Systems and methods for ML-based engineering library translation and integration and for structure and file format mapping

Abstract

Embodiments of the present disclosure relate to systems and methods for ML-based engineering library translation and integration and for structure and file format mapping. The method comprises: using an artificial intelligence / machine learning AI / ML model to map content between an interface of a first entity for interacting with other entities and an interface of a second entity for interacting with other entities, and / or classifying the content of the interface of the first entity and / or the interface of the second entity; and obtaining, from the AI / ML model, a first output indicating a result of the mapping of the content and / or a result of the classification of the content.

Description

Technical Field

[0001] The present invention relates to systems and methods for machine learning-based engineering library translation and integration, and for structure and file format mapping. Background Art

[0002] Today, in the field of process and automation engineering (PAEng) libraries, several engineering projects are typically based on various different libraries, structures, and file formats. As a result, the PA library landscape is likely to not undergo major changes. In particular, a reduction in the number of libraries is not expected, and there may not be any agreement on using a particular library / standard either. The reason is especially that customers require special libraries for their specific processes. Therefore, the challenge of mapping or at least classifying content including concepts, naming, categories, relationships, parameters, units, and even entire method signatures or functions between libraries will remain. Summary of the Invention

[0003] However, in view of the above, manual matching, mapping, or classification ties up a large amount of resources and is very error-prone. For example, in order to map a concept from one library (e.g., MinLib) to another library (e.g., PC Device Lib), the data of interest (e.g., characterizing elements) such as names, categories, tags, relationships, parameters, units, and / or method signatures or functions must be matched or mapped (or at least classified). Therefore, support or even full automation of the library matching, mapping, or classification process is needed, because this will significantly reduce costs and / or errors. Thus, support or even full automation of the library matching, mapping, or classification process between several libraries of different content is needed, and costs and / or errors will be significantly reduced.

[0004] To solve one or more of these problems, in a first aspect, a method is provided that includes using an artificial intelligence / machine learning AI / ML model to map content between an interface of a first entity for interacting with other entities and an interface of a second entity for interacting with other entities, and / or to classify the content of the interface of the first entity and / or the interface of the second entity; and obtaining a first output from the AI / ML model indicating the result of the mapping of the content and / or the result of the classification of the content.

[0005] That is, given interface A of a first entity and interface B of a second entity, the names or other handles for exposing variables, methods, functions, or other items of interface A can be input into an AI / ML model. Based at least in part on the output of the AI / ML model, the names or other handles for causing interface B to most likely expose the corresponding items can then be determined. In many cases, the output of the AI / ML can point directly to one name or other handle of interface B. However, this is not required. Benefits can be achieved once the output of the AI / ML model allows narrowing down the names or other handles of interface B that can correspond to a particular item exposed by interface A. For example, the output of the AI / ML model can result in, out of 100 names or other handles in interface B, only 10 being likely to be a proper match for the name or other handle of interface A that references a particular item.

[0006] Thus, even if only some of the names or other handles in interface A can be automatically mapped to names or other handles in interface B, and the mapping can also result in more than one name or other handle in interface B for a particular item, this already reduces the manual effort required to convert the names and / or handles exposed by interface A that reference a particular item into the names and / or handles exposed by interface B that reference the same item.

[0007] In this way, compared to manual mapping or categorization of content between different entities or of different entities, the method can allow for a reduction in the amount of resources and / or errors. In addition, the method can also allow for being applied to third-party entities, such as third-party libraries, structures, or files, for example in cases where other MTPs should be integrated into a toolchain.

[0008] There can be many reasons why it is impractical or even impossible to unify the interfaces used within an industrial plant. One reason is that in a complex plant, a change in one place always risks causing problems in other unexpected places that are directly or indirectly linked to it. In addition, the field devices in the plant may have such old dates that there is no longer software support for changing the interface, while the hardware field devices themselves are still in working condition. In some industries, such as nuclear power or other regulatory sectors, where operations rely on government-issued licenses, the conditions of these licenses may impose further restrictions on what may still be changed and what must remain unchanged after the license is issued. In an extreme example, even the font and color in which the control software in a nuclear power plant displays something to an operator may not be changed without permission.

[0009] In one example, the first entity can be at least one of a first engineering library, a first standard, a first structure, and a first file format, where the second entity can be at least one of a second engineering library, a second standard, a second structure, and a second file format.

[0010] In another example, the method may include: using pre-trained embeddings to map content between an interface of a third entity for interacting with other entities and an interface of a fourth entity for interacting with other entities, and / or classifying the content of the interface of the third entity and / or the interface of the fourth entity; obtaining a second output from the pre-trained embeddings indicating the result of the mapping of the content and / or the result of the classification of the content; training and operating an AI / ML model in shadow mode based on the received user feedback related to the second output and / or based on labeled data utilized related to the second output; and if the AI / ML model in shadow mode reaches a predetermined maturity, extending the pre-trained embeddings through the AI / ML model. For example, an embedding can be a representation of an input expressed by a digital feature vector. Once such a feature vector is available, it can be used as an input for all types of AI / ML models.

