Method and system for automatically determining a structure of a bill of process

Abstract

Systems and a method for determining a BOP structure. Input data on a workpackage and input data on a schema are received; wherein the workpackage comprises BOM data. The system provides access to a LLM module pre-trained with manufacturing data. The received input data is processed into a format comprising chunks of text understandable by the LLM module. The LLM module is fed with the processed input data to obtain, as output data, a set of determined BOP structures.

Description

METHOD AND SYSTEM FOR AUTOMATICALLY DETERMINING A STRUCTURE OF A BILL OF PROCESS TECHNICAL FIELD

[0001] The present disclosure is directed, in general, to computer-aided design,visualization, and manufacturing (“CAD”) systems, product lifecycle management (“PLM”) systems, product data management (“PDM”) systems, production environment simulation, and similar systems, that manage data for products and other items (collectively, “Product Data Management” systems or PDM systems). BACKGROUND OF THE DISCLOSURE

[0002] In the field of manufacturing process planning of a product, a key aspect is thedetermination of the Bill of Materials (“BOM”) and of the Bill of Process (“BOP”) whereby the BOM details the components that form a specific product, and the BOP details the planned production approach for the same specific product. For example, the BOM details the part pieces of the product to be produced and the BOM details the tasks which tell what BOM piece to produce in which station in the factory and which manufacturing tool to use.

[0003] Depending on the intended target users, there are various types of BOMs forexample the Engineering BOM (“E-BOM”) and the Manufacturing BOM (“M-BOM”). The E-BOM emphasizes how materials and parts relate to the product’s functional design, the M-BOM reflects how the materials and parts / components build a complete product, hereinafter each one referred also as X-BOM or simply BOM.

[0004] Nowadays, schema retrieval, creation, determination and management of BOPsare challenging tasks.

[0005] Operators who generate the BOPs need expertise, knowledge, and a significanttime investment. To create a BOP, they typically need to have a comprehensiveunderstanding of the manufacturing process including assembly sequence, work instructions, and quality control measures.

[0006] To optimize the tasks of BOP determination, manufacturing companies typicallymake use of solutions such as standardization of document structures, centralized repositories to store previous E-BOM and M-BOM, collaboration and knowledge sharing, digital tools, and software systems e.g. ERP, PLM systems or more in general PDM systems.

[0007] Such ERP, PLM, PDM systems provide specific functionalities for X-BOM / BOP management. For example, PLM systems provides comprehensive tools for managing product development and the associated data such as tools for creating and managing BOM / BOP structures, version control, and collaboration.

[0008] Notwithstanding the above existing solutions and the above systemfunctionalities, for an operator, the task of determining a BOP structure of a product is still a cumbersome and error-prone activity.

[0009] Improved techniques for determining a BOP structure are therefore desirable.SUMMARY OF THE DISCLOSURE

[0010] Various disclosed embodiments include methods, systems, and computerreadable mediums for determining a BOP structure. A method comprises receiving input data on a workpackage and input data on a schema; wherein the workpackage comprises BOM data. The method further comprises providing access to a Large Language Model (“LLM”) module pre-trained with manufacturing data. The method further comprises processing the received input data into a format comprising chunks of text understandable by the LLM module. The method further comprises feeding the LLM module with the processed input data to obtain, as output data, a set of determined BOP structures.

[0011] Various disclosed embodiments include methods, systems, and computerreadable mediums for providing a trained a LLM module to determine a BOP structure.A method comprises providing a training dataset including data of few-shot examples on BOM and BOP pairs and a vectorized database of workpackage and schema libraries. The method further comprises teaching the model of the LLM module to return from input data comprising workpackage and a schema a corresponding BOP structure matching the workpackage. The method further comprises providing the model of the LLM module to be used to automatically determine a BOP structure set.

[0012] A data processing system comprising a processor and an accessible memory ordatabase is also disclosed, wherein the data processing system is configured to implement any of the previously described methods.

[0013] The present invention proposes also a non-transitory computer-readable mediumencoded with executable instructions that, when executed, cause one or more data processing systems to perform any of the previously described methods.

[0014] Various disclosed embodiments include methods for producing a product byperforming the bill of processes of the determined BOP structure set.

[0015] The foregoing has outlined rather broadly the features and technical advantagesof the present disclosure so that those skilled in the art may better understand the detailed description that follows. Additional features and advantages of the disclosure will be described hereinafter that form the subject of the claims. Those skilled in the art will appreciate that they may readily use the conception and the specific embodiment disclosed as a basis for modifying or designing other structures for carrying out the same purposes of the present disclosure. Those skilled in the art will also realize that such equivalent constructions do not depart from the spirit and scope of the disclosure in its broadest form.

[0016] Before undertaking the DETAILED DESCRIPTION below, it may beadvantageous to set forth definitions of certain words or phrases used throughout this patent document: the terms “include” and “comprise,” as well as derivatives thereof, mean inclusion without limitation; the term “or” is inclusive, meaning and / or; the phrases “associated with” and “associated therewith,” as well as derivatives thereof, may mean toinclude, be included within, interconnect with, contain, be contained within, connect to or with, couple to or with, be communicable with, cooperate with, interleave, juxtapose, be proximate to, be bound to or with, have, have a property of, or the like; and the term “controller” means any device, system or part thereof that controls at least one operation, whether such a device is implemented in hardware, firmware, software or some combination of at least two of the same. It should be noted that the functionality associated with any particular controller may be centralized or distributed, whether locally or remotely. Definitions for certain words and phrases are provided throughout this patent document, and those of ordinary skill in the art will understand that such definitions apply in many, if not most, instances to prior as well as future uses of such defined words and phrases. While some terms may include a wide variety of embodiments, the appended claims may expressly limit these terms to specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] For a more complete understanding of the present disclosure, and the advantagesthereof, reference is now made to the following descriptions taken in conjunction with the accompanying drawings, wherein like numbers designate like objects, and in which:

[0018] Figure 1 illustrates a block diagram of a data processing system in which anembodiment can be implemented. Figure 2 schematically illustrates a flowchart for determining a BOP structure in accordance with disclosed embodiments.

[0019] Figure 3 schematically illustrates a block diagram for determining a BOPstructure in accordance with disclosed embodiments.

[0020] Figure 4 schematically illustrates a block diagram for BOP generation inaccordance with disclosed embodiments.

[0021] Figure 5 schematically illustrates a block diagram for data pre-processing inaccordance with disclosed embodiments.

[0022] Figure 6 schematically illustrates a block diagram for data vectorizing inaccordance with disclosed embodiments.

