/11/29/whats-a-semantic-model-and-why-should-we-care/

13. OMG – Catalog of Data Distribution Services (DDS) Specificationshttp://www.omg.org/technology/documents/dds_spec_catalog.htm

14. Robin Bloor on Information-oriented Architecturehttp://www.dataintegrationblog.com/robin-bloor/heart-information-oriented-architecture-middleware/

15. Alan Brown, An introduction to Model Driven Architecturehttp://www.ibm.com/developerworks/rational/library/3100.html

16. Savigent Catalyst xMhttp://savigent.com/products/catalyst-xm/

17. Model-based systems in manufacturinghttp://savigent.com/casestudies/

18. The Enterprise Service Bus, re-examinedhttp://www.ibm.com/developerworks/websphere/techjournal/1105_flurry/1105_flurry.html




Glossary, Acronyms and Abbreviations

TERM/ACRONYM/ABBREVIATION DEFINITION

API Application Program Interface(s)

AtomPub Atom Publishing Protocol

B2MML Business To Manufacturing Markup Language. An XML schema document is an XML specifica-tion file used to describe and enforce XML data structures. The B2MML schema is used in con-junction with XML tools enforce that a B2MML XML file has implemented the B2MML data def-initions correctly. The schema can be used to validate B2MML common data definition usageacross ERP, supply chain management systems and manufacturing systems

BI Business Intelligence. Category of applications and technologies for gathering, storing, analys-ing and providing access to data.

CAD Computer Aided Design. Highly specialized graphical software used to create 2-D and 3-D engi-neering specifications, blue prints etc. sometimes including parts lists and other relevant data.

DOT Department of Transportation

EDI Electronic Data Interchange

ERP Enterprise Resource Planning. A planning system for manufacturing, order entry, accounts re-ceivable and payable, general ledger, purchasing, warehousing, transportation and human re-sources. Organization-wide computer software system used to manage and coordinate all theresources, information and functions of a business. Example: SAP.

HAZOP HAZard and OPerability study . A structured and systematic examination of a planned or exist-ing process or operation in order to identify and evaluate problems that may present risks topersonnel or equipment, or prevent efficient operation. The HAZOP technique was initially de-veloped to analyze chemical process systems but has been extended to other types of systemsand also to complex operations and to software systems. A HAZOP is a qualitative techniquebased on guidewords and is carried out by a multi-disciplinary team (HAZOP team) during a setof meetings.

HTTP Hypertext Transfer Protocol

IM Information Model

IO Input/Output

IOA Information-Oriented Architecture

IOM Inventory Operations Management

ISA-95 ISA Standards, ISA-95: An International Society of Automation (ISA) Consensus Committee de-fining standards for enterprise and manufacturing integration. ISA-95 is the internationalstandard for the integration of enterprise and control systems. It is used to determine whatinformation has to be exchanged between finance and logistics systems, along with production,maintenance, inventory and quality systems. See also Manufacturing Operations Management.

JSON JavaScript Object Notation is a text-based open standard designed for human-readable datainterchange. It is derived from the JavaScript scripting language for representing simple datastructures and associative arrays, called objects. Despite its relationship to JavaScript, it is lan-guage-independent, with parsers available for many languages.

LIMS Laboratory Information Management System

Maintenance Op-erations Manage-ment (MnOM)

Maintenance Operations Management is defined as the collection of activities that coordinate,direct, and track the functions that maintain equipment, tools and related assets to ensuretheir availability for manufacturing and to ensure scheduling for reactive, periodic, preventiveor proactive maintenance.

ManufacturingOperations Man-agement (MOM)

MOM is used in this paper as the basis for the development of standard interfaces betweenEnterprise Resource Planning (ERP) systems and Manufacturing Enterprise Systems (MES).MOM decomposes its models into global, strategic, tactical and atomic functions. This func-tional decomposition is used as the basis of the Service lifecycle by identifying Services candi-dates.

MDM Master Data Management

MES Manufacturing Execution /Enterprise System

MnOM Maintenance Operations Management

MOA Message-Oriented Architecture




MOM Manufacturing Operations Management

MQ Message Queue

MSMQ Microsoft Message Queuing, a Message Queue implementation developed by Microsoft anddeployed in its Windows Server operating systems since Windows NT 4 and Windows 95. Thelatest Windows 7 also includes this component.

O&M Operations and Maintenance

OAGIS Open Application Group Integration Specification

OData Open Data Protocol (OData) standard, used to expose complex, unstructured execution data topeople and BI applications. The Open Data Protocol is a Web protocol for querying and updat-ing data that provides a way to free data from the silos that exist in applications today.

OI Operations intelligence

OWL Web Ontology Language

PLCs Programmable Logic Controller(s)

PLM Product Lifecycle Management (PLM): the process of managing the entire lifecycle for a productfrom conception through design, manufacture, service and disposal.

