Method and system for generating material lists and corresponding process lists for products

Abstract

A method and system for generating a list of materials and a corresponding process list for a product. To generate a material list and a corresponding process list for a product, a natural language processing module maps at least one document specifying the product into a potential representation that is embedded in a potential space. A bill of materials generator receives the potential representation as input and generates a bill of materials for the product. A user interface (UI) detects a user interaction with the generated bill of materials and creates a planned bill of materials depending on the user interaction. A process manifest generator receives the potential representation and the planned manifest of materials as input and builds a process manifest for the product. The method and system, or at least some embodiments thereof, provide semi-automatic generation of a material list and process list for a product using an AI engine. The method of generating this information using AI can significantly accelerate the process of calculating product cost and sustainable value compared to manual setting of a list of materials and / or a list of processes in a software system.

Description

TECHNICAL FIELD

[0001] The present application relates to generating a bill of materials and a corresponding bill of processes for a product. BACKGROUND

[0002] In industry, it is important to have transparency on the cost and sustainability value, such as carbon, of a product. Typically, transparency on cost and sustainability value is achieved by defining a bill of materials (BOM) in combination with a bill of processes (BOP), and then calculating and aggregating cost and sustainability value along the BOM / BOP structure of the product to obtain the total cost / carbon of the product. In this sense, the BOM / BOP can be seen as a digital twin of the product in terms of the materials and manufacturing processes needed to manufacture it, including any direct or indirect overheads contributing to it.

[0003] The requirement to calculate cost and sustainability value has been starting at the engineering and design phase of a product. When designing a new product, an engineer can encounter different potential ways on how to implement the product, which materials can be used, and which manufacturing processes can be required. By obtaining qualified information on the expected cost of different potential implementations, and how a particular implementation will impact the value of sustainability, the engineer is enabled to make design decisions at an early stage of the product development. In this way, products can be optimized for cost and their impact on our environment long before they enter manufacturing.

[0004] Another use case for product cost calculation is the estimation of product cost and sustainability value in the procurement department. In this field, a procurement user wants to quickly obtain qualified information on the cost and sustainability value of procured parts and use this information in price negotiations with potential suppliers.

[0005] Typically, when a procurement user is interested in estimating the cost of a product, he can not even have a bill of materials readily available to start calculating cost and sustainability value. Therefore, the procurement user needs to create a bill of materials from scratch in the system in order to calculate it bottom-up, so that he can use this calculation during price negotiations with suppliers.

[0006] For example, if a product is assembled of 200 different parts, the procurement user needs to create all 200 parts one by one in his calculation software in order to manually create a digital twin of the product.

[0007] In addition to the bill of materials, the bill of processes also needs to be set up manually. Each single manufacturing step needed to produce the product needs to be defined individually. This requires detailed knowledge of the different manufacturing techniques used in the production of that particular product.

[0008] On the other side of the procurement price negotiation, there is usually a supplier that needs to provide a detailed cost breakdown to his customer when replying to a request for quote. In addition to that, the supplier needs to make sure that certain key performance indicators regarding the profitability and risk associated with the project are met. In this way, reliable data can be provided to the management, who can decide based on these data whether a project should be acquired or not.

[0009] For internal project control, the bill of materials is usually created by the R&D department during the build and design phase of a product in a software tool like Teamcenter NX. Nevertheless, there are use cases where a detailed bill of materials might not be available in the early stages of the product lifecycle.

[0010] In addition, for this use case, the process list needs to be set up in a manual way, where each different manufacturing step needs to be defined. Setting up the manufacturing process needs to be performed by manufacturing experts and requires in-depth knowledge of the manufacturing technology.

[0011] The bill of materials and the process list are usually set up in a dedicated software for calculating the cost and sustainability value. There are several calculation tools on the market, starting from the widely used Microsoft Excel to more complex calculation tools such as Teamcenter Product Cost Management (TcPCM), Facton, aPriori, and others.

[0012] Usually, in any of these tools, the bill of materials and the process list need to be set up manually. In some cases, the bill of materials might be imported from a third-party system like Teamcenter NX or other 3D-based design software.