[0023] Figure 7 schematically illustrates a block diagram for querying BOP structuresin accordance with disclosed embodiments.

[0024] Figure 8 schematically illustrates a block diagram for training a module forformalizing a BOP query in accordance with disclosed embodiments.

[0025] Figure 9 schematically illustrates a block diagram for training a module forquerying a BOP structure in accordance with disclosed embodiments.

[0026] Figure 10 schematically illustrates a block diagram for training a module forgenerating a BOP structure in accordance with disclosed embodiments.

[0027] Figure 11 schematically illustrates a second example of block diagram fordetermining a BOP structure in accordance with disclosed embodiments. DETAILED DESCRIPTION

[0028] FIGURES 1 through 11, discussed below, and the various embodiments used todescribe the principles of the present disclosure in this patent document are by way of illustration only and should not be construed in any way to limit the scope of the disclosure. Those skilled in the art will understand that the principles of the present disclosure may be implemented in any suitably arranged device. The numerous innovative teachings of the present application will be described with reference to exemplary non-limiting embodiments.

[0029] Previous techniques for determining a BOP structure departing from the BOMstructures have some drawbacks.

[0030] The embodiments disclosed herein provide numerous technical benefits,including but not limited to the following examples.

[0031] Embodiments provide a smart system capable of optimizing the creation of anew BOP structure departing from E-BOM / M-BOM structures.

[0032] Embodiments provide a smart system capable of retrieving the most similarexisting BOP structure to the one about to be created based on X-BOM data in a fast and reliable manner.

[0033] Embodiments enable to automatically determine BOP structures departing fromBOM input data with high accuracy and relevance. Embodiments ensures that each BOM item is consumed.

[0034] Embodiments provide a smart system capable of generating a new BOP basedon natural text input and on E-BOM / M-BOM.

[0035] Embodiments allow easy retrieval and creation of a new BOP.

[0036] Embodiments enable BOP generation by means of search embeddings.

[0037] Embodiments makes use of the captured semantic patterns and relationshipsbetween different X-BOM / BOP schemas and structures.

[0038] Embodiments enhance the accuracy of the similarity search, ensuring that theretrieved BOP structures are truly relevant and closely aligned to the user's input.

[0039] Embodiments reduce the time and the efforts required for determining a BOPstructure.

[0040] Embodiments improve the quality of the product design and the productperformance.

[0041] Embodiments provide an improved user experience.

[0042] Indeed, prior art techniques for determining BOP structures do require by theusers many working hours and multiple data loadings. For example, in prior art techniques, users are often required to deal with multiple applications for managing the E-BOM, M-BOM, BOP data with multiple studies on each one. Examples of suchapplications include, but are not limited to, applications that manage that the M-BOM is consumed in the BOP, applications to search resource items in the resource library, applications to search for similar BOP structures in the workpackage libraries, applications to ensure the correctness of the process sequence, applications for load balancing a production line, applications for work instructions etc.

[0043] Embodiments streamlines the process, enabling users to access relevant BOPrecommendations efficiently and effectively.

[0044] In embodiments, users can provide inputs and receive relevant suggestions inintuitive and user-friendly manners.

[0045] Embodiments include a single running process via indexed and efficientdatabase calls that results in a new generated BOP.

[0046] Embodiments may advantageously result in a single click and some fine tuningsby the users.

[0047] Embodiments may reduce hardware consumption and the latency leading togains in terms of performances and sustainability.

[0048] Embodiments enable retaining the domain expertise and the domain knowledgewithin manufacturing companies.

[0049] Embodiments improve knowledge sharing, data consistency and / or datastandardization within manufacturing companies.

[0050] Embodiments, being based on a large amount of manufacturing training data,improve the scalability and the adaptability of the techniques for BOP determination.

[0051] Embodiments enable to determine a BOP structure in an efficient and intuitivemanner.

[0052] Figure 1 illustrates a block diagram of a data processing system 100 in which anembodiment can be implemented, for example as a PDM system particularly configured by software or otherwise to perform the processes as described herein, and in particular as each one of a plurality of interconnected and communicating systems as described herein. The data processing system 100 illustrated can include a processor 102 connected to a level two cache / bridge 104, which is connected in turn to a local system bus 106. Local system bus 106 may be, for example, a peripheral component interconnect (PCI) architecture bus. Also connected to local system bus in the illustrated example are a main memory 108 and a graphics adapter 110. The graphics adapter 110 may be connected to display 111.

[0053] Other peripherals, such as local area network (LAN) / Wide Area Network / Wireless (e.g. WiFi) adapter 112, may also be connected to local system bus 106. Expansion bus interface 114 connects local system bus 106 to input / output (I / O) bus 116. I / O bus 116 is connected to keyboard / mouse adapter 118, disk controller 120, and I / O adapter 122. Disk controller 120 can be connected to a storage 126, which can be any suitable machine usable or machine readable storage medium, including but are not limited to nonvolatile, hard-coded type mediums such as read only memories (ROMs) or erasable, electrically programmable read only memories (EEPROMs), magnetic tape storage, and user-recordable type mediums such as floppy disks, hard disk drives and compact disk read only memories (CD-ROMs) or digital versatile disks (DVDs), and other known optical, electrical, or magnetic storage devices.

[0054] Also connected to I / O bus 116 in the example shown is audio adapter 124, towhich speakers (not shown) may be connected for playing sounds. Keyboard / mouse adapter 118 provides a connection for a pointing device (not shown), such as a mouse, trackball, trackpointer, touchscreen, etc.

[0055] Those of ordinary skill in the art will appreciate that the hardware illustrated inFigure 1 may vary for particular implementations. For example, other peripheral devices, such as an optical disk drive and the like, also may be used in addition or in place of thehardware illustrated. The illustrated example is provided for the purpose of explanation only and is not meant to imply architectural limitations with respect to the present disclosure.

[0056] A data processing system in accordance with an embodiment of the presentdisclosure can include an operating system employing a graphical user interface. The operating system permits multiple display windows to be presented in the graphical user interface simultaneously, with each display window providing an interface to a different application or to a different instance of the same application. A cursor in the graphical user interface may be manipulated by a user through the pointing device. The position of the cursor may be changed and / or an event, such as clicking a mouse button, generated to actuate a desired response.

[0057] One of various commercial operating systems, such as a version of MicrosoftWindows™, a product of Microsoft Corporation located in Redmond, Wash. may be employed if suitably modified. The operating system is modified or created in accordance with the present disclosure as described.