POM Production Operations Management

QMS Quality Management System

QOM Quality Test Operations Management

RDB Relational database

RDF Resource Definition Framework

RDFS Resource Definition Framework Schema

RFC Remote Function Calls

SCADA Supervisory Control and Data Acquisition

Semantic Compu-ting

From Wikipedia: Semantic computing is a field of computing that combines elements of seman-tic analysis, natural language processing, data mining and related fields.Semantic computing addresses three core problems:1.Understanding the (possibly naturally-expressed) intentions (semantics) of users and express-ing them in a machine-process-able format2.Understanding the meanings (semantics) of computational content (of various sorts, includ-ing, but is not limited to, text, video, audio, process, network, software and hardware) and ex-pressing them in a machine-process-able format3.Mapping the semantics of user with that of content for the purpose of content retrieval,management, creation, etc.

Semantic DataModeling

From Wikipedia: Semantic data models.A semantic data model in software engineering has various meanings:1. It is a conceptual data model in which semantic information is included. This means that themodel describes the meaning of its instances. Such a semantic data model is an abstraction thatdefines how the stored symbols (the instance data) relate to the real world.2. It is a conceptual data model that includes the capability to express information that enablesparties to the information exchange to interpret meaning (semantics) from the instances, with-out the need to know the meta-model. Such semantic models are fact oriented (as opposed toobject oriented). Facts are typically expressed by binary relations between data elements,whereas higher order relations are expressed as collections of binary relations. Typically binaryrelations have the form of triples: Object-RelationType-Object. For example: the Eiffel Tower <islocated in> Paris.Typically the instance data of semantic data models explicitly include the kinds of relationshipsbetween the various data elements, such as <is located in>. To interpret the meaning of thefacts from the instances it is required that the meaning of the kinds of relations (relation types)is known. Therefore, semantic data models typically standardise such relation types. Thismeans that the second kind of semantic data models enable that the instances express factsthat include their own meaning. The second kind of semantic data models are usually meant tocreate semantic databases. The ability to include meaning in semantic databases facilitatesbuilding distributed databases that enable applications to interpret the meaning from the con-tent. This implies that semantic databases can be integrated when they use the same (stand-ard) relation types. This also implies that in general they have a wider applicability than rela-tional or object oriented databases.




Service A Service is a discrete reusable unit of business functionality. It leverages a remote method orfunction rather than data or information through a well-defined interface. Services can beshared within and outside of the business organization. It avoids business functionalities dupli-cation and the leverage of application silos.

Service Implemen-tation

Resource providing functionality and implementation details of a Service.

Service OrientedArchitecture (SOA)

SOA is an architecture style for organizing and utilizing the discrete functions (services) com-bined in the enterprise applications into interoperable, standard-based Services. These servicescan be combined and reused quickly to meet business needs. It allows enterprises to sharecommon application Services as well as information. SOA facilitates the integration and sharingof business Services by providing the required infrastructure tools and Services.

SKU Stock Keeping Unit

SOA Service-Oriented Architecture

SOP Standard Operating Procedures

SOR System of Record – an information system which is the authoritative data source for a givendata element or piece of information

SPARQL SQL Protocol and RDF Query Language

Strategic Function The set of methodologies that describes the difference in the process of extracting the addedvalue from the enterprise practice.

Tactical Function The set of techniques applied to extract the added value from the strategically direction of theenterprise and mapped to asset, organization and technology.

TIBCO TIBCO Software Inc. (NASDAQ: TIBX) is a provider of infrastructure software for companies touse on-premise or as part of cloud computing environments. Whether it's efficient claims ortrade processing, cross-selling products based on real-time customer behavior, or averting acrisis before it happens,

UML Unified Modeling Language

W3C World Wide Web Consortium

WMS Warehouse Management System




Authors

Dave NollerSenior Certified IT Architect, Manager of IBM SWG Mfg. Industry [email protected]

Tim HanisChief Architect - Integrated Information Core (IIC), Senior Technical Staff Member - Industry [email protected]

Michael FeldmanCTOSavigent [email protected]

Charlie GiffordChief Manufacturing Consultant21st Century Manufacturing Solutions [email protected]

Contributing Editors

Jimmy AsherEngagement ManagerSavigent [email protected]

Bill Bosler, P.E.PrincipleTexas Consultants, Inc+7-967-053-27-54, Skype/LinkedIn: [email protected]




Reviewers

Mark AlbanoSolution ConsultantHoneywell Process [email protected]

Scott P. BogueTechnical [email protected]

Eyad A. BuhulaigaSystems Integration Engineer Saudi Aramco+966 (3) [email protected]

Technology

B3.0 c2 role of semantic models in smarter industrial ops mom c 3 reprinted with permission 20130325