[0013] There are also solutions on the market that use AI models to predict the price of a product for certain manufacturing technologies like milling, drilling, and turning. Usually, these solutions require a 3D model in the form of a CAD or JT file and are designed specifically for certain manufacturing technologies. SUMMARY

[0014] It is an object of the present invention to identify one problem in the prior art and to find a technical solution to this problem.

[0015] The invention is defined in the independent claims. Further advantageous arrangements and embodiments of the invention are set out in the dependent claims.

[0016] According to the method for generating a bill of materials and a corresponding process list for a product, the following operations are performed by a module, wherein the module is a hardware module and / or a software module executed by one or more processors:

[0017] - mapping, by a natural language processing module, of at least one document specifying a product into a latent representation, the latent representation being an embedding in a latent space,

[0018] - generating, by a bill of materials generator receiving the latent representation as input, a bill of materials for the product,

[0019] - outputting, by a user interface, the generated bill of materials,

[0020] - detecting, by the user interface, a user interaction with the generated bill of materials and creating, depending on the user interaction, a curated bill of materials, and

[0021] - constructing, by a process list generator receiving the latent representation and the curated bill of materials as input, a process list for the product.

[0022] The system for generating a bill of materials and a corresponding process list for a product comprises the following modules, wherein a module is a hardware module and / or a software module executed by one or more processors:

[0023] - a natural language processing module configured for mapping of at least one document specifying a product into a latent representation, the latent representation being an embedding in a latent space,

[0024] - a bill of materials generator configured for receiving the latent representation as input and generating a bill of materials for the product,

[0025] - a user interface configured for outputting the generated bill of materials, detecting a user interaction with the generated bill of materials and creating, depending on the user interaction, a curated bill of materials, and

[0026] - a process list generator configured for receiving the latent representation and the curated bill of materials as input and constructing a process list for the product.

[0027] The following advantages and explanations are not necessarily the result of the target of the independent claims. Rather, they can be advantages and explanations that apply only to certain embodiments or variants.

[0028] The term “computer” is to be interpreted broadly as encompassing any electronic device having data processing capabilities. Examples include, but are not limited to, a personal computer, a server, a client, a programmable logic controller (PLC), a handheld device, a palmtop PC, a mobile device such as a smartphone, or other communication device capable of processing data. A computer can comprise one or more processors and memory units, such as volatile memory (e.g. RAM) or non-volatile memory (e.g. solid state drive, hard disk, flash memory).

[0029] In the context of embodiments of the present invention, the terms "memory," "memory cell," or "memory module" refer to any suitable storage medium. Examples include volatile memory (e.g., random access memory (RAM)) or non-volatile memory (e.g., solid-state drives, hard disks, flash memory cards, or optical disks).

[0030] While bills of materials (BOMs) are typically created during the design process, process lists are not available at this stage of product development. Despite both, BOMs and process lists are necessary to calculate and estimate the cost and sustainability value of a product.

[0031] This method and system, or at least some embodiments thereof, can enhance the bill of materials (BOM) by providing engineers with qualified information about cost and sustainability value through the manufacturing processes required to manufacture and assemble the product. This is particularly helpful if product engineers lack sufficient knowledge of the details of the required manufacturing processes during product development to manually define them in a manner that delivers qualified results. Furthermore, manually creating the BOM and / or process list in the system is extremely time-consuming and cumbersome, as any information needed for meaningful calculations must be manually entered into the system. In addition, a deep understanding of all the manufacturing processes required to produce and assemble the product is required. Different user groups (such as product developers and engineers, and purchasers) may not have a sufficiently good understanding of certain manufacturing technologies required to manufacture the product to manually create the manufacturing processes with adequate quality in their computational systems.

[0032] This method and system, or at least some embodiments thereof, utilizes an AI engine to provide semi-automatic generation of bills of materials (BOMs) and process lists for a product. Compared to manually setting up BOMs and / or process lists in a software system, using AI to generate this information can significantly accelerate the process of calculating product costs and sustainability value. In this way, the time from requesting a quote to submitting an actual offer can be significantly reduced. Furthermore, less knowledge is required for the typical user to create product cost and sustainability value calculations. Similarly, product engineers gain access to qualified information about the costs and environmental impacts of all potential product implementations, and can easily obtain the information needed to determine the optimal solution during development.