[0058] LAN / WAN / Wireless adapter 112 can be connected to a network 130 (not a partof data processing system 100), which can be any public or private data processing system network or combination of networks, as known to those of skill in the art, including the Internet. Data processing system 100 can communicate over network 130 with server system 140, which is also not part of data processing system 100, but can be implemented, for example, as a separate data processing system 100.

[0059] Figure 2 illustrates a flowchart for determining, by a data processing system, aBOP structure.

[0060] At act 205, the system receives input data on a workpackage and on a schema. Inembodiments, the workpackage comprises at least BOM data.

[0061] At act 210, the system provides access to a LLM module pre-trained withmanufacturing data. Examples of LLM models include, but are not limited by, GPT-xmodels of OpenAI. Examples of manufacturing data include, but are not limited to, BOM / BOP data, schema data, workpackage libraries, schema libraries, vectorized databases and other data. In embodiments, the LLM module may be on the cloud or on the premise.

[0062] At act 215, the received input data is processed into a format comprising chunkof texts understandable by the LLM module.

[0063] At act 220, the system feeds the LLM module with the processed input data toobtain as output data a set of determined BOP structures.

[0064] In embodiments, the system receives additional input data on a user query in ahuman language format. The system receives access to data of a vectorized database of workpackage and schema libraries. In embodiments, at act 220, the system feeds the LLM module with the processed input data, with the additional input data and with data from the vectorized database.

[0065] In embodiments, at act 220, the set of determined BOP structures by the LLMmodule is a set of most similar BOP structures. For example, the most similar BOP structure may be determined via a similarity search between embeddings of the input data and data received from the vectorized database. Examples of metrics for similarity search calculations include, but are not limited by, cosine similarity metrics or Euclidean distance metrics.

[0066] In embodiments, at act 220, the set of determined BOP structures by the LLMmodule is a set of newly generated BOP structures.

[0067] In embodiments, at act 205 the workpackage additionally comprises high levelBOP data and / or the method further including receiving access to data of the vectorized database for feeding the LLM module. In embodiments, at act 220, the set of determined BOP structures by the LLM module is a set of newly generated BOP structures. For example, the newly generated BOP structures are obtained by detailing a high-levelstructure present in the workpackage and / or by taking into account the most similar BOP structures found in the vectorized database via the similarity search.

[0068] Thus, with embodiments, vectorized databases may also conveniently be used tocomplement the knowledge already available with the pre-trained LLM module.

[0069] With embodiments, a product is produced by performing the bill of processes ofthe BOP structure with a BOP or a sub-BOP structure. With the term product it is herein meant a product or a sub-product like e.g. a part or a component of a product which can be produced with a BOP or a sub-BOP.

[0070] In embodiments, the terms “received / receive / receiving”, as used herein, caninclude retrieving from storage, receiving from another device or process, receiving via an interaction with a user or otherwise.

[0071] As used herein the term “workpackage context” or simply “workpackage”denotes what in the art may also be called “context” or “BOP context”. The workpackage includes the information that process planners need to accomplish their digital planning work. For example, the workpackage includes info on the BOM context, on the BOP context, on the variant of the product, on the version of the product / process the X- BOM / BOP, on the work areas (factory stations), etc. As used herein, the workpackage comprises at least data on the BOM structure. In embodiments, the workpackage comprises also data on a high-level BOP structure. The X-BOM / BOP structure is a hierarchal representation with “parent-children” relations of the X-BOM / BOP data. For example, such structure can be displayed as “a tree” in the graphical user interface (“GUI”). An example of GUI with a graphical representation of BOM, BOP structures is shown in Figure 5 and other examples of BOM, BOP structures are shown in Tables 1 and 2.

[0072] As used herein the term “schema” e.g. X-BOM / BOP schema or PLM, ERP,PDM schema, comprises information and data detailing the business model schema, the modeling of the objects in the database, e.g. the properties of the BOM materials, BOP processes, stations, resource objects, relations between entities, etc. In other words, theschema comprises metadata information at the database level. For example, the schemacomprises data specifying the names of the objects e.g. Process-Station, Process-Line,Welding-Operation, Robot, etc. Additionally, the schema comprises data specifying theproperty of each object type, e.g. Process-Station has “Allocated Time, Station-Height,etc.”, Welding-Operation has properties such as: “Welding-Time, Welding-Robot,Welding-Gun, etc”. In other words, the BOM schema is a representation of the BOMobject in the database model, for example, the database has a “Load Assistant” objecttype having several properties such as: Type, Name, Max-Load Weight, etc. Inembodiments, the schema defines the structure and the organization of the X-BOM / BOP, specifying the components, sub-assemblies, and their relationships within the product.

[0073] Typically, in the art and as used herein, the term “X-BOM” refers to the “real”data / structure of the user (the hierarchal / 3D representation of the BOM) and what is included in the workpackage. With the terms “X-BOM schema” in the art it is typically meant the metadata information of the X-BOM objects and not their specific instances. In the art, some manufacturing company use E-BOM and BOP without M-BOM. In embodiments, the BOP structure may include the assembly sequence sequency and the manufacturing work instructions. In embodiments, the BOP structure may include the operation name, operation allocated time, its station, tool name, etc.

[0074] For example, the PLM schema may include the information, the description ofthe industry and the object types of the user. In other words, the schema does not denote the real data but rather the explanation of the data, what the data is about, e.g. the property types of the objects and not their values, e.g. what “loading a part” means, their relations in the database.

[0075] In embodiments, the schema is provided as inference input to generate a BOPstructure aligned with the schema. For example, the customer PLM schema includes data on the relation between a given operation and a part in the in the user’s database.

[0076] In embodiments, the workpackage and the schema may come from a variety ofdifferent sources.

[0077] For example, the source of the schema may come from the user environment ofthe user application although it may not be explicitly provided by the users themselves. In embodiments, the schema may be extracted from the user database. For example, users may have in their user interface applications some objects and structures called activities, processes, and BOPs. In embodiments, some of these data may be customized and / or added by the user.

[0078] For example, once the user has her workpackage package open, the PLMschema can be extracted from the PLM system. For example, in embodiments, the user may click a button in her user application and the system takes data on the workpackage and on the schema and send them to the preprocessing module. In embodiments, separate APIs are called to extract the PLM or ERP schemas i.e. the metadata.

[0079] Algorithms of exemplary embodiments

[0080] The main algorithm phases and steps for automatically determining a BOPstructure are illustrated in the following exemplary embodiments.

[0081] Figure 3 schematically illustrates a block diagram for determining a BOPstructure in accordance with disclosed embodiments.