[0011] In particular, the operation of the AI / ML model to be trained in shadow mode with the pre-trained embeddings or another "active" model may include feeding all inputs into the "active" model to the AI / ML model to be trained, and optimizing the parameters characterizing the behavior of the AI / ML model to be trained with the goal of improving its performance. The performance can be measured, for example, using feedback from process / plant engineers or some other domain experts. At the same time, the "active" model whose output has been used in production at least to a limited extent remains "frozen". That is, the parameters characterizing the behavior of the active model remain unchanged.

[0012] In addition, in another example, the third entity may be at least one of a third engineering library, a third standard, a third structure, and a third file format, and the fourth entity may be at least one of a fourth engineering library, a fourth standard, a fourth structure, and a fourth file format.

[0013] In another example, the obtaining of the first output may be based on a one-step method or a two-step method, where the first output may be the result of a mapping according to the one-step method or a mapping according to the two-step method. The one-step method may include: given the input and output provided via the respective interfaces, using an AI / ML model to map content between an interface of a first entity and an interface of a second entity, where the input and output are related to the content. The two-step method may include: given the input and output provided via at least one of the respective interfaces, using an AI / ML model to classify the content of the interface of the first entity and / or the content of the interface of the second entity, where the input and output are related to the content; using an AI / ML model to map the classified input and the classified output to the interface of the first entity and / or the interface of the second entity.

[0014] In addition, in another example, the mapping and / or classification of the given input and output may include: inputting a first entity input and / or a first entity output related to the content of an interface of a first entity into an AI / ML model, where the first entity input and / or the first entity output is related to a first content element of the first entity, and the first content element may be at least one of a first name, a first category, a first concept, a first relationship, and a first parameter, and the first entity may represent a source entity. Such mapping and / or classification may further include: receiving a model output from the AI / ML model based on the first entity input and / or the first entity output, where the model output is related to a second content element different from the first content element, and the second content element may be at least one of a second name, a second category, a second concept, a second relationship, and a second parameter, and / or where the model output may be based on the training of the AI / ML model. In addition, such mapping and / or classification may further include: for the model output, determining a match with a second entity input and / or a second entity output, where the second entity input and / or the second entity output is related to the content of an interface of a second entity, and the content of the interface of the second entity is related to at least one of a name, a category, a concept, a relationship, and a parameter, where the second entity may represent a target entity, and where the model output may be associated with a first output.

[0015] In addition, in another example, the determination of the match may include at least one of the following: determining a one-to-one match between the first entity input and / or the first entity output and the second entity input and / or the second entity output; determining a one-to-many match between the first entity input and / or the first entity output and the second entity input and / or the second entity output; determining a many-to-one match between the first entity input and / or the first entity output and the second entity input and / or the second entity output; and determining a many-to-many match between the first entity input and / or the first entity output and the second entity input and / or the second entity output.

[0016] In another example, using the AI / ML model may be based on or may include leveraging one of the following: obtaining an embedding representation from a joint embedding space and performing a nearest neighbor search based thereon, named entity recognition NER (e.g., based on Transformer-NN), and graph neural network Graph-NN.

[0017] One advantage of graph neural networks is that they can encode the relationships between content items, specifically, one item can be computed from other items. For example, if interface A exposes voltage and current, and interface B exposes power, then the graph neural network can encode that the power is the product of the voltage and the current.

[0018] In addition, in another example, the method may further include using a knowledge-based component associated with the AI / ML model, the knowledge-based component using an underlying knowledge representation system (such as a knowledge graph), possibly in combination with a Graph-NN, to utilize graph-containing information via the interface of the first entity and / or the interface of the second entity; and further considering the utilized graph-containing information for mapping and / or classification related to the interface of the first entity and / or the interface of the second entity.

[0019] In addition, in another example, the method may further include: if the result indicated by the first output includes an uncertainty value equal to or higher than a predetermined uncertainty threshold, providing the result to the user for confirmation; and feeding back the decision received from the user on whether to accept the result to the knowledge base associated with the AI / ML model and / or the AI / ML model.

[0020] In another example, the content may be related to at least one of the following: concepts, naming, categories, relationships, parameters, method signatures, and method functions.

[0021] According to a second aspect, there is provided a method including: using pre-trained embeddings to map content between an interface of a third entity for interacting with other entities and an interface of a fourth entity for interacting with other entities, and / or classifying content of the interface of the third entity and / or the interface of the fourth entity; obtaining a second output indicating a result of mapping of the content and / or a result of classification of the content from the pre-trained embeddings; training and operating an artificial intelligence / machine learning (AI / ML) model in a shadow mode based on the received user feedback related to the second output and / or based on labeled data utilized related to the second output; and if the AI / ML model in the shadow mode reaches a predetermined maturity, extending the pre-trained embeddings by the AI / ML model.

[0022] According to a third aspect, there is provided a control device for a mapping and / or classification control system, the control device being configured to perform the method of the first aspect and / or the second aspect above.

[0023] According to a fourth aspect, there is provided a mapping and / or classification control system configured to perform the method of the first aspect and / or the second aspect above.

[0024] According to a fifth aspect, there is provided an industrial automation system including the control device of the third aspect and / or the mapping and / or classification control system of the fourth aspect.

[0025] The method of the first aspect and / or the second aspect may be computer-implemented. Optional features of the first aspect may form part of any one of the second to fifth aspects, with necessary modifications.

[0026] According to a sixth aspect, there is provided a computing system configured to perform the method of the first aspect and / or the second aspect.