[0033] The method and system, or at least some embodiments thereof, are capable of cost and value engineering for complex assembly parts that are not tied to a particular manufacturing technology and may assemble a variety of parts using different manufacturing technologies.

[0034] For these more complex assembly parts, predicting prices at a high level is not suitable for price negotiations between customers and suppliers. To negotiate costs and sustainability value, a more granular level of detail is needed, which can be provided by calculating costs at the level of the different BOM items (parts / materials) and BOPs (Breakpoints in Production) required to manufacture the product.

[0035] Furthermore, for use cases where design decisions have already been made during development, simple projected product price figures may not be sufficient. A more granular level of cost and sustainability value can help engineers decide on the specific materials to use for their designs.

[0036] Description of the Implementation Examples

[0037] In embodiments of the method and system, a process inventory generator has been trained to construct a process inventory that maximizes one or more key performance indicators.

[0038] Implementations of the method and system include additional operations such as outputting a planned bill of materials, a process list, and / or at least one key performance indicator.

[0039] In embodiments of the method and system, at least one document is at least one semi-structured document or at least one document described in a free form in natural language.

[0040] In embodiments of the method and system, the natural language processing module includes an encoder for the transformer, a long short-term memory architecture, or a pre-trained language model.

[0041] In embodiments of the method and system, the generated bill of materials, the planned bill of materials, and the process list are directed acyclic graphs.

[0042] In embodiments of the method and system, the process manifest generator is a reinforcement learning agent, particularly a graph convolutional policy network, which iteratively constructs a process manifest by adding nodes and edges that specify the desired tasks and their dependencies.

[0043] In embodiments of the method and system, a reinforcement learning agent has been trained with rewards based on the validity of the generated process list and at least one key performance indicator.

[0044] The computer program product includes instructions that, when executed by a computer, cause the computer to perform the method.

[0045] Supply equipment storage and / or provide computer program products. Attached Figure Description

[0046] The foregoing and other aspects of the invention are best understood from the following detailed description when read in conjunction with the accompanying drawings. The drawings illustrate presently preferred embodiments for illustrating the objectives of the invention. However, the invention is not limited to the specific means disclosed. Embodiments may be combined with each other. Furthermore, embodiments may be combined with any of the features described above. Unless otherwise stated, identical reference numerals between the drawings denote the same features or functionally identical elements. The following figures are included in the drawings:

[0047] Figure 1 The first embodiment is shown.

[0048] Figure 2 Another embodiment is shown.

[0049] Figure 3 An architecture of one embodiment is shown.

[0050] Figure 4 A simplified example of document D is shown, which contains product specifications, the corresponding generated Bill of Materials (BOM-TR), and an excerpt of the corresponding generated Bill of Materials (BOP-DAG) for illustrative purposes only.

[0051] Figure 5 A flowchart illustrating a possible exemplary embodiment is shown. Detailed Implementation

[0052] In the following description, various aspects of the invention and its embodiments will be described. However, those skilled in the art will understand that the embodiments may be practiced using only some or all of them. Specific figures and configurations are set forth for purposes of explanation in order to provide a thorough understanding. However, it will also be apparent to those skilled in the art that the embodiments can be practiced without these specific details.

[0053] The modules described herein may include hardware modules, software modules, or a combination thereof. A software module may consist of software libraries, procedures, subroutines, functions, or any other portion of software code suitable for implementing the module's functionality, depending on the programming paradigm. In some embodiments, specific functionality may be implemented exclusively in hardware (e.g., a processor such as an ASIC or FPGA), while other functionality may be implemented in software.

[0054] Example description

[0055] Figure 1 The illustration shows an exemplary architecture of a computer implementation of an embodiment of the present invention, including:

[0056] (101) Computer system;

[0057] (102) A processor used to execute instructions;

[0058] (103) A memory used to store data and program instructions;

[0059] (104) A computer program (product) comprising instructions for carrying out embodiments of the present invention; and

[0060] (105) User interface, used to present the results of the embodiments of the present invention or to receive user input.