[0082] As shown in Figure 3, the system 310 receives as input data the workpackagecontext data or workpackage 302 – preferably from an operator or user 301 typically a process planner– and it receives as input data also the schema data 303. In a pre-process module 309, at least the workpackage data 302 is pre-processed into formatted data 304 comprising chunks of text understandable by a previously trained LLM module 305. The LLM module 305 receives the text formatted data as input data 304 and returns as output a set of determined BOP structures 306. The LLM module 305 comprises a set of LLM modules. In embodiments, the LLM module 305 comprises a chain of LLMs (not shown). In embodiments, the LLM module 305 may optionally receive as input data a user query 307 in a human language format. In embodiments, the LLM module 305 may optionally exchange data with a vectorized database 308. In embodiments, the output BOP data 306 of the LLM module 305 may preferably be a newly generated BOP structure. In otherembodiments, the output BOP data 306 of the LLM module 305 may be a list of BOP structures (or their IDs for database retrieval) most similar to the requested user query 307.

[0083] Figure 4 schematically illustrates a block diagram for BOP generation inaccordance with disclosed embodiments.

[0084] As shown, the system 410 receives as input data the workpackage data 302 –preferably from a user 301 – and the schema data 303. As previously described, the workpackage and the schema data 302, 303 may come from different sources. Type of formatting of the input data 302, 303, 404 include, but are not limited to, XML format, Jason format, Excel format, text format and other types of formats. In embodiments, in a pre-process module 409, interacting with the PLM system 408, at least the workpackage data 302 is pre-processed into formatted data comprising chunks of text understandable by a previously trained LLM module 405. The LLM module 405 receives input data 404 and returns as output a newly generated BOP structure 406. The LLM module 405 comprises a set of LLM modules. The LLM module 405 includes an embedding module 415 whereby the preprocessed workpackage and schema data are embedded in a vectorized representation data 414. Types of embeddings performed in the embedding module 415 includes word embeddings whereby individual words are represented as vectors (e.g. word2vec technique or other techniques), sentence embeddings whereby full sentences are represented as vectors and bag of words embeddings whereby text is represented as a bag of words, where each word is assigned a unique ID. Examples of tools for embedding include, but are not limited to, OpenAI-Embedding from Open AI, Bedrock Embedding from AWS, Cohere Embedding, GPT4ALL from Hugging-Face and others.

[0085] In embodiments, the LLM module 405 comprises a chain of LLMs (not shown).For example, the LLM module 405 comprises two LLM modules (not shown) a Re-Rank LLM module and a Generator LLM module. The Re-Rank LLM module may for example comprise a GPT model, LLM agents and LLM chains and it receives as input the vectorized workpackage and schema data 414 and the closest BOM, BOP structures418 identified in the vectorized database 438 after the similarity search 428 has been performed between the workpackage embeddings and the data from the vectorized database 438. The vectorized database 438 may for example be a customer database which comprises two vectorized databases, the vectorized database of the library packages (e.g. BOM, BOP) and a vectorized database for the schema (e.g. PLM schema). In embodiments, other examples of libraries to be vectorized may be the classification and resources libraries. The Re-Rank LLM module returns as output the closest BOM, BOP structures which is one of the inputs of the Generator LLM module. In embodiments, the Generator LLM module comprises a GPT model, LLM agents and LLM chains. The Generator LLM module receives as inputs the vectorized workpackage and schema data 414 and the closest vectorized BOM, BOP structures 418. The Generator LLM module returns as output the newly generated BOP structure 406.

[0086] In embodiments, the user 301 provides input data 302 by loading theirworkpackage 302 which includes at least the BOM digital representation – e.g. E-BOM and / or M-BOM - where the data is preferably represented as hierarchal-tree and as 3D visualization. The workpackage loaded by the user may include also the contextual data (e.g. configuration), the plant digital representation (e.g. lines, stations of the factory, resources, tools needed to build the product) and, optionally, the current high-level BOP or in progress BOP. In embodiments, the output provided by the LLM module 405 is a detailed BOP 406 whereby the detailed BOP structure shall preferably consume all the BOM items using resources (e.g. tools and machines) of the plant. Additionally, the detailed BOP structure preferably includes additional information such as timing “planned time”, “simulated time” etc.

[0087] Figure 5 schematically illustrates a block diagram for data pre-processing inaccordance with disclosed embodiments.

[0088] In embodiments, a pre-processing phase is performed to prepare the inputdataset. In embodiments, a pre-processing phase is performed both for the training of the LLM modules as well as the input for inference during runtime prediction. For example, the user 301 loads the workpackage data 302 - as shown in the window 512 of the GUI -into a PLM system 408. The PLM system 408 receives the data on the schema 303 and it passes relevant PLM data 503 to the pre-processing module 509. For example, the workpackage data 302 may be exported into a data file 502 which may e.g. be in an Excel format 522. The pre-processing module 509 processes the input data 502, 503 by splitting and converting such data to obtain output data 504 organized in text chunks of operations / processes 514. The pre-processing module 509 returns as output data 504 the workpackage data in a format comprising chunk of texts understandable by the LLM module.

[0089] Figure 6 schematically illustrates a block diagram for data embedding inaccordance with disclosed embodiments.

[0090] In embodiments, the vectorized databases 308, 438, 608, 618 are obtained viaembedding techniques and tools in an embedding module 611, 612 preferably in a pre- preparation phase. Vectorized databases are dedicated database enabling to save and search embedding vectors resulting from the embedding models. Examples of vectorized databases include, but are not limited by, FAISS, Chroma, Weaviate, and other types of VecDB databases.

[0091] In Figure 6 two types of vectorized databases are shown, the vectorized databasefor the PLM schema 608 and the vectorized database for the library packages 618. In embodiments, vectorized databases – which are used during training phase - may come from customer specific libraries or generic standard libraries. In embodiments, vectorized databases - used during inference phase - may come from customer specific libraries or generic standard libraries.

[0092] In embodiments, the system extracts relevant information from the schemadatabase 603, it processes the data into raw-text (in Xml, Jason format or any other text format) and then sends the processed information data to the embedding module 611 which converts this information data to high-dimensional numeric vectors data 612.

[0093] The schema 603 data may comprise data on BOM schema, BOP schema,requirement schema, other schema and / or any other schema relating to the businessmodel for the PLM system, ERP system or any other PDM system. Such input data 603 is transformed into vectors 612 data via embedding techniques within the embedding module 611 which may comprise pre-processing steps and embedding tool steps to obtain a vectorized database 608 for the PLM schema.