[0027] According to a seventh aspect, there is provided a computer program (product) comprising instructions which, when executed by a computing system, cause the computing system to be able to perform or cause the computing system to perform the method of the first aspect and / or the second aspect.

[0028] According to an eighth aspect, there is provided a computer-readable (storage) medium comprising instructions which, when executed by a computing system, cause the computing system to be able to perform or cause the computing system to perform the method of the first aspect and / or the second aspect. The computer-readable medium may be transient or non-transient, volatile or non-volatile.

[0029] By using an AI / ML model for mapping and / or classification purposes, as outlined above and described in more detail below, compared to such a manual process, it is possible to at least reduce resource requirements and the occurrence of mapping errors and / or classification errors.

[0030] A “(process) automation system” refers to an industrial or production device comprising one or more pipelines, production lines, and / or assembly lines for converting one or more isolates into products and / or for assembling one or more components into a final product.

[0031] As used herein, the term “acquire” may include, for example, receiving from another system, device, or process via interaction with a user; loading or retrieving from a storage or memory; measuring or capturing using a sensor or other data acquisition device.

[0032] As used herein, the term “determine” encompasses a variety of actions and may include, for example, calculating, computing, processing, deriving, investigating, looking up (e.g., looking up in a table, database, or another data structure), ascertaining, etc. In addition, “determine” may include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory), etc. Moreover, “determine” may include parsing, selecting, picking, establishing, etc.

[0033] The indefinite article “a” or “an” does not exclude a plurality. In addition, the articles “a” and “an” as used herein shall generally be construed to mean “one or more” unless otherwise stated or clearly indicated as the singular form from the context.

[0034] Unless otherwise stated, or clearly apparent from the context, the phrases "one or more of A, B, and C", "at least one of A, B, and C", and "A, B, and / or C" as used herein are intended to represent all possible permutations of one or more of the listed items. That is, the phrase "A and / or B" means (A), (B), or (A and B), and the phrase "A, B, and / or C" means (A), (B), (C), (A and B), (A and C), (B and C), or (A, B, and C).

[0035] The term "comprising" does not exclude other elements or steps. In addition, the terms "comprising", "including", "having", etc. may be used interchangeably herein.

[0036] The present invention may include one or more aspects, examples, or features, either alone or in combination, whether specifically disclosed in such combination or alone. Any optional feature or sub-aspect of one of the above aspects is suitably applied to any other aspect.

[0037] With reference to the detailed description provided hereinafter, the above aspects will become apparent and be elucidated. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] A detailed description will now be given, by way of example only, with reference to the accompanying drawings, in which:[[]]END]]

[0039] Figure 1 An example overview of an AI / ML model-based mapping and / or classification process including a feedback loop is shown;

[0040] Figure 2 An example method related to an AI / ML model-based mapping and / or classification process is shown; and

[0041] Figure 3 Another example method related to an AI / ML model-based mapping and / or classification process is shown. DETAILED DESCRIPTION

[0042] To address the above drawbacks in the prior art, the use of AI / ML to (further) automate the mapping and transformation of, for example, libraries, structures, and file formats is particularly disclosed herein. ML / AI methods that may be used (not limited to these ML / AI methods) are: for example, if an underlying information model with concepts and relationships is available, obtaining a joint embedding representation and performing a nearest neighbor search among them, named entity recognition (NER) (possibly based on a transformer NN), and graph-neural network (Graph-NN).

[0043] The following are non-limiting examples of a method overview:

[0044] Utilize pre-trained LLMs / embeddings for out-of-the-box library content classification or mapping, e.g., particularly when there is no labeled data, i.e., it is not known how library 1 (lib1) items map to library 2 (lib2) items.

[0045] Utilize labeled data or human expert feedback starting in shadow mode / shadow deployment to train (i.e., calibrate and / or fine-tune) one's own (e.g., newly set up) AI / ML model, so that the AI / ML model is continuously improved and becomes an expert in, e.g., specific library-to-library (lib-2-lib) conversions.

[0046] Additionally, knowledge-based components can also use underlying knowledge representation systems, such as the MTP ontology, eBase graphs, possibly combined with Graph-NN, to utilize information containing graphs.

[0047] In cases of high uncertainty (i.e., when the uncertainty level may exceed a predetermined level), mapping cases can be presented to an expert (i.e., a domain expert user) for confirmation, and the expert / user's decision (regarding whether the result associated with the high uncertainty is correct (or acceptable)) is fed back into the knowledge base or the AI / ML model.

[0048] Thus, an AI / ML model is provided to map content, e.g., between two (engineering) libraries. Additionally, a flexible multi-model approach is provided to handle completely unknown lib-2-lib mappings as well as partially known lib-2-lib mappings, and to continuously improve the entire AI / ML-based library integration system. Additionally, flexible ideas for deployment in critical environments are provided, which request / allow user feedback based on the AI / ML-based output uncertainty.

[0049] Thus, the advantageous technical effects are achieved in such a way that manual matching, which binds a large amount of resources and is very error-prone, can be avoided (or at least reduced). Thus, supporting (or even fully automating) the library mapping process will significantly reduce costs and / or errors. Additionally, the advantageous technical effects are achieved in such a way that, e.g., when other MTPs are to be integrated into a given toolchain, the application of, e.g., third-party libraries, files, and structures can be enabled.

