Method and system for generating a bill of materials and a corresponding work plan list for a product

Abstract

To generate a bill of materials (BOM) and a corresponding routing list for a product, a natural language processing module (NLP-M) maps at least one product-specifying document into a latent representation (LR), which is embedded in the latent space. A BOM generator (BOM-G) receives the latent representation (LR) as input and generates a BOM (BOM-TR) for the product. A user interface (UI) captures user interactions with the generated BOM (BOM-TR) and creates a curated BOM (BOM-C) based on these interactions. A routing list generator (BOP-G) receives the latent representation (LR) and the curated BOM (BOM-C) as input and constructs a routing list (BOP-DAG) for the product.The method and the system, or at least some of their embodiments, provide for the semi-automatic generation of the bill of materials (BOM) and the work plan for a product using an AI engine. Compared to manually creating a BOM and / or a work plan within a software system, using AI to generate this information can significantly accelerate the process of calculating a product's costs and sustainability values. This can substantially reduce the time from request for quotation to the submission of the actual quotation.

Description

field of technology

[0001] This registration relates to the generation of a bill of materials and a corresponding work plan list for a product. State of the art

[0002] In industry, transparency regarding the costs and sustainability values ​​(such as carbon footprint) of products is crucial. Typically, this transparency is achieved by defining a bill of materials (BOM) in combination with a bill of processes (BOP). Costs and sustainability values ​​are then calculated and aggregated along the BOM / BOP structure of a product to obtain its total cost and carbon footprint. In this sense, the BOM / BOP can be considered the digital twin of the product, encompassing the materials and manufacturing processes required to produce it, including all direct and indirect overhead costs.

[0003] The need to calculate costs and sustainability values ​​begins as early as the design and development phase of a product. When designing a new product, engineers may encounter various potential ways to realize it, which materials can be used, and which manufacturing processes might be required. By obtaining qualified information about the expected costs of different potential realizations, as well as values ​​regarding how these specific realizations impact sustainability, engineers are empowered to make design decisions at an early stage of product development. In this way, products can be optimized for cost and their environmental impact long before they go into production.

[0004] Another application of product cost calculation is the estimation of product costs and sustainability values ​​in purchasing departments. In this area, a purchasing user wants to quickly obtain qualified information about the costs and sustainability values ​​of purchased parts and use this information in price negotiations with potential suppliers.

[0005] Often, when a purchasing user is interested in estimating the cost of a product, they may not even have an existing bill of materials to begin calculating costs and sustainability values. Therefore, the purchasing user must create the bill of materials from scratch within the system, working from the bottom up, so they can use this calculation in price negotiations with suppliers.

[0006] For example, if the product is assembled with 200 different parts, the purchasing user must create all 200 parts individually in their costing software to manually create a digital twin of the product.

[0007] In addition to the bill of materials, the work plan must also be set up manually. Every single manufacturing step required to produce the product must be defined individually. This requires detailed knowledge of the various manufacturing techniques used in the production of this specific product.

[0008] On the other side of the purchase price negotiations is typically a supplier who must provide their customers with a detailed cost breakdown when responding to requests for proposals. Furthermore, the supplier must also ensure that specific performance indicators regarding profitability and the risks associated with a project are met. This provides management with reliable data on which to base their decision whether or not to acquire a project.

[0009] For internal project controlling, the bill of materials is often created by the R&D department during a design and development phase of a product using software tools such as Teamcenter NX.

[0010] However, there are still use cases in early phases of the product life cycle where a detailed bill of materials may not yet be available.

[0011] Even in this use case, the work plan list must be set up manually, defining each individual manufacturing step. Setting up the manufacturing process must be carried out by manufacturing experts and requires in-depth knowledge of manufacturing techniques.

[0012] The bill of materials and the work plan are typically set up in specially developed software for calculating costs and sustainability values. Several costing tools are available on the market, ranging from the widely used Microsoft Excel to more sophisticated tools such as Teamcenter Product Cost Management (TcPCM), Facton, aPriori, and others.

[0013] Typically, the bill of materials (BOM) and work plan list must be set up manually in each of these tools. In some cases, the BOM can be imported from third-party systems such as Teamcenter NX or other 3D-based design software.