[0061] In an embodiment of the invention, computer program 104 includes program instructions for implementing the method. Computer program 104 is stored in memory 103 of computer system 101. In an embodiment, computer system 101 may use processor 102 to execute the program instructions of computer program 104. The results of executing the program instructions may be presented on user interface 105, stored in memory 103, or saved to another suitable storage medium, such as an external storage device or a cloud-based system.

[0062] Figure 2 The illustration shows another sample structure of a computer implementation of an embodiment of the present invention, including:

[0063] (201) Supply equipment;

[0064] (202) Computer programs (products);

[0065] (203) Computer networks;

[0066] (204) Computing system;

[0067] (205) Mobile devices.

[0068] In an embodiment, the supply device 201 stores a computer program 202 including program instructions for implementing embodiments of the present invention. The supply device 201 provides the computer program 202 via a computer network 203 such as the Internet. For example, a computing system 204 or a mobile device 205 such as a smartphone may load the computer program 202 and execute its program instructions to implement embodiments of the present invention.

[0069] The supply device 201 can act as a network interface, such as a server, cloud-based storage system, or other remote computing infrastructure, enabling the computing system 204 or mobile device 205 to download or access computer programs 202 via computer network 203.

[0070] Alternatively, the supply device 201 may function as a computer-readable storage medium. For example, the supply device 201 may be a physical storage medium for storing the computer program 202, such as an SD card, USB drive, external hard drive, or optical disc. The supply device 201 may be physically connected to the computing system 204 or mobile device 205, such as via a USB interface or SD card slot, enabling these devices to load and execute the computer program 202.

[0071] Preferably, Figures 3 to 5 The embodiments shown can be used as follows Figure 1 or Figure 2 The structure shown is used to implement this.

[0072] Vaswani et al.'s "Attention is all you need" (arXiv:1706.03762v5[cs.CL], available on the internet on March 21, 2023 at https: / / doi.org / 10.48550 / arXiv.1706.03762) discloses an encoder that is part of a transformer architecture, which is, for example, a pre-trained transformer model. The entire contents of that document are incorporated herein by reference.

[0073] Hochreiter et al.'s "Long short-term memory" (Neuralcomputation 9.8(1997):1735-1780) discloses a Long Short-Term Memory (LSTM) architecture. The entire contents of that paper are incorporated herein by reference.

[0074] Zhu et al.'s "A survey on deep graph generation: Methods and applications" (arXivpreprint arXiv:2203.06714(2022)) discloses methods for deep graph generation. The entire contents of that paper are incorporated herein by reference.

[0075] You et al.'s paper, "Graph convolutional policy network for goal-directed molecular graph generation" (Advances in neural information processing systems 31 (2018)), discloses a graph convolutional policy network. The entire contents of that paper are incorporated herein by reference.

[0076] The embodiments described below propose a semi-interactive method for generating a Bill of Materials (BOM) and Bill of Processes (BOP) for a given product based on a (unstructured or semi-structured) text document containing product feature specifications. The embodiments are data-driven (i.e., they can leverage a previously specified set of BOM, BOP, and associated specification documents) and may optionally rely on existing, specially built software for BOM / BOP definition as a means of evaluating the effectiveness (and optional additional KPIs) of the generated BOM and BOP during training.

[0077] In the embodiments described below, the bill of materials for a given product is represented by a tree structure, where each (sub)component of the product corresponds to a node in the tree. Nodes closer to the root node correspond to higher-level components consisting of smaller subcomponents (their child nodes), which may in turn have their own subcomponents at the next (lower) level of the tree. In particular, the root represents the final product, and the raw materials required to produce that final product are given by the leaves.

[0078] In the embodiments described below, the process list specifies the order of manufacturing steps required to produce the product. Similar to the BOM, the BOP can also be represented as a directed acyclic graph (DAG), where each operation is a node in the graph, and the directionality of the edges specifies the order in which the operations are performed. This means that leaf nodes correspond to the initial operations in the manufacturing process, and the root node indicates the final operations required to complete the product.

