A pattern-based approach to ontology-supported semantic data integration

Andreas A. Jordan Wolfgang Mayer Georg Grossmann Markus Stumptner

Advanced Computing Research Centre University of South Australia, Mawson Lakes Boulevard, Mawson Lakes, South Australia 5095, Email: [email protected] Email: (wolfgang.mayer|georg.grossmann)@unisa.edu.au, [email protected]

Abstract RSM2, ISO 159263 and Gellish4 standards are all used to represent information about engineering assets and Interoperability between information systems is associated maintenance information. However, there rapidly becoming an increasingly difficult issue, in exists considerable variation in the scope and level of particular where data is to be exchanged with exter- detail present in each standard, and where the infor- nal entities. Although standard naming conventions mation content covered by different standards over- and forms of representation have been established, laps, their vocabulary and representation used vary significant discrepancies remain between different im- considerably. Furthermore, standards like ISO 15926 plementations of the same standards and between dif- are large yet offer little guidance on how to represent ferent standards within the same domain. Further- particular asset information in its generic data struc- more, current approaches for bridging such hetero- tures. Heterogeneities arising from continued evolu- geneities are incomplete or do not scale well to real- tion of standards have further exacerbated the prob- world problems. lem. As a consequence, tool vendors formally support We present a pragmatic approach to interoperabil- the standard, yet their implementations do not inter- ity, where ontologies are leveraged to annotate proto- operate due to different use of the same standard or typical information fragments in order to facilitate varying assumptions about the required and optional automated transformation between fragments in dif- information held in each system. Not all standards ferent standards and representations. We discuss our are supported by formal ontologies that would allow approach based on an example drawn from the oil and one to overcome these heterogeneities automatically. gas industry. Where formal models are present, their focus is typ- ically on describing individual entities and relation- 1 Introduction ships, whereas dealing with issues arising from their use in larger representation structures, and scalabil- A growing number of organisations are interested in ity problems stemming from application of inference sharing information between their internal informa- mechanisms, are often left aside. tion systems and externally with partners or the pub- The aim of our work is to build a bridge that lic sector. A core issue facing these organisations ties together the heterogeneous standards and pro- is the information management systems they employ vides automated translation between their informa- have been built without supporting a common stan- tion models and concrete implementations. Our hy- dard for the underlying data models used for storing pothesis is that this goal can be achieved only if the information and their interfaces. Consequently, creat- information representation fragments used by differ- ing custom-built software that translates between dif- ent tools and standards are captured such that pos- ferent interfaces and data models is a predominantly sible translations between equivalent fragments can manual task, and current solutions for bridging het- be inferred automatically. To resolve the interoper- erogeneities between information systems do not scale ability problem not only with respect to terminologi- well to real-world problems. cal heterogeneities but also with respect to heteroge- Although standardisation of domain models, in- neous usage patterns by different organisations, exist- terfaces, and information representation can help to ing standards must be complemented with a formal reduce this burden, standardisation alone has been in- model of the elementary “information fragments” and sufficient to overcome the interoperability challenge. possible use of these fragments within each standard This is predominantly due to the fact that the stan- and organisation. Whereas approaches for translating dards themselves are heterogeneous and sometimes a semi-formal model like Gellish into natural language suffer from ambiguity. In the Engineering domain exist, we aim to extend these ideas to formal com- alone, there exist several competing standards and plex structures. This requires a formal model of the de-facto standards set by large tool vendors. For ex- building blocks, henceforth “information fragments”, 1 in the source and target representation and mech- ample, in the oil and gas industry, the MIMOSA , anisms to translate between “similar” fragments in Copyright c 2011, Australian Computer Society, Inc. This pa- different standards. per appeared at the Seventh Australasian Ontology Workshop For this purpose, we intend to construct an on- AOW 2011, Perth, Western Australia. Conferences in Research tology that reconciles the concepts and relationships and Practice in Information Technology (CRPIT), Vol. 48, in each standard into a common model and formally Vladimir Estivill-Castro and Gillian Dobbie, Ed. Reproduc- express the information fragments found in each stan- tion for academic, not-for-profit purposes permitted provided this text is included. 