[0094] In embodiments, the system extracts relevant information 602 from historicalworkpackage libraries with workpackage structures, it processes the data into raw-text (e.g. in Xml format, Jason format and / or any other text format) and then sends the processed information data to the embedding module 621 which converts this information data into high-dimensional numeric vectors data 622. The workpackage data 602 preferably includes data on packages of E-BOM, M-BOM, BOP, in general library packages with end customer data e.g. historical packages library from the system, from the admin or from background activity. The input workpackage data 602, and the input schema data 603 are transformed into vectors via pre-processing and embedding tools steps by the embedding module 621 to obtain a vectorized database 618 for the library packages.

[0095] For example, in embodiments, engineering data such as BOM / BOP datastructures are converted into raw text (e.g. in Xml format, Jason format and / or any other text format) and then by using an embedding model e.g. OpenAI Embedding convert such data is converted into numeric vectors which can be stored in vectorized database which are conveniently capturing semantic relationship of the various engineering data.

[0096] Figure 7 schematically illustrates a block diagram for querying BOP structuresin accordance with disclosed embodiments.

[0097] A user 301 requests a user query 707 to a system 710 in a human languageformat. The user query 707 is transformed into a formalized query 717 by the Formalize- Query module 730. The Formalize-Query module 730 is a pre-trained LLM module capable of formalizing the user query. A formalized query 717 is a query which is a structured query - a PLM / ERP / PDM model-oriented query - herein simply called PLM-oriented query. An example of user’s query 707 is “Find me an operation loading a partover 6 Kg” and an example of formalized query or PLM query 717 is “Find ‘object-type’=‘Operation’ and has child of ‘object-type’=‘Part’ and child has ‘Item Weight’>=6”. In embodiments, the formalize query module 730 receives as input data, via the PLM system 408, data on the schema 703 and data on the workpackage 702. Additionally, in embodiments, the data on the schema 703 and data on the workpackage 702 are pre-processed by a pre-process module 709 in suitably formatted data 719 to feed the BOP query LLM module 740. The Query LLM module 740 returns as output 706 a set of IDs of BOP structures meeting the user’s query 707. The BOP Query LLM module 740 is a pre-trained customized LLM module comprising LLM agents and Chain of Thought prompting (“CoT”).

[0098] In embodiments, in case the user has additionally data on historical BOMs andBOPs – for example in a workpackage library and in a workpackage VecDB (not shown) - and they are working on an on-going product via an on-going BOM / BOP, the system via the Query LLM module 740 determines the most similar historical BOP structures, whereby, for example, the LLM module 740 returns a list of IDs 706 of relevant BOP results, then the application opens the BOP objects of the IDs from the database or library using its own API and / or SDK.

[0099] As above illustrated, the BOP structures 406, 706 can be determined by a set ofpre-trained LLM modules 405, 730, 740 via query by the LLM modules 730, 740 (e.g. as shown in the exemplary embodiment of Figure 7) or by generating a new BOP structure by a set of pre-trained LLM modules 405 (e.g. as shown in the exemplary embodiment of Figure 4). Examples embodiments for training such set of LLM modules 405, 730, 740, are illustrated below by means of Figures 8 to 10.

[0100] Figure 8 schematically illustrates a block diagram for training a module forformalizing a BOP query in accordance with disclosed embodiments.

[0101] The LLM model used for inference in the formalize query module 730 of Figure7 is pre-trained. Figure 8 shows an exemplary diagram for training such a Formalize Query LLM module 830. To train the LLM model of the Formalize Query module 830, the system utilizes few shots examples 840. Simplified few shot examples are shown in Table 1 below.

[0102] { "input": "Find me an operation longer than 10 sec", "output": "Find me node. 'Item type'= 'Operation' node. 'Time' >10 sec } { "input": "find an operation loading a part with load assist", "output": "find node. 'Item type'= 'Operation' node.children include … }, { "input": "find an operation loading a part over 6kg", "output": "find node. 'Item type'= 'Operation’ and node.children … }

[0103] Table 1: few shot examples to train the Formalize Query module 830

[0104] To train the LLM model of the Formalize Query module 830 (whichcorresponds to the LLM model module 730 used for inference in Figure 7), the system checks 850 if the outcome data 831 of the formalized query are satisfactory otherwise itrepeats 860 until the results are good enough. An example of user query 807 is “Findme an operation loading a part over 6 Kg” and an example of formalized query 831, 832is “Find ‘object-type’=‘Operation’ and has child of ‘object-type’=‘Part’ and child has‘Item Weight’>=6”.

[0105] Figure 9 schematically illustrates a block diagram for training a module forquerying a BOP structure in accordance with disclosed embodiments.

[0106] The LLM model used for inference in the module 740 of Figure 7 is a pre-trained model. Figure 9 shows an exemplary diagram for training such a BOP Query LLM module. In embodiments, to train the BOP Query module 960, the system utilizes few shot examples 941 and manufacturing documents 942.

[0107] In embodiments, the system receives workpackage data 902 and a user queryrequest 907 by a user 901.

[0108] The BOP Query module 960 comprises a customized LLM e.g. LLM agents andCoT. The BOP Query module 960 receives as input data the formalized query 931 and the pre-processed data 934 for example workpackage 902 and schema 903. An exampleof user query 907 is “Find me an operation loading a part over 6 Kg” and an example offormalized query 931, is “Find ‘object-type’=‘Operation’ and has child of ‘object-type’=‘Part’ and child has ‘Item Weight’>=6”.

[0109] To train the LLM model of the module 960, the system checks 950 if theoutcome data the BOP IDs 932 of the BOP query are satisfactory otherwise it repeats 970 until the new resulting data 933 are good enough. The outcome data 932 is the closes BOP structure matching the user query e.g. an ID or list of IDs of BOP structures or BOP objects meeting the user query 907.

[0110] Figure 10 schematically illustrates a block diagram for training a module forgenerating a BOP structure in accordance with disclosed embodiments.

[0111] The LLM models used for inference in the module 305 of Figure 3 or in themodule 405 of Figure 4 are models pre-trained with manufacturing data. Figure 10 shows an embodiment of exemplary diagram for training a module 1005 for generating a new BOP structure 1006. In embodiments, to train the BOP generation module 1005, the system utilizes few-shot examples 1041 e.g. pairs of X-BOM structures with corresponding BOP structures. The few shot examples 1041 comprise examples of BOP structures generated departing from BOM data (such as e.g. E-BOM and M-BOM). The BOM generation module 1005 is trained also with vectorized data coming from the vectorized library packages database 1022. This vectorized database 1022 is obtained by processing data in the embedding module 1021 to obtain embeddings whereby the input data comprise the PLM schema 1003 and library packages 1012 i.e. E-BOM, M-BOM, BOP packages for example from customers and from Q&A libraries. The BOP generation module 1005 comprises a customized LLM comprising e.g. GPT model, LLM agents, LLM chains and Re-Ranks.