[0014] There are also solutions on the market that use AI models to predict the price of a product using specific manufacturing technologies such as milling, drilling, and turning. These solutions typically require a 3D model in the form of a CAD or JT file and are specifically designed for certain manufacturing technologies. Summary of the invention

[0015] The object of the present invention is to identify a problem in the prior art and to find a technical solution for it.

[0016] The invention is defined in the independent claims. Further advantageous arrangements and embodiments of the invention are set out in the respective dependent claims.

[0017] According to the procedure for generating a bill of materials and a corresponding work plan list for a product, the following operations are performed by modules, where the modules are hardware modules and / or software modules executed by one or more processors: - Mapping, by means of a natural language processing module, at least one document specifying a product into a latent representation that is an embedding in latent space, - Generating a bill of materials for the product using a bill of materials generator that receives the latent representation as input, - Output, through a user interface, the generated bill of materials, - Capturing, through the user interface, user interactions with the generated bill of materials and creating a curated bill of materials based on these user interactions, and - Construct, by means of a work plan list generator that receives the latent representation and the curated bill of materials as input, a work plan list for the product.

[0018] The system for generating a bill of materials and a corresponding work plan list for a product comprises the following modules, where the modules are hardware modules and / or software modules executed by one or more processors: - a natural language processing module configured to map at least one product-specifying document into a latent representation that is an embedding in latent space, - a bill of materials generator configured to receive the latent representation as input and to generate a bill of materials for the product, - a user interface configured to output the generated bill of materials, to capture user interactions with the generated bill of materials, and to create a curated bill of materials based on user interactions, and - a work plan list generator configured to receive the latent representation and the curated bill of materials as input and to construct a work plan list for the product.

[0019] The following advantages and explanations are not necessarily the result of abandoning the independent claims. Rather, they may be advantages and explanations that apply only to certain embodiments or variants.

[0020] The term "computer" is to be interpreted broadly and encompasses any electronic device with data processing capabilities. Examples include, but are not limited to, personal computers, servers, clients, programmable logic controllers (PLCs), handheld devices, pocket PCs, mobile devices such as smartphones, or other communication devices capable of processing data. A computer may comprise one or more processors and memory units, such as volatile memory (e.g., RAM) or non-volatile memory (e.g., solid-state drives, hard disk drives, flash memory).

[0021] In the context of embodiments of the invention, the terms "memory", "memory unit" 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 discs).

[0022] While the bill of materials (BOM) is typically created during the design process, the work plan (WP) is not available at this stage of product development. However, both the BOM and the WP are needed to calculate and estimate the product's costs and sustainability values.

[0023] The process and the system, or at least some of their embodiments, can provide an engineer with qualified information about costs and sustainability values ​​by extending the bill of materials to include the manufacturing processes required for the production and assembly of the product. This is particularly helpful when a product engineer, during product development, lacks sufficient knowledge of the details of the required manufacturing processes to define them manually in a way that yields qualified results. Manually creating the bill of materials and / or the work plan within a system is also very time-consuming and cumbersome, as all the information needed for a meaningful cost calculation must be entered manually. Furthermore, in-depth knowledge of all the manufacturing processes required for the production and assembly of the product is necessary.Different user groups, such as product developers and engineers, as well as purchasers, may not have sufficient knowledge of certain manufacturing technologies required to produce the product in order to manually create the manufacturing processes in their costing system with sufficient quality.

[0024] The method and system, or at least some of their embodiments, provide for the semi-automatic generation of the bill of materials (BOM) and the work plan for a product using an AI engine. Compared to manually creating a BOM and / or a work plan within a software system, using AI to generate this information can significantly accelerate the process of calculating a product's costs and sustainability values. This can considerably reduce the time from request for quotation to the submission of the actual quotation. Furthermore, the typical user will require less expertise to create a product cost and sustainability value calculation.Similarly, product engineers can obtain qualified information about costs and environmental impacts for all potential product realizations and easily get the information they need to choose the optimal solution during the development phase.

[0025] The method and system, or at least some of their embodiments, are capable of addressing the cost and value engineering of complex assembled parts that are not tied to a specific manufacturing technology and can assemble various parts from different manufacturing technologies.