[0050] According to various examples, possible AI / ML-methods for using AI / ML to map or classify, e.g., library concepts, file formats, and / or structures include (but are not limited to these examples): (1) obtaining a joint embedding representation and performing a nearest neighbor search based on it, (2) named entity recognition (NER) (e.g., based on Transformer or x-LSTM), (3) Graph-NN (e.g., if concepts and relationships are available).

[0051] (1) and (2) can both use pre-trained (out-of-the-box) models (such as LLMs) or out-of-the-box embeddings and apply them to map the content of lib1 to the content of lib2. In this way, for most fields, such as the least common denominator for different components (such as valves / controllers / tanks), plus mapping lib1 to lib2, a fairly reliable / clear mapping from lib1 to lib2 is already available. The remaining parts can be grayed out and / or deactivated, and / or can trigger feedback from experts (users).

[0052] In addition, experts (users) can continuously request feedback on the automatic mapping, which is provided, for example, by the AI / ML methods used.

[0053] Meanwhile, in shadow mode (e.g., without the user knowing it), the feedback can be used to train (or fine-tune) (one's own, e.g., newly established) expert models (such as (expert) AI / ML models), and once a predetermined confidence level is reached (i.e., once the expert model is understood, expected, and / or considered to be sufficiently confident or reliable (e.g., reaching a predetermined threshold related to the correctness of the results of the provided mapping)), the expert model can be deployed.

[0054] Reference Figure 1 , which shows an exemplary overview of an AI / ML model-based mapping and / or classification process including a feedback loop. More specifically, a mapping from library 1 to library 2, i.e., a lib1 to lib2 mapping, is shown.

[0055] Now refer to Figure 1 , in step 1, the pre-trained embedding 110 is used to map the inputs and outputs (IO) of library PC Device Lib 120 and MinLib / eBase 130. Figure 1 The indicator number 1 in indicates mapping between library PC Device Lib 120 and MinLib / eBase 130 by using the IO mapper 140. The obtained embedding 110 can indicate and provide results with, for example, nearest neighbor search. Figure 1 The indicator number 3 in indicates that the mapping result and / or classification result can be output.

[0056] In step 2, for example, in the case where the output obtained from the embedding 110 is incorrect, the factory / process engineer (such as the user) 150 gives feedback. Figure 1 The indicator number 4 in indicates feedback and learning from the feedback.

[0057] In step 3, when the embedding 110 provides results, the AI / ML model (in Figure 1is shown as mapper 160) and is trained and operated in shadow mode. Such training and operation can be performed on (different) unstructured text, where the categories and relationships of the (different) unstructured text are based on feedback. The term "different" can be understood as being different (i.e., not the same) compared to the result (or output) provided by the embedding 110. The unstructured text can be any representational element, such as the name, description, parameters, and / or categories of the IO.

[0058] In step 4, given their inputs and outputs, the trained system (i.e., the trained mapper 160) is used to map two libraries (e.g., library PC Device Lib 120 and MinLib / eBase 130, or any different libraries / multiple libraries). This can be done using a two-step method or a one-step method.

[0059] In the two-step method, the mapper 160 can be used to classify the individual IOs of the libraries (e.g., library PC Device Lib 120 and MinLib / eBase 130) into their corresponding standard categories. The mapper 160 (the mapper category) then maps the standardized IOs to different libraries.

[0060] In the one-step method, the mapper 160 directly gives, for example, the mapping between two libraries as output.

[0061] In step 5, once the shadow model reaches a certain maturity, the embedding 110 can be extended or replaced by the shadow model, i.e., extended or replaced by the mapper 160 (previously trained and / or operated in shadow mode). Figure 1 The indicator number 5 in indicates that the performance of the embedding 110 and the mapper 160 can be compared. Based on the result of this comparison, for example, with respect to a predetermined performance threshold to be achieved by the mapper 160, it can be determined to extend the embedding 110 with the mapper 160.

[0062] In step 6, the mapper 160 learns from the feedback of the factory / process engineer 150 in the background.

[0063] Now refer to Figure 2 , example methods related to the mapping and / or classification process based on an AI / ML model are shown according to various examples.

[0064] In step S210, the method includes using an artificial intelligence / machine learning AI / ML model to map content between an interface of a first entity for interacting with other entities and an interface of a second entity for interacting with other entities, and / or classify the content of the interface of the first entity and / or the interface of the second entity.

[0065] It should be noted that such an AI / ML model can be represented as referred to above Figure 1Such a mapper 160 is outlined. In addition, the interface of the entity may be represented as above with reference to Figure 1 The library PC Device Lib 120 or MinLib / eBase 130. In addition, the term content may include any element included in the interface of an entity or in an entity, with respect to which input can be input into (the interface of) the entity and / or output can be obtained from (the interface of) the entity. For example, for the purpose of explanation only, reference is made to Figure 1 As shown in the library, the content of library MinLib / eBase 130 may have content including a content element named "Interlock Tr". Therefore, the output obtained from library MinLib / eBase 130 may be related to the content / content element named "Interlock Tr". Such output may be associated with a specific characterization element, including, for example, a specific name, category, relationship, parameter, structure and / or file format. In contrast, library PC Device Lib 120 may have content including a content element named "Interlock". Therefore, the input to library PC Device Lib 120 may be related to the content / content element named "Interlock". Such input may be associated with a specific characterization element, including, for example, a specific name, category, relationship, parameter, structure and / or file format. Therefore, the AI / ML model may be used to map the output of library MinLib / eBase 130 (e.g., source library) related to the content element named "Interlock Tr" to the input of library PC Device Lib 120 (e.g., target library) related to the content element named "Interlock". Therefore, given their corresponding inputs and outputs, a mapping between two different libraries can be achieved. This mapping is Figure 1 is shown as an example.