3http://15926.org/home/tiki-index.php 1 http://www.mimosa.org/ 4http://www.gellish.net/ dard in terms of the common model. Our focus will be on representing information fragments and their con- Upper Ontology crete representation in different standards such that A A equivalent representations in different standards can Information MIMOSA Fragment ISO be constructed automatically. Data Model B Ontology B Data Model We seek to investigate formal models and infer- 3 MIMOSA MIMOSA Fragment ISO ISO ence procedures to determine which are adequate RDL Templates 2 instances 4 Templates RDL for capturing the intent and composition of concrete 1 5 data structures on each end of the translation pro- MIMOSA Data ISO Data cess while reusing as much as possible existing on- tologies capturing domain-specific entities and their relations. Whereas (domain-specific) ontologies pro- Figure 1: Schematic of the overall translation process vided by several standards exhaustively capture do- on the example of translation between the MIMOSA main concepts and relations, we intend to develop and ISO 15926 standards. mechanisms for reasoning about translations between different representations. The resulting ontology will help to build adequate unifying ontology. Each fragment can be seen as a translation systems by providing developers of in- parametric statement that, when instantiated with tegration tools a catalogue of possible translations suitable values and objects for its parameters, ex- between different systems, either at run-time or at presses a certain meaning in the concrete representa- design time, when translations for information frag- tion corresponding ontology. By combining fragment ments in the source representation must be formed. instances representing template instances in the tar- Our ontology will also be able to verify completeness get representation, suitable translation for a given set of a given translation by identifying fragments and of template instances in the source representation can constraints in the source model that cannot be trans- be devised. lated precisely into the target model, and fragments that have ambiguous or multiple conflicting transla- Ontology of Information Fragments. Central tions. to our approach is the unifying ontology of informa- Our research will be predominantly driven by a tion fragments and translation operators depicted in case study within the oil and gas industry, where in- the centre of figure 1. It extends a suitable upper on- formation models held in MIMOSA, ISO 15926, and tology in order to foster reuse and to ensure its over- RSM are to be synchronised among a number of sys- all organisation adheres to well-established knowledge tems. representation principles. Each fragment in our on- In this paper we make the following contributions: tology is linked with its corresponding template(s) in • we outline our overall approach to automated one or more source or target ontologies. Note that not fragment-based information integration, every fragment will have corresponding templates in each data source. Therefore, it is necessary to find • present our approach to linking existing tax- combinations of fragments that are available in the onomies and ontologies to our model, target ontology that correspond to fragments found in the information source. Furthermore, it may be • show how to capture information fragments and necessary to query information from the target in or- associated translation operations in an ontology, der to obtain information (such as object identifiers of and existing objects) while instantiating target fragments. We intend to construct an ontology that captures • show how the resulting ontology can be applied the properties and constraints of fragments in a for- to infer automated translation between source mal manner such that equivalent instances of frag- and target representations. ments can be identified and synthesised. In order to The paper is organised as follows. We present our reason about transformations, it is necessary to cap- overall approach in Section 2 and introduce a running ture the constraints that govern a template’s use in example in Section 3. We discuss related work in the target ontology and the constraints on its param- Section 4 before outlining future work and concluding eters. In addition, a library of conversion operators the paper in Section 5. that transform between representations of basic data types will be included in order to enable automated construction of equivalent fragment instances where 2 Approach data types differ. Our intent is not to reinvent yet another ontology of entities and relationships, but to Our approach to overcoming heterogeneities in differ- reuse as much as possible existing ontologies and their ent standards is based on the construction of a uni- formal underpinnings and focus on the representation fying ontology that comprises the information frag- of information fragments, their parameters and con- ments used in source and target standards. The in- straints. gredients to our approach are depicted in Figure 