[0112] In embodiments, exemplary algorithms for training a LLM module to determinea BOP structure, may comprise the following steps: providing a training dataset includingdata few-shot examples on BOM and BOP pairs and a vectorized database of workpackage and schema libraries; teaching the model of the LLM module to return from input data comprising workpackage and a schema a corresponding BOP structure matching the workpackage; and, providing the model of the LLM module to be used to automatically determine a BOP structure.

[0113] Figure 11 schematically illustrates a second example of block diagram fordetermining a BOP structure in accordance with disclosed embodiments.

[0114] In embodiments, the system uses search embeddings 1101 to feed the customer'sdata into the LLM system 1110. The system is able of suggesting similar BOP structures 1106 also based on natural language inputs from the user (not shown). An example ofnatural language inputs received by the user could be a text like “generate for me themanufacturing process of this sub-BOM based on my current stage” (e.g. audio or text) or the user may press a button and the system opens a User Interface (“UI”) form where the user fills the data about the BOM, and / or the historical BOM and the current BOP.

[0115] Embodiments provide a smart system which generates a new BOP based on auser’s input (not shown) and on E-BOM / M-BOM. For example, a natural language text may be that the user requests a free-text query in plain English to generate a BOP of sub- component of E-BOM and M-BOM. For instance, the user opens the X-BOM and the in- progress BOP. User selects some component from the X-BOM and asks «suggest me a way to create a process for this component» e.g. in audio or text. Accordingly, with embodiments, the system generates a sub-process in the context of the BOP to assemble the selected X-BOM.

[0116] In embodiments, the customer’s data may comprise BOM, BOP, configurationdata which is fed automatically to the LLM system. In embodiments, the system asks the user to provide a link to the historical BOM / BOP data or the system uses SDK / APIs to get such data e.g. the user provides an ID of the previous BOM / BOP.

[0117] The system via the LLM module 1110 is able to analyze and to understand thespecific customer’s requirements, providing relevant recommendations to create a newBOP 1106. For example, some customers’ requirements include naming requirements like e.g. a naming requirement that each manufacturing process starts with: <ProductName>-<ver>-<Name> for instance: “BMW6ix-45A-Loading”. Advantageously, theengine of the system takes care of such naming requirements and other customer requirements and determines a BOP structure with such requirements.

[0118] The search embeddings 1101 are representations of data that capturerelationships. The system reads the user data 1121, 1222 e.g. the schema of the E-BOM, M-BOM and converts in an embedding module 1135, 1136 this data into raw data e.g. key-value pairs. The embedding mechanism converts raw-text data into high dimension vector of numbers. The system then use search embedding techniques 1140 to find similarity between the data.

[0119] According to a pre-process phase, from the E-BOM / M-BOM schema 1121 anembedding is performed also by taking into account the available libraries 1150. An Artificial Intelligence (“AI”) module may preprocess in an embedding module 1135 the data 1122 from the available E-BOM / M-BOM and mapped BOP libraries 1150 within the customer's scope. In embodiments, E-BOM / M-BOM structure data within the libraries are processed based on a schema 1121, and the processed data is then saved in an embedding database (not shown). In embodiments, customers may be able to customize or to extend a default schema with their own preferences. In the pre-processing phase, the defined schema 1121 e.g. E-BOM / M-BOM schema is processed by an embedding module 1136.

[0120] According to a similarity search inference, the most similar BOP structure 1106is retrieved e.g. ID of the BOP structure. Based on input data 1121, 1102, the AI module may retrieve the closest / most similar BOP structure between the vector representation of a hypothetic BOP whereby the hypothetic BOP structure is generated by an LLM module 1110.

[0121] In embodiments, steps of the similarity search phase include:- transforming a hypothetical BOP structure into a vector representation using an LLM module;- performing a vector comparison between the vector representation of the hypothetical BOP structure and the vector representation of the existing BOM / BOP structures stored in the embedding database (not shown); - performing a similarity calculation 1140 whereby a similarity metric is applied to measure the similarity between the vector representations. Examples of similarity metrics include, but are not limited by, cosine similarity, Euclidean distance etc. - retrieving the BOP structure 1106 with the highest similarity score is retrieved as the closest / most similar match 1106 to the hypothetical BOP.

[0122] In summary, the system retrieves the most similar BOP from the existinglibrary based on the vector representation of a hypothetical BOP generated by the LLM module.

[0123] In embodiments, the input data 1121, 1102 for the similarity search includesthe BOM schema 1121 and inference input i.e. the current workpackage context 1102. In embodiments, the BOM schema 1121 is the schema of the E-BOM / M-BOM. In embodiments, the current workpackage context 1102 is the current work-package data loaded by the user and it may include BOM CAD files, additional external BOM / BOP data, contextual BOP data. The E-BOM / M-BOM CAD files or other type of BOM description files comprise detailed information about the individual components, including specifications, dimensions, materials, and other relevant attributes. The BOM / BOP external data can also be included as input 1102 and comprises information such as supplier details, manufacturing processes, cost data, lead times, and any other relevant information helpful for identification of most similar BOP. Contextual BOP data may include information like industry domain, product category, intended application, or any other contextual details helpful for identification of most similar BOP.

[0124] In embodiments, based on the similarity input and by including the retrievedBOP an additional AI module e.g. a LLM module, previously trained with given templates, prompt engineering, few-shot learning examples, and designated output schemas, generates a new BOP 1106, connected to the given E-BOM / M-BOMspecifications. In embodiments, the new BOP may be generated without the similarity input.

[0125] Exemplary embodiments of similarity search of BOMs for determining BOPs

[0126] In simplified exemplary embodiments, steps for performing a similarity searchfor determining a BOP structure are illustrated below with the help of Tables 2 to 5.

[0127] Table 2 is an example of a first user workpackage for producing a standardwheel, more specifically it includes an E-BOM structure eBOM#1 and a correspondingBOP structure BOP#1 for the standard wheel.

[0128] Table 3 is an example of a second user workpackage for producing anadvanced wheel, more specifically it includes an E-BOM structure ebOM#2 and acorresponding high level and a detailed BOP structure BOP#2.

[0129] Table 4 shows the vector representations of the calculated embeddingsBOM#1-v, BOM#2-v, BOP#1-v of the corresponding BOM / BOP structures eBOM#1,eBOM#2, BOP#1.