[0066] In step S220 , the method further includes obtaining a first output from the AI / ML model indicating a result of mapping of the content and / or a result of classification of the content.

[0067] It should be noted that such a first output may indicate such a mapping between the content elements "Interlock Tr" and "Interlock" as described above. In addition, such a first output may indicate such a mapping between the content elements "Interlock Tr" and "Interlock" as described above. Figure 1 The result of such a mapping (and / or classification) between the libraries PC DeviceLib 120 and MinLib / eBase 130 is shown.

[0068] It should also be noted that the above reference Figure 2The method outlined allows for library-to-library mapping (lib-2-lib-mapping) and standard-to-standard mapping (standard-2-standard-mapping). For example, in the case of mapping (or classifying content) between two different entities (or between the corresponding interfaces of two different entities), library-to-library mapping is obtained when the two different entities are two different libraries. Additionally, if content is mapped (or classified) between two different entities (or between the corresponding interfaces of two different entities), where the two different entities are two different standards (rather than two different libraries), standard-to-standard mapping is obtained. Similarly, the two different entities can represent different structures or different file formats. Examples of libraries include, for example, MinLib and PC Device Lib. Examples of standards include, for example, the module type encapsulation "MTP" (VDI-2658), eBase, and DEXPI.

[0069] Furthermore, according to various examples, the first entity can be at least one of a first engineering library or library (which can have first content), a first standard, a first structure, and a first file format, and wherein the second entity can be at least one of a second engineering library or library (which can have second content), a second standard, a second structure, and a second file format.

[0070] However, it should be noted that the entities are not limited to these examples. Instead, as outlined above with reference to Figure 2 and method steps S210 and S220, such a first entity and / or second entity can be understood as an entity having an interface for interacting with other entities.

[0071] Moreover, according to various examples, the method can further include: for example, before using an AI / ML model, using pre-trained embeddings to map content between the interface of a third entity for interacting with other entities and the interface of a fourth entity for interacting with other entities, and / or classifying the content of the interface of the third entity and / or the interface of the fourth entity; obtaining a second output from the pre-trained embeddings indicating the result of the mapping of the content and / or the result of the classification of the content; training and operating the AI / ML model in shadow mode based on the received user feedback related to the second output, and / or based on the labeled data utilized related to the second output; and, if the AI / ML model in the shadow mode reaches a predetermined maturity, extending the pre-trained embeddings through the AI / ML model.

[0072] It should be noted that the mapping by using pre-trained embeddings can be understood such that such a "mapping" can be generated by, for example, obtaining an embedding of the content by using pre-trained embeddings and performing a nearest neighbor search based on that embedding to identify the "best match" neighbor, i.e., the nearest neighbor (e.g., the nearest content / embedding for the selected content / embedding).

[0073] It should also be noted that such pre-trained embeddings can represent the embeddings 110 as referred to above Figure 1 described. In addition, the third entity and the fourth entity can represent the first entity and / or the second entity or entities different therefrom. That is, for example, the first entity, which can be at least one of a first engineering library (or libraries), a first standard, a first structure, and a first file format, can be the same as the third entity, and the third entity can be, for example, at least one of a third engineering library (or libraries), a third standard, a third structure, and a third file format. Thus, the first engineering library, the first standard, the first structure, and the first file format can be the same as the third engineering library, the third standard, the third structure, and the third file format, respectively. The same applies to the second entity with respect to the fourth entity.

[0074] In addition, the term "extension" can be understood such that the extension can, for example, consist in using the original model (i.e., the pre-trained embeddings), but with additional information included in a feedback-based model (i.e., a (trained) AI / ML model). However, the extension can also be understood as not further using the original model, i.e., the pre-trained embeddings, but using a feedback-based model, i.e., a (trained) AI / ML model. Thus, for example, when appropriate, the two models can be used together, i.e., combined with each other, so that, for example, a synergistic effect can be achieved, or the two models can be used separately.

[0075] In addition, according to various examples, such a third entity can be at least one of a third engineering library or libraries (which can have third content), a third standard, a third structure, and a third file format, and wherein the fourth entity can be at least one of a fourth engineering library or libraries (which can have fourth content), a standard, a fourth structure, and a fourth file format.

[0076] In addition, according to various examples, the obtaining of the first output can be based on a one-step method or a two-step method, and the first output is the result of mapping according to the one-step method or mapping according to the two-step method. The one-step method can include: given the input and output provided via the corresponding interfaces, using an AI / ML model to map the content between the interface of the first entity and the interface of the second entity, where the input and output are related to the content. The two-step method can include: given the input and output provided via at least one of the corresponding interfaces, using an AI / ML model to classify the content of the interface of the first entity and / or the content of the interface of the second entity, where the input and output are related to the content; and using an AI / ML model to map the classified input and the classified output to the interface of the first entity and / or the interface of the second entity.