1: Naturally, effective fragment representation and information fragments and transformations are repre- translation depends on a unified model of the under- sented in a unifying central ontology, and fragments lying entities and relationships. We intend to obtain are linked to concrete representations, so-called “tem- such a model by extending a suitable upper ontol- plates”, in their source and target representations. ogy with concepts and relationships relevant to our domain. These will be drawn from the data mod- Fragments and Templates. We will use the term els underpinning each standard (arrows labelled A “template” to refer to a particular parametric repre- in Figure 1), along with a unified taxonomy of en- sentation of a chunk of information in the source and tities/relationships defined in each standard (arrows target representation (which may or may not be sup- labelled B). ported by an ontology), and will use the term “frag- We intend to leverage as much as possible exist- ment” to refer to its parametric representation in our ing large scale ontologies, especially those that have been developed by the oil and gas industries, but the oil and gas industry for representing and managing latter cannot cover all bases. While the use of the the life-cycle data of assets. Whereas ISO 15926 is ISO15926 ontology has been suggested as a universal predominately used for the exchange of design infor- upper ontology, it has been shown to suffer from sig- mation, MIMOSA is mainly used in the operations nificant intrinsic shortcomings Smith (2006), as well and maintenance area. Both standards provide a data as being based on a formalism that has been shown model that specifies a generic structure to represent to make it difficult to concisely express engineering asset information, along with a “Reference Data Li- knowledge Felfernig et al. (2003). To express stan- brary” where the available asset types and their prop- dard engineering data sheets, ISO15926 therefore re- erties are defined. lies on a first order logic platform different from its We consider the frequent use case of hand-over core ontology specified in Part 2 of the standard, for- of design information to operations and maintenance mally specified but not implemented, and subject to that occurs when, for example, an asset is installed creating classic “impedance mismatch” situations for on site. Although both standards are designed to modellers as different formalisms have to be joined in overcome the interoperability problem, currently all one ISO15926 application. translation between these two standards is performed Whereas we anticipate that lexical matching and manually. specialisation relationships will allow us to largely In the following we will briefly illustrate the differ- identify corresponding entities in this domain, we ex- ences in representation between the two standards, pect that more involved procedures and manual in- and show that a common framework of information tervention will be necessary in order to address dis- fragments and transformation operations can over- crepancies arising from different standards and dif- come these discrepancies. ferent granularity of representation. Our work will For purposes of illustration, we assume that trans- thus highlight where established standards agree and lation between the MIMOSA CCOM 3.35 and the where further work is needed to reconcile modelling ISO 15926 (ISO 2003) data models is to be performed. heterogeneities. However, our approach will be applicable in scenarios where data represented in multiple standards, such as Translation Process. The overall approach is RSM and XMpLant6, must be utilised jointly to per- summarised as follows, as shown by the red path in form a complete translation. For example, the cur- Figure 1: First, the information in the source repos- rent ISO 15926 standard must be complemented with itory (MIMOSA) is read and templates are instan- additional geometry information stored in XMpLant tiated accordingly (Step 1). Each template is then data structures in order to synthesise a complete rep- translated into its counterpart fragment in our unify- resentation to be consumed by an RSM-compliant ing ontology (Step 2), where equivalent fragments in tool. the target representation are subsequently identified and synthesised (Step 3). Once a representation in 3.1 MIMOSA terms of the target fragments has been obtained, each fragment is translated into its corresponding template The data model of MIMOSA has been defined us- in the target representation (Step 4), which is then ing the language of the Unified Modelling Language emitted into the target system (Step 5). (UML). The various entities in the data model are One may argue that suitable templates/fragments represented as classes and the permitted relationships and their translations should not be defined explicitly between the various entities as associations between but instead be inferred directly from the ontologies classes. The data model is generic in that it allows underlying the source and target systems. Unfortu- for extensions to the model by means of a reference nately, such ontologies do not always exist explicitly, library comprising the asset types, units of measure, cover only the information part