[0130] Table 5 shows the results of the calculated cosine similarity table for theBOM / BOP structures BOM#1, BOM#2, BOP#1.

[0131] As shown in the example taken from Tables 2 to 5, the two E-BOM structureseBOM#1, eBOM#2 are considered to be similar to each other since they have a cosine similarity value greater than 0.6 (see metric computation results in Table 5 below). Inembodiments, the system determines for eBOM#2, an existing or a new BOP which isbased on the similarity to BOP#1.

[0132] In embodiments, the trained LLM module has been trained with manufacturingdata comprising for example, historical data from libraries with dataset sample <input data; output data> pairs like e.g. <X-BOM; BOP>; <Existing sub-BOM; Matching Sub- Process>; <sub E-BOM + sub M-BOM; sub BOP>. In embodiments, during inference, the LLM module determines a BOP structure based on data of existing BOMs, either by suggesting an existing similar BOP responding to a user query or by suggesting a newlygenerated BOP structure also by taking into account the found most similar BOP structure. For example, the user loads into the system the new X-BOM, the LLM module determines a sub-BOP for the given sub-X-BOM. In embodiments, the system finds via a similarity search the most similar old X-BOM and then it provides to the user the relevant old BOP. In embodiments, the system may also feed a LLM module again with the new X-BOM and the old similar BOP so that the system is able to generate a new BOP.

[0133] eBOM#1: description: standard wheel 1. Wheel x4 (4 wheels) - id: wheel-001 1.1 Rim - id: rim-001 - desc: outer circular part, holding the tire - material: alloy - size: 17 inches 1.1.1 Spoke x 24 - id: spoke-001 - desc: metal rods connecting the rim to the hub - material: steel - quantity: 24 1.2 Tire - id: tire-001 - desc: rubber part, in contact with the road - type: all-season radial - size: 225 / 60R17 1.3 Hub Assembly - id: hub-assembly-001 - desc: central part, connecting it to the vehicle's axle - material: steel 1.3.1 Bearings x2 - id: bearing-001 - desc: bearings for smooth rotation (within the hub assembly) - material: steel - quantity: 2 BOP#1: 1. Wheel Process - id: wheel-process-001 - desc: BOP, entire process to assemble the wheel 1.1. Rim- id: rim-process-001 - desc: assemble rim BOM… - consumption: rim-001 - tool: tool-rim-0001 1.2. Spoke- id: spoke-process-001 - consumption: spoke-001 x 24 - tool: tool-spoke-001 1.3.Tire - id: tire-process-001 - consumption: tire-001 - tool: tool-tire-0001 1.4. Hub Assembly1.5. Wheel Assembly1.6. Quality Inspection

[0134] Table 2: simplified eBOM#1 and BOP#1 for producing the standard wheel

[0135] eBOM#2: - description: advanced wheel “TerraGrip Off-Road Wheel” designed for adventurous off-road enthusiasts seeking rugged performance & durability 1. TerraGrip Off-Road Wheel - id: terra-grip-001 - desc:… 1.1.TerraGrip Rim - id: terra-grip-rim-001 - desc: robust rim engineered for off-road challenges, featuring reinforced construction & protection - material: heavy-duty steel - thickness: zzz - size: 18 inch 1.2.TerraGrip All-Terrain Tire - id: terra-grip-tire-001 - desc: Specially designed all-terrain tire with aggressive tread for superior traction on various surfaces. - material: rugged rubber compound - protection-level: 2 1.3.TerraGrip Hub Assembly - id: terra-grip-hub-assembly-001 - desc: The hub assembly tailored for off-road adventures, providing robust performance & reliability. - material: heavy-duty cast iron - load: 234 - water-resistance: 12 1.3.1 TerraGrip Locking Differential -id: terra-grip-locking-diff-001- desc: locking differential mechanism for improved off-road traction & control - material: high-strength steel gears - control-type: manual / auto BOP#2: (high level BOP) 1. TerraGrip Off-Road Wheel Process - id: terra-grip-001 - desc: BOP, entire process to assemble TerraGrip Off-Road Wheel BOP#2: (detailed BOP) 1. TerraGrip Off-Road Wheel - id: terra-grip-001 - desc: BOP, entire process to assemble TerraGrip Off-Road Wheel 1.1.Process: Rim Manufacturing - id: terra-grip-trim-process-001 - desc: - consumed eBOM Item: terra-grip-001 - Tools / Equipment: - Heavy-duty steel rim manufacturing equipment - Beadlock design machinery 1.2.Process: All-Terrain Tire Production - id: terra-grip-tire-bop-001 - desc: bla - consumed eBOM: terra-grip-tire-001 - Tools / Equipment: - Rugged rubber compound mixing and molding machinery - Advanced tread pattern design - Sidewall protection integration 1.3.Process: Hub Assembly Process - id: terra-grip-hub-assembly-process-001 - desc: assemble the hub assembly eBOM using - consumed eBOM Item: terra-grip-hub-assembly-001 - Tools / Equipment: - Heavy-duty cast iron hub manufacturing equipment - Sealed bearing installation - Axle and CV joint assembly 1.3.1. Sub-Process: Locking Differential Installation - id: lock-diff-bop-001 - Relevant eBOM Item: lock-diff-bop-001 - Tools / Equipment: - High-strength steel gear installation - Locking differential mechanism setup 1.4.Process: Wheel Assembly - id: wheel-assembly-process-001 - desc: assemble all pieces together - consumed eBOM Items: TerraGrip Rim, TerraGrip All-Terrain Tire, TerraGrip Hub Assembly (with Locking Differential) - Tools / Equipment: -Wheel assembly fixture- Torque wrench for secure assembly 1.5.Process: Quality Inspection 1.6.Comprehensive inspection station

[0136] Table 3: simplified eBOM#2 and BOP#2 for producing the advanced wheel.eBOM#1-v: [-0.26631, 0.45589, -0.054918, -0.31348, -0.16587, 0.029911, 0.073169, -0.18743, - 0.10874, -0.057634, 0.0625, -0.23889, 0.28262, -0.2281, -0.10088, 0.14192] eBOM#2-v: [-0.052983, 0.36244, -0.043515, -0.28736, -0.30588,-0.12885, 0.10736, -0.1093, -0.042424, -0.05617, 0.20799, -0.14509, 0.33379, -0.11133, 0.013576, 0.10575] BOP#1-v: [-0.24305, 0.33712, 0.13151, -0.29161, 0.087039,-0.01543, 0.011432, -0.32261, -0.092132, 0.0021029, 0.2443, -0.14418, 0.2176, -0.2414, 0.18339, 0.10454]

[0137] Table 4: calculated embeddings BOM#1-v, BOM#2-v, BOP#1-vebom#1 ebom#2 BOP#1ebom#1 1.0000 0.6319 0.3933ebom#2 0.6319 1.0000 0.3806BOP#1 0.3933 0.3806 1.0000

[0138] Table 5: calculated cosine similarity table

[0139] Of course, those of skill in the art will recognize that, unless specificallyindicated or required by the sequence of operations, certain steps in the processes described above may be omitted, performed concurrently or sequentially, or performed in a different order.