[0077] It should be noted that such a one-step method and a two-step method can represent the one-step method and the two-step method as outlined above with reference to Figure 1 Step 4.

[0078] In addition, according to various examples, the mapping and / or classification of the given input and output can include: inputting the first entity input and / or the first entity output related to the content of the interface of the first entity into the AI / ML model, where the first entity input and / or the first entity output are related to the first content element of the first entity, and the first content element is at least one of a first name, a first category, a first concept, a first relationship, and a first parameter, and the first entity represents the source entity; receiving a model output from the AI / ML model based on the first entity input and / or the first entity output, where the model output is related to a second content element different from the first content element, and the second content element is at least one of a second name, a second category, a second concept, a second relationship, and a second parameter, and / or where the model output is based on the training of the AI / ML model; and determining a match with the second entity input and / or the second entity output for the model output, where the second entity input and / or the second entity output are related to the content of the interface of the second entity, and the content of the interface of the second entity is related to at least one of a name, a category, a concept, a relationship, and a parameter, and the second entity represents the target entity. The model output can be associated with the first output.

[0079] It should be noted that such a "first entity output" can represent an output related to the content element "Interlock Tr" as outlined above with reference to Figure 1 and 2 where the "second entity input" can represent an input related to the content element as outlined above with reference to Figure 1 and 2Input related to the content element "Interlock" as outlined. The "model output" can be understood to represent the "first entity output" and / or the "second entity input". Thus, the "model output" can be understood to represent the result of mapping the "first entity output" related to the content element "Interlock Tr" to the "second entity input" related to the content element "Interlock".

[0080] In addition, according to various examples, the determination of a match can include at least one of the following: determining a one-to-one match between the first entity input and / or the first entity output and the second entity input and / or the second entity output; determining a one-to-many match between the first entity input and / or the first entity output and the second entity input and / or the second entity output; determining a many-to-one match between the first entity input and / or the first entity output and the second entity input and / or the second entity output; and determining a many-to-many match between the first entity input and / or the first entity output and the second entity input and / or the second entity output.

[0081] In addition, according to various examples, the use of an AI / ML model can be based on or can include leveraging one of the following: obtaining a joint embedding representation via natural language processing (NLP) methods and performing a nearest neighbor search based thereon; named entity recognition NER (e.g., transformer-based); and graph neural network Graph-NN.

[0082] However, it should be noted that the use of an AI / ML model may not be understood to be limited to using one of joint embedding and nearest neighbor search, named entity recognition NER, and graph neural network Graph-NN. In fact, for example, the use of the AI / ML model outlined above with reference to Figure 2 and method steps S210 and S220 should not be understood to be limited to one (or several) specific methods.

[0083] Additionally, according to various examples, the method can further include using a knowledge-based component associated with the AI / ML model, the knowledge-based component using an underlying knowledge representation system, possibly in combination with Graph-NN, to leverage graph-containing information via the interface of the first entity and / or the interface of the second entity; and further considering the graph-containing information leveraged for mapping and / or classification related to the interface of the first entity and / or the interface of the second entity.

[0084] Additionally, according to various examples, the method may further include: if the result indicated by the first output includes an uncertainty value equal to or higher than a predetermined uncertainty threshold [the output mapping result or classification result obtained from the AI / ML model is uncertain], then providing the result to the user for confirmation; and feeding back the decision received from the user on whether to accept the result to the knowledge base associated with the AI / ML model and / or the AI / ML model.

[0085] Additionally, according to various examples, the content may be related to at least one of the following: concepts, naming, categories, relationships, parameters, method signatures, and method functions. It should be noted that such content may be understood to be obtained (e.g., received, accessed, provided, and / or output) in / through the interface of the corresponding entity.

[0086] Now referring to Figure 3 , according to various examples, an example method related to the mapping and / or classification process based on an AI / ML model is illustrated.

[0087] In step S310, the method includes using pre-trained embeddings to map content between the interface of a third entity for interacting with other entities and the interface of a fourth entity for interacting with other entities, and / or classifying the content of the interface of the third entity and / or the interface of the fourth entity.

[0088] In step S320, the method includes obtaining a second output indicating the result of the mapping of the content and / or the result of the classification of the content from the pre-trained embeddings.

[0089] In step S330, the method includes training and operating an artificial intelligence / machine learning AI / ML model in shadow mode based on the received user feedback related to the second output and / or based on the labeled data utilized related to the second output.

[0090] In step S335, if the AI / ML model in the shadow mode reaches a predetermined maturity ("yes" in S335), then the method includes, in step S340, extending the pre-trained embeddings through the AI / ML model.

[0091] It should be noted that the AI / ML model may represent such a mapper 160 as shown in Figure 1 and / or such an AI / ML model as shown in Figure 2 . Additionally, the pre-trained embeddings may represent the embeddings 110 as shown in Figure 1 and / or the pre-trained embeddings as shown in Figure 2 . Additionally, the content and the entities may be understood to represent as referred to above in Figure 2The content and entities outlined above. Thus, the third entity can be, for example, the same as the first entity (outlined above), and the fourth entity can be the same as the second entity (outlined above). Additionally, steps S330, S335, and S340 can represent at least a portion of the process outlined above with reference to Figure 1 Steps 4 and 5 of such a process outlined above.