[0079] Figure 3 The architecture used in the embodiments described below is shown. At a very high level, the first embodiment can be considered as consisting of the following steps:

[0080] The Natural Language Processing (NLP-M) module processes documents containing product feature specifications. NLP-M compresses the document content into a latent representation (LR), which is an embedding in the latent space.

[0081] The Bill of Materials (BOM) generator BOM-G receives the latent representation LR as input and generates a Bill of Materials (BOM)-TR, which is a tree-structured representation of the bill of materials corresponding to the document.

[0082] Since both the Natural Language Processing (NLP) module (NLP-M) and the Bill of Materials (BOM) generator (BOM-G) introduce noise into the process, the User Interface (UI) then allows the User (U) to inspect the generated BOM-TR and correct any errors at hand, resulting in a curated BOM-C. Optionally, the UI employs a recommendation system, using one of the algorithms disclosed in US2023 / 0273573A1, to suggest alternative components to the User (U) that are being considered for replacement. The entire contents of US 2023 / 0273573A1 are incorporated herein by reference.

[0083] The potential representation LR and the planned Bill of Materials (BOM-C) are then passed to the Process Inventory Generator (BOP-G). The BOP-G has been trained to construct process inventories that (a) correspond to the products specified in the document, (b) are valid, and (c) (optionally) maximize a set of KPIs corresponding to ecological or economic objectives (e.g., the price of the product or its CO2 footprint).

[0084] The steps and modules of various embodiments will be described in more detail below.

[0085] An embodiment consists of the following building blocks and aspects:

[0086] The document set SD contains documents D1, D2, ..., D N Product specifications are documents that can be semi-structured or described in free form using natural language. Product specifications can also be requirements or include requirements.

[0087] Natural Language Processing Module NLP-M The NLP-M module can map each document in a document set to a corresponding latent representation (LR), which is the embedding of the corresponding document in a d-dimensional latent space. For example, the NLP-M module can be implemented using an encoder or LSTM with a transformer as disclosed in the references cited above. If the documents in the document set SD follow a known structure, the NLP-M module can include several corresponding preprocessing and feature engineering sub-modules. The NLP-M module can be trained on various tasks, including document completion. Depending on the documents in the document set SD, the NLP-M module can also be a pre-trained language model, which may or may not be further fine-tuned.

[0088] In the first variant, each document in the document set SD specifies the entire product and its sub-components. In this case, the latent representation LR is created from a single document. In the second variant, the document set as a whole specifies the product and its sub-components, potentially including requirements. In this case, the latent representation LR is created from the entire document set SD, which will be described further below.

[0089] Bill of Materials (BOM) Generator Taking the latent representation (LR) of the document as input, a Bill of Materials (BOM-TR) is generated, which is a tree-structured representation of the bill of materials corresponding to the document. This means that... Represents a set of finite trees.

[0090] Each node in the generated Bill of Materials (BOM-TR) represents a specific (sub)component of the product specified in the document. Each node has attributes that describe the (sub)component at the technical level, such as material type, size, etc. The values ​​of these attributes can then be used by downstream blocks of the current embodiment as additional implicit contextual information (e.g., the fact that the plate (component) is made of high carbon steel may make it necessary to use a specific drilling machine in the corresponding process list entry).

[0091] Regarding the implementation of the Bill of Materials (BOM) generator BOM-G, the current embodiment is agnostic when it comes to the choice of its generative model, provided that the generation process is conditioned on the context given by the latent representation LR. Possible choices for the generative model include autoregressive, variational autoencoders, normalized flow, generative adversarial networks (GANs), or diffusion-based models, such as those described in "A survey on deep graph generation: Methods and applications" by Zhu, Yanqiao, et al. (arXiv preprint arXiv:2203.06714(2022)). The entire contents of that document are incorporated herein by reference.

[0092] The user interface (UI) allows users to modify the generated Bill of Materials (BOM-TR) to produce the planned Bill of Materials (BOM-C).

[0093] BOP-G Random Process List Generator Map the potential representation LR of each document together with the planned Bill of Materials (BOM-C) to a process list (BOP-DAG), which is a process list that specifies the corresponding BOM-C (i.e., ...). A directed acyclic graph (DAG) representing a set of finite DAGs.