but not its concrete and other domain-specific entities that may be as- representation, or are not sufficiently complete to fa- signed to properties of classes in the generic data cilitate all needed inferences. Therefore, our approach model. The resulting model is serialised into an XML relies on a pragmatic template-based approach, which document and can be transmitted via XML Web Ser- also helps when dealing with issues such as tools that vice interfaces defined in the standard. do not fully conform or abuse the official standards. Figure 2 depicts a partial model of a property at- Furthermore, current formalisations are not typically tached to an asset. The labels ‘A’ through ‘E’ in- rich enough to overcome significant differences in rep- dicate instantiated templates used to represent the resentation, such as property-based vs. relationship- property ‘Ambient operating temperature range’ for based models, in particular if source and target use the attached asset, i.e. a pump. The templates la- different levels of granularity for certain aspects of an belled ‘A’ and ‘B’ associate an Attribute with a type. entity. Templates labelled ‘C’ and ‘D’ associate the attribute to numerical values whereas the templates labelled ‘E’ are needed to constrain the magnitude of each at- Validation. The ontology and transformation op- tribute type to ‘Degrees Celsius’. erators can be leveraged to verify the consistency and completeness of translation. Formal analysis of the possible translation sequences can establish if there 3.2 ISO 15926 are multiple possible representations in the target rep- resentation, or if some parts of the source represen- This standard evolved from a smaller project begin- tation could not be translated. In both cases, the ning in 1991 in an effort to provide a generic data ontology may need to be augmented in order to elim- model for representing assets throughout their life cy- inate unwanted translations and improve coverage. cle. One of the goals of this standard is “to facili- tate integration of data to support the life-cycle ac- tivities and processes of process plants.” (ISO 2004). 3 Case Study For this purpose, it relies on a generic data model

5http://www.mimosa.org/?q=resources/specs This case study is based on the MIMOSA and the 6 ISO 15926 standards used predominantly within the http://www.noumenon.org.uk/category.php?id=NA== Figure 2: MIMOSA CCOM: Ambient Temperature Range

Figure 3: ISO 15926: Ambient Temperature Range where all properties and facts are expressed as rela- a classification relationship (LowerBoundOfProper- tionships between objects. A catalogue of well-known tyRangeTemplate labelled ‘IsoA’ and UpperBoundOf- object classes and relationship classes is defined in PropertyRangeTemplate labelled ‘IsoB’ respectively), its so-called “Reference Data Library”, an extensi- and each boundary property is linked to its value ble catalogue accessible via Web Service interfaces. and scale via a ternary relationship (LowerUpperMag- Similar to the MIMOSA library of component types, nitudeOfPropertyRange labelled ‘IsoC’). The entire the ISO reference library associates unique identi- model is composed from templates, where shared pa- fiers with each well-known entity in the reference li- rameter assignments manifest as overlapping regions brary, thus defining a public “vocabulary” that shall in the figure. facilitate interoperability. In addition, the standard specifies well-known “templates” that express elemen- 3.3 Transformation tary statements within the object model, and defines a data exchange format in terms of XML and the Comparing the MIMOSA and the ISO 15926 rep- W3C’s RDF framework. resentation, it is easy to see that both models fol- Figure 3 shows a partial model of the same asset low fundamentally different approaches. Automated property expressed in the ISO 15926 data model. It translation between the two would be challenging at can be seen that numerous templates are needed in best, as many elements, such as PropertyQuantifi- order to state that the property is an interval express- cation and ClassOfIdentification, do not have corre- ing a temperature in the Celsius scale7. Minimum sponding counterparts in the other model. Further- and maximum bounds of the interval are denoted by more, the libraries providing reference data in both standards differ in many aspects. In our example, 7Note for clarity not all templates identified are highlighted the modelling of temperature scales differs, where MI- Similar to the previous statements, we annotate MOSA follows an indirect approach using a base type mapped reference library entities with a predicate and a reference type. that indicates their origin. This will be needed to However, these differences can be hidden by trans- ensure that only entities are emitted that exist in the lating each of the templates into the unified in- target standard. We write SC to denote the scale formation fragment ontology. In this intermediate Degrees Celsius. representation, only the intent of the template and From analysis of the source MIMOSA data, tem- the meaning of its parameters are captured, while plates have been recognised and translated into cor- representation-specific details are omitted. responding fragments, yielding the