[0092] Additionally, according to various examples, a control device for a mapping and / or classification control system is disclosed, wherein the control device is configured to perform the method outlined above with reference to Figures 1 to 3 The method outlined above. For example, the control device can include a processor and a memory for storing instructions, which when executed by the processor, can cause the control device to, for example, perform the method steps outlined above with reference to Figure 2 and 3 The method steps outlined above. For such an execution, the control device can include several functional parts, for example, a mapping and / or classification part for performing the process of step 210 according to Figure 2 and an acquisition part for performing the process of step 220 according to Figure 2 Additionally and / or alternatively, the control device can include several functional parts, for example, an embedding part for performing the process of step 310 according to Figure 3 an acquisition part for performing the process of step 320 according to Figure 3 a training and / or operation part for performing the process of step 330 according to Figure 3 and an extension part for performing the process of step 340 according to Figure 3 Furthermore, such parts can be understood as representing devices for performing specific functions or parts configured to perform specific functions.

[0093] Additionally, according to various examples, a mapping and / or classification control system is disclosed, which is configured to perform the method and / or method steps outlined above with reference to Figures 1 to 3 The method and / or method steps outlined above.

[0094] Additionally, according to multiple examples, an industrial automation system is disclosed, which includes a control device and / or a mapping and / or classification control system.

[0095] Additionally, according to various examples, a computer-readable medium including instructions is disclosed, which when executed by a computing system, causes the computing system to perform the method and / or method steps outlined above with reference to Figures 1 to 3 The method and / or method steps outlined above.

[0096] Furthermore, according to various examples, the method and / or method steps outlined with reference to Figures 1 to 3 can be computer-implemented.

[0097] In addition, according to various examples, there is provided a computing system configured to perform the methods and / or method steps outlined above with reference to Figures 1 to 3 the computing system.

[0098] In addition, according to various examples, there is provided a computer program (product) comprising instructions that, when executed by a computing system, enable the computing system to perform or cause the computing system to perform the methods and / or method steps outlined above with reference to Figures 1 to 3 the computing system.

[0099] Any unit, module, circuit system, or method described herein may be implemented using hardware, software, and / or firmware configured to perform any of the operations described herein. The hardware may include one or more processor cores, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), an application specific standard product (ASSP), a system on a chip (SOC), a complex programmable logic device (CPLD), and the like. The software may be implemented as a software package, code, instructions, an instruction set, and / or data recorded on at least one transient or non-transient computer-readable storage medium. The firmware may be implemented as code, instructions, or an instruction set and / or hard-coded data in a memory device (e.g., a non-volatile memory device).

[0100] If implemented in software, the functions may be stored or transmitted as one or more instructions or code on a computer-readable medium. The computer-readable medium includes computer-readable storage media. The computer-readable storage media may be any available storage media accessible by a computer. By way of example and not limitation, such computer-readable storage media may include FLASH storage media, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and that is accessible by a computer. Disk and optical disks as used herein include compact disk (CD), laser disk, optical disk, digital versatile disk (DVD), floppy disk, and Blu-ray disk (BD), where disks typically reproduce data magnetically, while optical disks typically reproduce data optically with a laser. In addition, propagated signals may be included within the scope of computer-readable storage media. The computer-readable medium also includes a communication medium that includes any medium that facilitates transfer of a computer program from one place to another. For example, a connection may be a communication medium. By way of example, if software is transferred from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technology such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technology such as infrared, radio, and microwave is included in the definition of communication medium. Combinations of the above should also be included within the scope of computer-readable media.

[0101] The applicant hereby discloses in isolation the features of each individual described herein and any combination of two or more such features to the extent that such features or combinations can be implemented by a person skilled in the art based on the common general knowledge from the specification as a whole, regardless of whether such features or combinations of features solve any of the problems disclosed herein, and are not limited to the scope of the claims. The applicant points out that aspects of the present invention may include any such individual features or combinations of features.

[0102] It must be noted that embodiments of the present invention are described with reference to different categories. In particular, some examples are described with reference to methods, while other examples are described with reference to apparatuses. However, a person skilled in the art will conclude from the specification that, unless otherwise stated, any combination between features belonging to different categories is also considered to be disclosed by this application, in addition to any combination of features belonging to one category. However, all features can be combined to provide more synergistic effects than a simple sum of the features.

[0103] Although the present invention has been described in detail and illustrated in the accompanying drawings and the foregoing description, such description and illustration should be regarded as exemplary rather than restrictive. The present invention is not limited to the disclosed embodiments. By studying the drawings, the disclosure, and the appended claims, a person skilled in the art can understand and implement other variations of the disclosed embodiments.

[0104] The fact that certain measures are recited in mutually different dependent claims does not mean that a combination of these measures cannot be used advantageously.

[0105] Any reference signs in the claims shall not be construed as limiting the scope.

[0106] Reference signs:

[0107] 110: Embedding;

[0108] 120, 130: Libraries;

[0109] 140: IO Mapper;

[0110] 150: User;

[0111] 160: Mapper

Claims

1. A method, comprising: Using an artificial intelligence / machine learning AI / ML model to map content between an interface of a first entity for interacting with other entities and an interface of a second entity for interacting with other entities, and / or to classify content of the interface of the first entity and / or the interface of the second entity; And Obtaining a first output from the AI / ML model, the first output indicating a result of the mapping of the content and / or a result of the classification of the content.