[0094] Regarding the implementation of the process manifest generator BOP-G, the current embodiment solves the process manifest generation problem via a reinforcement learning agent, such as the Graph Convolutional Policy Network (GCPCN) described in "Graph convolutionalpolicy network forgoal-directed molecular graph generation" (Advances in neural informationprocessing systems 31 (2018)) by You, Jiaxuan et al. The entire contents of that paper are incorporated herein by reference.

[0095] In this setup, the reinforcement learning agent’s task is to iteratively build a process list BOP-DAG by adding nodes and edges that specify the required tasks and their dependencies.

[0096] Furthermore, the current embodiment sets up an artificial environment that assigns reward (RW) to the reinforcement learning agent based on the validity of the generated Bill of Materials (BOP)-DAG and (optionally) KPIs. The current embodiment uses specially constructed software for specifying the BOP to perform soft validity checks (and, when a planned Bill of Materials (BOM-C) is received as an optional input OI, to verify the corresponding BOM / BOP pair against a rule set that imposes restrictions on certain manufacturing steps applied to components made of specific material types) and obtain the values ​​of the associated KPIs.

[0097] During training, the parameters of the reinforcement learning agent are trained to maximize the reward RW. During deployment, the reinforcement learning agent iteratively samples steps from the trained policy to be added to the process list BOP-DAG until a stopping action is determined. This process results in a randomized mapping h.

[0098] In another embodiment, the document set SD contains documents for each specific part of the product. The document set SD as a whole represents the product for which a corresponding bill of materials and process list should be provided. Thus, the exact workflow is as follows:

[0099] First, the documents D1, D2, ..., D contained in the document set SD N They are fed into the Natural Language Processing module NLP-M to produce embeddings f(D1), f(D2), ..., f(D N Subsequently, the current embodiment applies the permutation-invariant pooling operator Θ(f(D1),f(D2),…,f(D)). N ))=:x, this operator gathers information from all the embeddings into Figure 3 The potential representation shown is in LR. Potential candidates for Θ are summation, averaging, or max-pooling operators.

[0100] Second, the current embodiment will potentially represent LR x Feed it to the Bill of Materials (BOM) generator BOM-G to generate the Bill of Materials (BOM-TRg(x) = :T).

[0101] Third, the generated Bill of Materials (BOM-TR) is output via the user interface (UI). When the generated BOM-TR needs correction, the UI detects user interaction and, with the help of specially designed software (e.g., TcPCM, Teamcenter product cost management), generates a planned BOM-C.

[0102] Fourth, the planned Bill of Materials (BOM-C) and potential representation LR x Feeded to the process list generator BOP-G h The process list generator BOP-G h Generate process list BOP-DAG G It is a graphical representation of the process list corresponding to the Bill of Materials (BOM-C) of the plan.

[0103] Figure 4 A simplified example of document D is shown, which contains product specifications, a corresponding generated bill of materials (BOM-TR) for document D, and an excerpt of the corresponding generated process list (BOP-DAG) for illustrative purposes only.

[0104] Figure 5 A flowchart illustrating a possible exemplary embodiment is shown.

[0105] In the first operation (1), the natural language processing module maps at least one document of the specified product into a latent representation, which is an embedding in the latent space.

[0106] In the second operation (2), the bill of materials generator, which receives the potential representation as input, generates the bill of materials for the product.

[0107] In the third operation (3), the user interface outputs the generated bill of materials.

[0108] In the fourth operation (4), the user interface detects user interactions with the generated bill of materials and creates a planned bill of materials based on the user interactions.

[0109] In the fifth operation (5), the process list generator, which receives the potential representation and planned material list as input, constructs the process list of the product.

[0110] Embodiments of this method may be executed by one or more processors, which may include microcontrollers, microprocessors, application-specific integrated circuits (ASICs), or neuromorphic microchips (e.g., neuromorphic processor units). These processors may be part of a computing device, such as a smartphone, tablet, laptop computer, server, or control system in a cloud-based environment.

[0111] The above embodiments of the method can be implemented via a computer program product comprising one or more computer-readable storage media storing instructions executable by one or more processors of a computing system. Execution of the instructions causes the computing system to perform actions corresponding to the operations described above.