following facts in We illustrate the approach based on a transforma- the unified ontology: tion from MIMOSA to ISO 15926. Figure 2 shows a partial model of a temperature range attribute at- MimB(Ax, T RAMax) tached to an asset type. The outlined elements indi- MimB(An, T RAMin) cate two template instances, MimA and MimB. The former represents the statement that the attribute’s MimA(Ax, 220,SC) value is 220 with unit Degrees Celsius, and the lat- MimA(An, −90,SC) ter expresses that the attribute’s type is Temperature, Rated Ambient Maximum. Both, scale unit and type We use Ax and An to refer the properties corre- stem from MIMOSA’s reference data library which is sponding to the two attributes present in the example assumed to have been mapped into our unified ontol- model. TRAMin represents the type Temperature, ogy. Rated Ambient Minimum in the unified reference li- Assume that the ontology provides the following brary ontology. statements about properties and their types: Given these fragments corresponding to the information fragments derived from MIMOSA tem- MagnitudeOfP roperty(x1, x2, x3) → plate instances, we can now infer the equivalent P roperty(x1) ∧ ArithmeticNumber(x2) ∧ Scale(x3) fragments corresponding to ISO15926 template UpperBoundOfP ropertyRange(x, y) → instances that are to be emitted in the target model: From MimB(Ax, T RAMax) we obtain Range(x) ∧ P roperty(y) P ropertyT ype(Ax, T RAMax) and, by the sub- P ropertyT ype(x, y) → P roperty(x) ∧ T ype(t) type axiom, P ropertyT ype(Ax, T emperature). Since Temperature is a type available in the Furthermore, assume that the reference data libraries target standard, this yields fragment IsoProp- in both standards have been aligned and reconciled Type(Ax,Temperature), which can be instantiated into a unified ontology as well. For brevity, we only as template directly in the target model. Similarly, show the subset necessary to illustrate our example: MimA(Ax, 220,SC) yields IsoD(Ax, 220,SC). Furthermore, P ropertyT ype(Ax, T RAMax) entails T ype(T RAMax) that (∃r)UpperBoundOfP ropertyRange(r, Ax), which allows us to instantiate an ISO15926 template SubT ypeOf(T RAMax, T emperature) associated with IsoA(r0, Ax). Note that the exis- P ropertyT ype(x, T RAMax) → tential quantification forces us to select a suitable r0 representing an interval property in the target ∃r : UpperBoundOfP ropertyRange(r, x) model. For “green fields” transformations, where P ropertyT ype(x, t) ∧ SubT ypeOf(t, u) → the target model is empty, a suitable entity can be P ropertyT ype(x, u) generated. Otherwise, the target endpoint must be queried to obtain the needed entity. The MIMOSA information fragments corresponding to the templates can then be characterised as follows: 4 Related Work MimosaT ype(T RAMax) Gellish and the ISO 15926 standard both have con- MimB(x, y) ↔ P ropertyT ype(x, y), MimosaT ype(y) structed comprehensive ontologies and data models for representing engineering assets, processes, and as- MimA(x, y, z) ↔ MagnitudeOfP roperty(x, y, z) ∧ sociated meta-data. Common to both is the assump- MimosaT ype(z) tion that all data not represented in other forms is converted into the standard’s ontology. In the case Term TRAMax represents the mapped attribute type of ISO 15926, simple tools have been developed that Temperature, Rated Ambient Maximum. We use allow one to create simple mappings between entities predicate MimosaType to indicate that a particular and their properties. However, this requires a labori- element is mapped from the MIMOSA reference li- ous process and leaves the actual translation largely brary. This will be needed when translating from to manual coding. In contrast, our approach covers ISO15926 to MIMOSA. complex translations in terms of entire patterns and Similarly, let the ISO15926 fragments correspond- aims to automatically synthesise a suitable target rep- ing to the templates shown in Figure 3 be represented resentation and validate the translation. as follows: Sv´abZamazalˇ & Scharffe (2009) proposes a trans- formation service capable of generating alternative IsoT ype(T emperature) modelling choices while retaining the intended mean- IsoT ype(SC) ing of each transformed pattern. In contrast, we are concerned with translating information expressed in IsoD(x, y, z) ↔ MagnitudeOfP roperty(x, y, z) ∧ multiple formal languages and not to transform en- IsoT ype(z) tire ontologies. More importantly, the authors assume that corresponding patterns are already given, or can IsoA(x, y) ↔ UpperBoundOfP ropertyRange(x, y) easily be identified from purely structural criteria in IsoP ropT ype(x, y) ↔ P ropertyT ype(x, y) ∧ IsoT ype(y) the source and target ontologies. Currently most state-of-the-art ontology matching CCOM data model. Both will be formalised within techniques focus on the matching of “atomic” con- the unified ontology. Third, planning and rewriting cepts and do not address translation between multi- techniques will be investigated in order to create a entity concrete representations which require consid- reasoning framework that is effective in synthesising ering context (Ritze et al. 2009). These techniques are and validating translations as well as being scalable typically unable to resolve ontological heterogeneities to real problems