2. The method according to claim 1, Wherein the first entity is at least one of the following: a first engineering library, a first standard, a first structure, and a first file format; And wherein the second entity is at least one of the following: a second engineering library, a second standard, a second structure, and a second file format.

3. The method according to claim 1 or 2, further comprising: Using pre-trained embeddings to map content between an interface of a third entity for interacting with other entities and an interface of a fourth entity for interacting with other entities, and / or to classify content of the interface of the third entity and / or the interface of the fourth entity; Obtaining a second output from the pre-trained embeddings, the second output indicating a result of the mapping of the content and / or a result of the classification of the content; Training and operating the AI / ML model in shadow mode based on received user feedback regarding the second output and / or based on labeled data utilized in relation to the second output; And If the AI / ML model in shadow mode reaches a predetermined maturity, extending the pre-trained embeddings through the AI / ML model.

4. A method, comprising: Using pre-trained embeddings to map content between an interface of a third entity for interacting with other entities and an interface of a fourth entity for interacting with other entities, and / or to classify content of the interface of the third entity and / or the interface of the fourth entity; Obtaining a second output from the pre-trained embeddings, the second output indicating a result of the mapping of the content and / or a result of the classification of the content; Training and operating an artificial intelligence / machine learning AI / ML model in shadow mode based on received user feedback regarding the second output and / or based on labeled data utilized in relation to the second output; And If the AI / ML model in shadow mode reaches a predetermined maturity, extending the pre-trained embeddings through the AI / ML model.

5. The method according to claim 3 or 4, Wherein the third entity is at least one of the following: a third engineering library, a third standard, a third structure, and a third file format; And wherein the fourth entity is at least one of the following: a fourth engineering library, a fourth standard, a fourth structure, and a fourth file format.

6. The method according to any one of claims 1 to 5, wherein the obtaining of the first output is based on a one-step method or a two-step method, and the first output is a result of a mapping according to the one-step method or a mapping according to the two-step method, The one-step method comprises: Given the input and output provided via the respective interfaces, map the content between the interface of the first entity and the interface of the second entity using the AI / ML model, where the input and the output are related to the content; The two-step method includes: Given the input and output provided via at least one of the respective interfaces, classify the content of the interface of the first entity and / or the content of the interface of the second entity using the AI / ML model, where the input and the output are related to the content; And Use the AI / ML model to map the classified input and the classified output to the interface of the first entity and / or the interface of the second entity.

7. The method according to claim 6, wherein the mapping and / or the classification given the input and the output includes: Input a first entity input and / or a first entity output related to the content of the interface of the first entity into the AI / ML model, where the first entity input and / or the first entity output is related to a first content element of the first entity, The first content element is at least one of a first name, a first category, a first concept, a first relationship, and a first parameter, and the first entity represents a source entity; Based on the first entity input and / or the first entity output, receive a model output from the AI / ML model, where the model output is related to a second content element different from the first content element, the second content element is at least one of a second name, a second category, a second concept, a second relationship, and a second parameter, and / or where the model output is based on the training of the AI / ML model; And For the model output, determine a match with a second entity input and / or a second entity output related to the content of the interface of the second entity, where the content of the interface of the second entity is related to at least one of a name, a category, a concept, a relationship, and a parameter, and where the second entity represents a target entity, Where the model output is associated with the first output.

8. The method according to claim 7, wherein the determination of the match includes at least one of the following: Determine a one-to-one match between the first entity input and / or the first entity output and the second entity input and / or the second entity output; Determine a one-to-many match between the first entity input and / or the first entity output and the second entity input and / or the second entity output; Determine a many-to-one match between the first entity input and / or the first entity output and the second entity input and / or the second entity output; And Determine a many-to-many match between the first entity input and / or the first entity output and the second entity input and / or the second entity output.

9. The method according to any one of claims 1 to 8, wherein using the AI / ML model includes utilizing one of the following: joint embedding and nearest neighbor search, named entity recognition (NER), and graph neural network (Graph-NN).

10. The method according to any one of claims 1 to 9, further comprising: using a knowledge-based component associated with the AI / ML model, the knowledge-based component using an underlying knowledge representation system to utilize graph-containing information via the interface of the first entity and / or the interface of the second entity; and further considering the utilized graph-containing information for the mapping and / or the classification related to the interface of the first entity and / or the interface of the second entity.

11. The method according to any one of claims 1 to 10, further comprising: if the result indicated by the first output includes an uncertainty value equal to or higher than a predetermined uncertainty threshold, providing the result to a user for confirmation; and feeding back a decision received from the user on whether to accept the result to a knowledge base associated with the AI / ML model and / or to the AI / ML model.

12. The method according to any one of claims 1 to 11, wherein the content is related to at least one of concepts, naming, categories, relationships, parameters, units, method signatures, and method functions.

13. A controller for a mapping and / or classification control system, the controller being configured to perform the method according to any one of claims 1 to 12.

14. A mapping and / or classification control system configured to perform the method according to any one of claims 1 to 12.

15. An industrial automation system comprising the controller according to claim 13 and / or the mapping and / or classification control system according to claim 14.

16. A computer-readable medium comprising instructions that, when executed by a computing system, cause the computing system to perform the method according to any one of claims 1 to 12.

Images