[0026] For these more complex assembled parts, high-level price prediction is not suitable for price negotiations between customer and supplier. Negotiations regarding costs and sustainability values ​​require a more granular level of detail, which can be provided by calculating costs at the level of the various BOM (parts / materials) and BOP (manufacturing steps) items required to manufacture the product.

[0027] Even for the use case of making design decisions during development, a simple figure for the expected product price might not suffice. Having cost and sustainability values ​​at a more granular level can help the engineer decide on the use of specific materials for their design. Description of the embodiments

[0028] In one embodiment of the method and the system, the work plan list generator was trained to construct the work plan list in such a way as to maximize one or more performance indicators.

[0029] One embodiment of the method and the system includes the additional operation of outputting the curated bill of materials, the work plan list and / or the at least one performance indicator.

[0030] In one embodiment of the method and the system, the at least one document is at least a semi-structured document or at least a free-form description in natural language.

[0031] In one embodiment of the method and the system, wherein the natural language processing module includes the encoder of a transformer, a long-short-term memory architecture, or a pre-trained language model.

[0032] In one embodiment of the method and the system, the generated bill of materials, the curated bill of materials, and the work plan list are directed acyclic graphs.

[0033] In one embodiment of the method and system, the work plan list generator is a reinforcement learning agent, in particular a graph convolutional policy network, which iteratively constructs the work plan list by adding nodes and edges that specify required tasks and their dependencies.

[0034] In one embodiment of the method and system, the reinforcement learning agent was trained with a reward based on the validity of the generated work plan list and at least one performance indicator.

[0035] The computer program product, which includes instructions that, when the program is executed by a computer, cause the computer to perform the procedure.

[0036] The deployment device stores and / or deploys the computer program product. Brief description of the drawings

[0037] The foregoing and other aspects of the present invention are best understood from the following detailed description when read in conjunction with the accompanying drawings. For the purpose of illustrating the invention, the drawings show embodiments that are currently preferred. However, the invention is not limited to the specifically disclosed instrumentation. The embodiments can be combined with one another. Furthermore, the embodiments can be combined with any of the features described above. Unless otherwise indicated, identical reference numerals denote the same features or functionally identical elements between the drawings. The drawings include the following figures: Fig.Figure 1 shows a first embodiment, Fig. 2 shows another embodiment, Fig. Figure 3 shows an architecture of one embodiment, Fig. Figure 4 shows a simple example of a document D containing product specifications, a generated bill of materials BOM-TR corresponding to document D, and, for illustration purposes only, an excerpt of a corresponding generated work plan list BOP-DAG. Fig. Figure 5 shows a flowchart of a possible exemplary embodiment.

[0038] The following description details various aspects of the present invention and its embodiments. However, it will be apparent to those skilled in the art that embodiments can be implemented with only some or all of these aspects. For illustrative purposes, specific figures and configurations are presented to provide a comprehensive understanding. However, it will also be apparent to those skilled in the art that the embodiments can be implemented without these specific details.

[0039] The modules described herein may comprise hardware modules, software modules, or a combination thereof. A software module may consist of a software library, a procedure, a subroutine, a function, or, depending on the programming paradigm, any other part of the software code suitable for implementing the module's functionality. In certain embodiments, specific functionalities may be implemented exclusively by hardware (e.g., a processor such as an ASIC or FPGA), while other functionalities may be implemented by software. Description of the examples

[0040] Fig. Figure 1 illustrates an exemplary architecture for the computer-implemented execution of embodiments of the invention, comprising: (101) A computer system; (102) a processor for executing instructions; (103) a memory for storing data and program instructions; (104) a computer program (product) containing instructions for carrying out embodiments of the invention; and (105) a user interface for displaying the results of embodiments of the invention or for receiving user input.

[0041] In one embodiment of the invention, the computer program 104 includes program instructions for implementing embodiments of the method. The computer program 104 is stored in the memory 103 of the computer system 101. In embodiments, the computer system 101 can execute the program instructions of the computer program 104 using the processor 102. Results generated by the execution of the program instructions can be displayed on the user interface 105, stored in the memory 103, or saved on another suitable storage medium, such as external storage devices or cloud-based systems.