[0112] The functions, actions, or tasks shown in the figures or described herein can be performed in response to one or more sets of instructions stored in or on a computer-readable storage medium. The functions, actions, or tasks can be independent of a particular type of instruction set, storage medium, processor, or processing strategy, and can be performed individually or in combination by software, hardware, integrated circuits, firmware, microcode, etc. Similarly, processing strategies can include multiprocessing, multitasking, parallel processing, etc.

[0113] The invention has been described in detail with reference to embodiments and examples. However, variations and modifications can be made within the spirit and scope of the invention as covered by the claims. The phrase "A, B, and / or C" as an alternative expression may specify the use of one or more of A, B, and C.

[0114] For clarity, it should be understood that the use of “a” or “an” in this application includes the possibility of multiple possibilities, and the term “comprising” does not exclude additional steps or elements.

[0115] In this patent application, nouns and pronouns referring to people generally do not specify a particular gender.

Claims

1. A computer-implemented method for generating a bill of materials and a corresponding bill of processes for a product, wherein the following operations are performed by modules, and wherein the modules are hardware modules and / or software modules executed by one or more processors: - mapping (1), by a natural language processing module (NLP-M), of at least one document specifying a product into a latent representation (LR), which is an embedding in a latent space, - generating (2), by a bill of materials generator (BOM-G) receiving the latent representation (LR) as input, a bill of materials (BOM-TR) for the product, - outputting (3), by a user interface (UI), the generated bill of materials (BOM-TR), - detecting (4), by the user interface (UI), a user interaction with the generated bill of materials (BOM-TR) and creating a curated bill of materials (BOM-C) depending on the user interaction, and - constructing (5), by a bill of processes generator (BOP-G) receiving the latent representation (LR) and the curated bill of materials (BOM-C) as input, a bill of processes (BOP-DAG) for the product.

2. The method according to claim 1, - wherein the bill of processes generator (BOP-G) has been trained to construct a bill of processes (BOP-DAG) that maximizes one or more key performance indicators.

3. The method according to claim 2, - with additional operations outputting the curated bill of materials (BOM-C), the bill of processes (BOP-DAG) and / or the one or more key performance indicators.

4. The method according to any of the preceding claims, - wherein the at least one document is at least one semi-structured document or at least one free-form description in a natural language.

5. The method according to any of the preceding claims, - wherein the natural language processing module (NLP-M) comprises an encoder of a transformer, a long short-term memory architecture or a pre-trained language model.

6. The method according to any of the preceding claims, - wherein the generated bill of materials (BOM-TR), the curated bill of materials (BOM-C) and the bill of processes (BOP-DAG) are directed acyclic graphs.

7. The method according to claim 6, - wherein the bill of processes generator (BOP-G) is a reinforcement learning agent, in particular a graph convolutional policy network, which iteratively constructs the bill of processes (BOP-DAG) by adding nodes and edges specifying required tasks and their dependencies.

8. The method according to claim 7, - wherein, - wherein the reinforcement learning agent has been trained with a reward (RW) based on a validity of the generated bill of processes (BOP-DAG) and the one or more key performance indicators.

9. A system for generating a bill of materials and a corresponding bill of processes for a product, with the following modules: - a natural language processing module (NLP-M) configured for mapping (1) at least one document specifying a product into a latent representation (LR), which is an embedding in a latent space, - a bill of materials generator (BOM-G) configured for receiving the latent representation (LR) as input and generating (2) a bill of materials (BOM-TR) for the product, - a user interface (UI) configured for outputting (3) the generated bill of materials (BOM-TR), detecting (4) a user interaction with the generated bill of materials (BOM-TR), and creating a curated bill of materials (BOM-C) depending on the user interaction, and - a process order generator (BOP-G) configured for receiving the latent representation (LR) and the curated bill of materials (BOM-C) as input and constructing (5) a process order (BOP-DAG) for the product.

10. A computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method according to one of the method claims.

11. A supply device for a computer program product according to the preceding claim, wherein the supply device stores and / or provides the computer program product.

Images