requiring incremental translation be- between atomic concepts and a target ontology where tween multiple endpoints. the intended meaning is defined using multiple con- cepts or concepts with one or more attributes or rela- tionships. Ritze et al. (2009) investigate matching of References complex correspondences by means of patterns com- Berger, S., Grossmann, G., Stumptner, M. & Schrefl, prised of an atomic concept and attribute. M. (2010), Metamodel-Based Information Integra- In contrast, our work focuses on larger fragments tion at Industrial Scale, in ‘ACM/IEEE Interna- comprised of multiple elements, which may not be tional Conference on Model Driven Engineering represented in the target ontology or may be rep- Languages and Systems (MoDELS)’, Vol. LNCS resented differently. Therefore, our aim is to find 6395, Springer-Verlag, pp. 153–167. and translate between corresponding complex tem- plates. Further, our work characterises the key prop- Felfernig, A., Friedrich, G., Jannach, D., Stumptner, erties of templates into a “fragment ontology” and M. & Zanker, M. (2003), ‘Configuration knowledge investigates the composition of fragments to synthe- representation for semantic web applications’, Arti- sise complete translations into multiple different rep- ficial Intelligence for Engineering Design, Analysis resentations. Template translations can then be used and Manufacturing 17(1), 31–50. Special issue on as a semantic foundation for formal mapping-based Configuration. translation tools as described in Berger et al. (2010). Hamdi et al. (2010) are concerned with identify- Hamdi, F., Reynaud, C. & Safar, B. (2010), Pattern- ing refactorings in ontologies based on change pat- based mapping refinement, in P. Cimiano & terns, where refactoring operations and implied map- H. Pinto, eds, ‘Knowledge Engineering and Man- pings are derived from lexical similarity and struc- agement by the Masses’, Vol. 6317 of Lecture Notes tural subsumption. Although our aim is to transfer in Computer Science, Springer Berlin / Heidelberg, information rather than refactor ontologies, some of pp. 1–15. 10.1007/978-3-642-16438-5 1. their pattern predicates may serve as inspiration for our ontology. ISO (2003), Industrial Automation Systems and In- The patterns defined in Scharffe (2009) form tegration – Integration of Lifecycle Data for Pro- a rather generic classification of correspondences cess Plants Including Oil and Gas Production Fa- (attributes, classes, and relations). The work cilities – Part 2: Data Model, International Stan- seems mostly concerned with representing the hetero- dards Organisation, Geneva, Switzerland. Ref. No. geneities as patterns. The focus is on relatively simple ISO15926-2:2003(E). “patterns” in concept hierarchies without comprehen- ISO (2004), Industrial Automation Systems and In- sive domain axiomatisation. tegration – Integration of Lifecycle Data for Pro- cess Plants Including Oil and Gas Production 5 Conclusion Facilities – Part 1: Overview and fundamen- tal principles, International Standards Organisa- In this paper we have described a potential solution tion, Geneva, Switzerland. Ref. No. ISO15926- to address problems arising from the need to trans- 1:2004(E). fer information between multiple ontology-supported yet incompatible information systems. Given compre- Ritze, D., Meilicke, C., Sv´abZamazal,ˇ O. & Stuck- hensive heterogeneities in data models and meaning enschmidt, H. (2009), A pattern-based ontology alike, current approaches to schema- and ontology- matching approach for detecting complex corre- matching are unable to overcome all difficulties. spondences, in ‘Proc. of Int. Workshop on Ontology We proposed an approach based on a unified on- Matching (OM)’. tology of “information fragments” that represent ab- stractions of concrete representations of chunks of in- Scharffe, F. (2009), Correspondence Patterns Repre- formation in different data models. Including asso- sentation, PhD thesis, University of Innsbruck. ciated translation operators between primitive data Smith, B. (2006), Against idiosyncrasy in ontol- types, our model can support automated planning ogy development, in ‘Proceedings Formal Ontol- techniques in order to derive transformation opera- ogy in Information Systems (FOIS’06)’, IOS Press, tions from one representation into another. The same pp. 15–26. framework will also allow us to identify aspects of data models that cannot be translated adequately, Sv´abZamazal,ˇ O. & Scharffe, F. (2009), Pattern- hence aiding the developer in revising or extending based ontology transformation service, in ‘Proceed- the translation ontology and/or the underlying stan- ings of the 1st International Conference on Knowl- dards and their ontologies. edge Engineering and Ontology Development’. There are multiple avenues to follow-up on our preliminary investigations. First, ontology matching tools will need to be evaluated to assess which are most suitable to infer (mis-)alignment between ex- isting reference data libraries of ISO 15926 and MI- MOSA. As a result, an aligned unified ontology com- prising the entities and relationships of both stan- dards will be built. Second, the templates defined in the ISO 15926 data model will need to be comple- mented with templates sourced from the MIMOSA