[0042] Fig. Figure 2 illustrates a further example structure for the computer-implemented execution of embodiments of the invention, comprising: (201) A staging device; (202) a computer program (product); (203) a computer network; (204) a computer system; (205) a mobile device.

[0043] In one embodiment, the provisioning device 201 stores a computer program 202 containing program instructions for executing embodiments of the invention. The provisioning device 201 makes the computer program 202 available via a computer network 203, such as the Internet. For example, a computing system 204 or a mobile device 205, such as a smartphone, can load the computer program 202 and execute its program instructions to implement embodiments of the invention.

[0044] The provisioning device 201 can serve as a network interface, such as a server, a cloud-based storage system, or other remote computing infrastructure, which enables the computing system 204 or the mobile device 205 to download or access the computer program 202 over the computer network 203.

[0045] Alternatively, the deployment device 201 can serve as a computer-readable storage medium. For example, the deployment device 201 can be a physical storage medium, such as an SD card, a USB drive, an external hard drive, or an optical disc, that stores the computer program 202. The deployment device 201 can be physically connected to the computing system 204 or the mobile device 205, for example, via a USB interface or an SD card slot, enabling these devices to load and execute the computer program 202.

[0046] Preferably, the ones in the Fig. 3 to 5 embodiments shown with a structure as in Fig. 1 or Fig. 2 will be implemented.

[0047] Vaswani et al.: “Attention is all you need”, arXiv:1706.03762v5 [cs.CL], available online at https: / / doi.org / 10.48550 / arXiv.1706.03762 on March 21, 2023, discloses an encoder that is part of a Transformer architecture, the latter being, for example, a pretrained Transformer model. The entire contents of this document are hereby incorporated by reference.

[0048] Hochreiter et al.: “Long short-term memory.”, Neural computation 9.8 (1997): 1735–1780, reveals a Long-Short-Term Memory (LSTM) architecture. The entire content of this document is hereby incorporated by reference.

[0049] Zhu et al.: “A survey on deep graph generation: Methods and applications.” arXiv preprint arXiv:2203.06714 (2022), discloses methods for deep graph generation. The entire content of this document is hereby incorporated by reference.

[0050] You et al., “Graph convolutional policy network for goal-directed molecular graph generation.” Advances in neural information processing systems 31 (2018), reveals a graph-convolutional policy network. The entire content of this document is hereby incorporated by reference.

[0051] The embodiments described below propose a semi-interactive approach to generating a bill of materials (BOM) and a work plan list for a given product, based on a (unstructured or semi-structured) text document containing the product's feature specifications. The embodiments are data-driven (i.e., they can utilize a collection of previously specified BOMs, BOPs, and their associated specification documents) and can optionally leverage existing, purpose-built BOM / BOP definition software as a means of evaluating the validity (and optionally additional KPIs) of the generated BOM and work plan list during training.

[0052] In the embodiments described below, a bill of materials for a given product is represented by a tree-like structure, with each (sub)component of the product corresponding to a node in this tree. Nodes closer to the root node correspond to higher-level components, which consist of smaller subcomponents (their child nodes), which in turn may 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 manufacture this final product are represented by the leaves.

[0053] In the embodiments described below, a work plan list specifies the sequence of manufacturing steps required to produce a product. Similar to BOMs, BOPs can also be represented as a directed acyclic graph (DAG), where each operation is a node in the graph and the edge direction indicates the order in which the operations are performed. This means that the leaf nodes correspond to the initial operations in the manufacturing process, and the root node indicates the final operation required to complete the product.

[0054] Fig. Figure 3 shows an architecture used by the embodiments described below. At a very high level, a first embodiment can be considered to consist of the following steps: A natural language processing module (NLP-M) processes a document containing feature specifications of a product. The NLP-M condenses the document's content into a latent representation (LR), which is an embedding in latent space.

[0055] A bill of materials generator (BOM-G) receives the latent representation LR as input and generates a bill of materials BOM-TR, which is a tree structure representation of a bill of materials corresponding to the document.