[0140] Those skilled in the art will recognize that, for simplicity and clarity, the fullstructure and operation of all data processing systems suitable for use with the present disclosure is not being illustrated or described herein. Instead, only so much of a data processing system as is unique to the present disclosure or necessary for an understanding of the present disclosure is illustrated and described. The remainder of the construction and operation of data processing system 100 may conform to any of the various current implementations and practices known in the art.

[0141] It is important to note that while the disclosure includes a description in thecontext of a fully functional system, those skilled in the art will appreciate that at least portions of the present disclosure are capable of being distributed in the form of instructions contained within a machine-usable, computer-usable, or computer-readable medium in any of a variety of forms, and that the present disclosure applies equally regardless of the particular type of instruction or signal bearing medium or storage medium utilized to actually carry out the distribution. Examples of machine usable / readable or computer usable / readable mediums include: nonvolatile, hard-coded type mediums such as read only memories (ROMs) or erasable, electrically programmable read only memories (EEPROMs), and user-recordable type mediums such as floppy disks, hard disk drives and compact disk read only memories (CD-ROMs) or digital versatile disks (DVDs).

[0142] Although an exemplary embodiment of the present disclosure has beendescribed in detail, those skilled in the art will understand that various changes, substitutions, variations, and improvements disclosed herein may be made without departing from the spirit and scope of the disclosure in its broadest form.

[0143] None of the description in the present application should be read as implyingthat any particular element, step, or function is an essential element which must be included in the claim scope: the scope of patented subject matter is defined only by the allowed claims.

Claims

WHAT IS CLAIMED IS:

1. A method for determining, by a data processing system, a Bill of Process structure, herein called BOP structure, comprising the following steps: a) receiving input data on a workpackage and input data on a schema; wherein the workpackage comprises Bill of Material data, herein called BOM data; b) providing access to a Large Language Model module pre-trained with manufacturing data, herein called LLM module; c) processing the received input data into a format comprising chunks of text understandable by the LLM module; d) feeding the LLM module with the processed input data to obtain, as output data, a set of determined BOP structures.

2. The method of claim 1 further comprises: - receiving additional input data on a user query in a human language format; - receiving access to data of a vectorized database of workpackage and schema libraries; - wherein in item c) the LLM module is fed with the processed input data, with the additional input data, with data from the vectorized database; - wherein in item d), the set of determined BOP structures by the LLM module is a set of most similar BOP structures to the input data.

3. The method of claim 2, the set of most similar BOP structures is determined by performing a similarity search between embeddings of the input data and data received from the vectorized database 4. The method of claim 1, wherein in item d), the set of determined BOP structures is a newly generated BOP structure.

5. The method of claim 1, wherein in item a) the workpackage additionally comprises high level BOP data and / or the method further including receiving access todata of the vectorized database for feeding the LLM module; and, wherein in item d), the set of determined BOP structures is a newly generated BOP structure.

6. A method for producing a product by performing the bill of processes of the BOP structure set determined with any of the previous claims.

7. A data processing system comprising: a processor; and an accessible memory, the data processing system particularly configured to: a) receive input data on a workpackage and input data on a schema; wherein the workpackage comprises BOM data; b) provide access to a LLM module pre-trained with manufacturing data; c) process the received input data into a format comprising chunks of text understandable by the LLM module; d) feed the LLM module with the processed input data to obtain, as output data, a set of determined BOP structures.

8. The data processing system of claim 7 further configured to: - receive additional input data on a user query in a human language format; - receive access to data of a vectorized database of workpackage and schema libraries; and, wherein: - in item c) the LLM module is fed with the processed input data, with the additional input data, with data from the vectorized database; - in item d), the set of determined BOP structures by the LLM module is a set of most similar BOP structures to the input data.

9. The data processing system of claim 8, the set of most similar BOP structures is determined by performing a similarity search between embeddings of the input data and data received from the vectorized database10. The data processing system of claim 7, wherein in item d), the set of determined BOP structures is a newly generated BOP structure.

11. The data processing system of claim 7, wherein in item a) the workpackage additionally comprises high level BOP data and / or the method further including receiving access to data of the vectorized database for feeding the LLM module; and, wherein in item d), the set of determined BOP structures is a newly generated BOP structure.

12. A non-transitory computer-readable medium encoded with executable instructions that, when executed, cause one or more data processing systems to: a) receive input data on a workpackage and input data on a schema; wherein the workpackage comprises BOM data; b) provide access to a LLM module pre-trained with manufacturing data; c) process the received input data into a format comprising chunks of text understandable by the LLM module; d) feed the LLM module with the processed input data to obtain, as output data, a set of determined BOP structures.

13. The non-transitory computer-readable medium of claim 12 further configured to: - receive additional input data on a user query in a human language format; - receive access to data of a vectorized database of workpackage and schema libraries; and wherein: - in item c) the LLM module is fed with the processed input data, with the additional input data, with data from the vectorized database; - wherein in item d), the set of determined BOP structures by the LLM module is a set of most similar BOP structures to the input data.

14. The non-transitory computer-readable medium of claim 13, the set of most similar BOP structures is determined by performing a similarity search between embeddings of the input data and data received from the vectorized database15. The non-transitory computer-readable medium of claim 12, wherein in item d), the set of determined BOP structures is a newly generated BOP structure.

16. The non-transitory computer-readable medium of claim 12, wherein in item a) the workpackage additionally comprises high level BOP data and / or the method further including receiving access to data of the vectorized database for feeding the LLM module; and, wherein in item d), the set of determined BOP structures is a newly generated BOP structure.

17. A method for providing, by a data processing system, a trained a LLM module to determine a BOP structure, comprising the following steps: - providing a training dataset including data of few-shot examples on BOM and BOP pairs and a vectorized database of workpackage and schema libraries; - teaching the model of the LLM module to return from input data comprising workpackage and a schema a corresponding BOP structure matching the workpackage: - providing the model of the LLM module to be used to automatically determine a BOP structure set.

Images