[0056] Since both the natural language processing module (NLP-M) and the bill of materials generator (BOM-G) introduce noise into the process, a user interface (UI) allows a user (U) to review the generated bill of materials (BOM-TR) and correct any errors, resulting in a curated bill of materials (BOM-C). Optionally, the UI uses a recommendation system to suggest alternative components to the one the user (U) is considering replacing, using one of the algorithms disclosed in US 2023 / 0273573 A1. The entire contents of US 2023 / 0273573 A1 are hereby incorporated by reference.

[0057] The latent representation LR and the curated bill of materials BOM-C are then passed to a bill of processes generator BOP-G. The bill of processes generator (BOP-G) has been trained to construct a bill of processes that (a) corresponds to the product specified in the document, (b) is valid, and (c) (optionally) maximizes a set of KPIs that correspond to environmental or economic goals (e.g., the price of the product or its carbon footprint).

[0058] The following section describes in more detail the respective steps and modules with regard to different embodiments.

[0059] One embodiment consists of the following building blocks and aspects: A set of documents D1, D2, ..., D NSD contains product specifications as documents, which can be semi-structured documents or free-form descriptions in natural language. The product specifications can also be requirements or contain requirements.

[0060] A natural language processing module NLP-M, f:D → ℝ dThe module is capable of mapping each document contained in the document set to a corresponding latent representation (LR), which is an embedding of the respective document in d-dimensional latent space. For example, the natural language processing module (NLP-M) can be implemented using the encoder of a transformer or an LSTM, as revealed 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 submodules. The NLP-M module can be trained for a variety of tasks, including document completion.Depending on the documents in the SD document set, the natural language processing module NLP-M may also be a pre-trained language model that can be further fine-tuned or not.

[0061] In the first variant, each document in the document set SD specifies an entire product and its subcomponents. In this case, the latent representation LR is created from a single document. In the second variant, the document set as a whole specifies a product and its subcomponents, possibly including requirements. In this case, the latent representation LR is created from the entire document set SD, as described below.

[0062] A bill of materials (BOM-G) generator, g:ℝd→J, It takes the latent representation LR of a document as input and generates a bill of materials BOM-TR, which is a tree structure representation of a bill of materials corresponding to the document. That is, T denotes the set of finite trees.

[0063] Each node in the generated BOM-TR represents a specific (sub)component of the product specified in the document. Each node has attributes that describe the (sub)component at a technical level, such as material type, dimensions, 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 a plate (sub)component is made of high-carbon steel may require the use of a specific drilling machine in a corresponding work plan entry).

[0064] Regarding the implementation of the BOM-G bill of materials generator, the current implementation is agnostic with respect to the choice of its generative model, as long as it is possible to condition the generation process to the context given by the latent representation LR. Possible choices for the generative model include autoregressive, variational autoencoder, normalizing flows, generative adversarial network (GAN), or diffusion-based models, as described in Zhu, Yanqiao, et al., “A survey on deep graph generation: Methods and applications.”, arXiv preprint arXiv:2203.06714 (2022). The entire contents of this document are hereby incorporated by reference.

[0065] A user interface (UI) allows a user (U) to modify the generated bill of materials (BOM-TR) to create a curated bill of materials (BOM-C).

[0066] A stochastic work plan list generator BOP-G, h:ℝd×J→G, maps the latent representation LR of each document together with the curated bill of materials BOM-C to a work plan list BOP-DAG, which is a directed acyclic graph (DAG) that specifies a work plan list corresponding to the curated bill of materials BOM-C (i.e., G denotes the set of finite DAGs).

[0067] Regarding the implementation of the work plan list generator BOP-G, the current embodiment solves the problem of work plan list generation using a reinforcement learning agent, such as a graph convolutional policy network (GCCPN), as described in You, Jiaxuan, et al., "Graph convolutional policy network for goal-directed molecular graph generation," Advances in neural information processing systems 31 (2018). The entire content of that document is hereby incorporated by reference.

[0068] In this context, the task of the reinforcement learning agent is to iteratively construct the work plan list BOP-DAG by adding nodes and edges that specify the required tasks and their dependencies.

[0069] Furthermore, the current embodiment establishes an artificial environment that assigns a reward (RW) to the reinforcement learning agent based on the validity of the generated work plan list BOP-DAG and (optional) KPIs. The current embodiment uses specially developed software to specify BOPs in order to perform a soft validity check (and, upon receiving the curated bill of materials BOM-C as an optional input OI, to verify the respective BOM / BOP pair against a set of rules that impose restrictions on the application of certain manufacturing steps to components made of specific material types) and to obtain the values ​​of the associated KPIs.

[0070] 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 work plan list BOP-DAG until a stop operation is selected. This process induces the random mapping h.

[0071] In another embodiment, the set of documents (SD) contains documents that each specify only a part of a product. The SD set as a whole represents the product for which a corresponding bill of materials and routing are to be provided. The exact workflow is then as follows: First, the documents in the set of documents D1, D2, ..., D NThe documents contained in SD were fed into the natural language processing module NLP-M to generate the embeddings f(D1),f(D2),...,f(D N ). Subsequently, the current embodiment applies a permutation-invariant pooling operator Θ(f(D1),f(D2),...,f(D) N ))=:x, which retrieves the information from all embeddings in the in Fig. The latent representation LR shown in the 3 is aggregated. Possible candidates for Θ are summation, averaging, or the max-pooling operator.

[0072] Secondly, the current embodiment feeds the latent representation LR, x, into the bill of materials generator BOM-G, g, in order to generate the bill of materials BOM-TR, g(x)=:T.

[0073] Third, the generated bill of materials (BOM)-TR, T, is output by a user interface (UI). If the generated BOM-TR needs to be revised, the user interface (UI) captures user interactions of a user (U) that generate a curated BOM-C, T̃, using specially developed software (e.g., TcPCM, Teamcenter Product Cost Management).

[0074] Fourth, the curated bill of materials BOM-C, T̃, and the latent representation LR, x, are fed into the routing list generator BOP-G, h, which generates a routing list BOP-DAG, G, which is a graph representation of the routing list corresponding to the curated bill of materials BOM-C.

[0075] Fig. Figure 4 shows a simple example of a document D containing product specifications, a generated bill of materials BOM-TR corresponding to document D, and, for illustration purposes only, an extract of a corresponding generated work plan list BOP-DAG.

[0076] Fig. Figure 5 shows a flowchart of a possible exemplary embodiment.

[0077] In a first operation (1), a natural language processing module maps at least one document specifying a product into a latent representation, which is an embedding in latent space.

[0078] In a second operation (2), a bill of materials generator, which receives the latent representation as input, generates a bill of materials for the product.

[0079] In a third operation (3), a user interface outputs the generated bill of materials.

[0080] In a fourth operation (4), the user interface captures user interactions with the generated bill of materials and creates a curated bill of materials depending on the user interactions.

[0081] In a fifth operation (5), a work plan list generator, which receives the latent representation and the curated bill of materials as input, constructs a work plan list for the product.

[0082] Embodiments of the method can be executed by one or more processors, which may include a microcontroller, a microprocessor, an application-specific integrated circuit (ASIC), or a neuromorphic microchip (e.g., a neuromorphic processing unit). These processors may be part of a computing device such as a smartphone, tablet, laptop, server, or control system in a cloud-based environment.

[0083] The embodiments of the method described above can be implemented via a computer program product comprising one or more computer-readable storage media on which instructions are stored that can be executed by one or more processors of a computing system. The execution of the instructions causes the computing system to perform operations corresponding to the operations of the method described above.

[0084] The functions, operations, or tasks depicted in or described herein can be performed in response to one or more sets of instructions stored in or on computer-readable storage media. The functions, operations, or tasks can be independent of the specific type of instruction set, storage medium, processor, or processing strategy, and can be performed by software, hardware, integrated circuits, firmware, microcode, and the like, alone or in combination. Likewise, processing strategies can include multiple processing, multitasking, parallel processing, and the like.

[0085] The invention has been described in detail with reference to embodiments and examples thereof. However, variations and modifications may be made within the spirit and scope of protection of the invention covered by the claims. The phrase "A, B and / or C" as an alternative expression may provide that one or more of A, B and C may be used.

[0086] For the sake of clarity, it should be noted that the use of “a” or “an” in this application includes the possibility of plurality and the term “comprehensive” does not exclude any additional steps or elements.

[0087] Regardless of the grammatical usage of the term, persons with male, female or other gender identities are included within the term. QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] US 2023 / 0273573

[0056] Cited non-patent literature

[0000] Vaswani et al.: “Attention is all you need”, arXiv:1706.03762v5 \[cs.CL], available on the Internet at https: / / doi.org / 10.48550 / arXiv.1706.03762 on 21 March 2023

[0047] Hochreiter et al.: “Long short-term memory.”, Neural computation 9.8 (1997): 1735-1780

[0048] Zhu et al.: “A survey on deep graph generation: Methods and applications.” arXiv preprint arXiv:2203.06714 (2022

[0049] You et al. „Graph convolutional policy network for goal-directed molecular graph generation.“ Advances in neural information processing systems 31 (2018

[0050] Zhu, Yanqiao, et al., „A survey on deep graph generation: Methods and applications.“, arXiv preprint arXiv:2203.06714 (2022

[0064] You, Jiaxuan, et al. „Graph convolutional policy network for goal-directed molecular graph generation“ beschrieben. Advances in neural information processing systems 31 (2018

[0067]

Claims

[1] Computer-implemented method for generating a bill of materials and a corresponding work plan list 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) at least one document specifying a product into a latent representation (LR) that is an embedding in latent space, - Generating (2) a bill of materials (BOM-TR) for the product using a bill of materials generator (BOM-G) that receives the latent representation (LR) as input, - Output (3) through a user interface (UI) of the generated bill of materials (BOM-TR), - Capture (4) user interactions with the generated bill of materials (BOM-TR) through the user interface (UI) and create a curated bill of materials (BOM-C) based on the user interactions, and - Construct (5) a bill of processes (BOP-DAG) for the product using a Bill of Processes Generator (BOP-G) that receives the latent representation (LR) and the curated bill of materials (BOM-C) as input. [2] Method according to claim 1, - wherein the bill of processes generator (BOP-G) was trained to construct the bill of processes (BOP-DAG) in such a way as to maximize one or more key performance indicators. [3] Method according to claim 2, - with the additional operation of outputting the curated bill of materials (BOM-C), the work plan list (BOP-DAG) and / or at least one performance indicator. [4] Method according to any one of the preceding claims, - where at least one document is at least a semi-structured document or at least a free-form description in natural language. [5] Method according to any one of the preceding claims, - where the natural language processing module (NLP-M) contains the encoder of a transformer, a long-short-term memory architecture, or a pre-trained language model. [6] Method according to any one of the preceding claims, - where the generated bill of materials (BOM-TR), the curated bill of materials (BOM-C) and the work plan list (BOP-DAG) are directed acyclic graphs. [7] Method according to claim 6, - where the work plan list generator (BOP-G) is a reinforcement learning agent, in particular a graph convolutional policy network, which iteratively constructs the work plan list (BOP-DAG) by adding nodes and edges that specify required tasks and their dependencies. [8] Method according to claim 7, - where the reinforcement learning agent was trained with a reward (RW) based on the validity of the generated work plan list (BOP-DAG) and at least one performance indicator. [9] System for generating a bill of materials and a corresponding work plan list for a product, comprising the following modules: - a natural language processing module (NLP-M) configured to map (1) at least one product-specifying document into a latent representation (LR) that is an embedding in latent space, - a bill of materials generator (BOM-G) configured to receive the latent representation (LR) as input and to generate (2) a bill of materials (BOM-TR) for the product, - a user interface (UI) configured to output (3) the generated bill of materials (BOM-TR), to capture (4) user interactions with the generated bill of materials (BOM-TR), and to create a curated bill of materials (BOM-C) depending on the user interactions, and - a work plan list generator (BOP-G) configured to receive the latent representation (LR) and the curated bill of materials (BOM-C) as input and to construct (5) a work plan list (BOP-DAG) for the product. [10] Computer program product comprising instructions which, when the program is executed by a computer, cause the computer to execute a method according to one of the method claims. [11] Provisioning device for the computer program product according to the preceding claim, wherein the provisioning device stores and / or provides the computer program product.

Images