Method and device for automatically configuring manufacturing execution system

Abstract

A method for automatically configuring a software system. In this case, the configuration request is processed by a plurality of software agents, each agent having a large language model (LLM). Each agent contributes to completion requests based on its role and professional knowledge (domain experts, application experts, configuration experts). The result is a new or adapted configuration of the software system.

Description

Technical Field

[0001] The present invention relates to a method for automating the configuration of a manufacturing execution system (MES) using artificial intelligence (AI), as well as an apparatus, a computer program, and a machine-readable storage medium. Background Technology

[0002] Manufacturing Execution System (MES) is a computer-aided system used in manufacturing to typically control the transformation from raw materials to finished products. The German terms "Produktionsleitsystem" (production guidance system) or "Fertigungsmanagementsystem" (manufacturing management system) are often used synonymously. Compared to similarly effective systems for production planning, such as ERP (Enterprise Resource Planning), MES is characterized by its direct connection to a distributed system of process automation, enabling real-time guidance, management, control, or monitoring of production through higher-level control of machines. This includes classic data acquisition and preparation, such as Production Data Acquisition (BDE) and Machine Data Acquisition (MDE), as well as all other processes that have a timely impact on the manufacturing / production process. In particular, MES is used for the continuous controlled execution of existing and effective plans and for providing feedback from the process.

[0003] DE 10 2007 045 926 A1 discloses the interface between manufacturing management systems and automation systems.

[0004] Currently, MES is configured or adapted manually. This task is typically performed by highly qualified application engineers who need to undergo very complex qualification processes and spend weeks or months in the factory to apply the corresponding products. The greatest cost arises primarily from the lengthy analysis of the production line and the stations used within it, as well as the collection and understanding of manufacturer-specific documentation for machine interfaces and production processes. The resulting understanding is then manually translated into MES system-specific configurations. Here, the configurations of various MES components are highly heterogeneous in the market and therefore complex. Thus, the disadvantage in this case is that, due to the high complexity and heterogeneity of the configurations, manually configuring and adapting MES systems requires a significant amount of time and manpower, making it virtually unscalable. Summary of the Invention

[0005] Therefore, the objective of this invention is to improve the scalability of configuring and adapting MES products.

[0006] Advantages of the present invention The main advantages of this invention include: reduced time and manpower spent configuring the MES system. Furthermore, the invention improves the scalability of the MES solution through automation. Another advantage is reduced error-proneness through automated configuration. Similarly, this invention enhances user-friendliness through an intuitive graphical user interface and the results output via that interface.

[0007] The disclosure of this invention In a first aspect, the present invention relates to a method for automating the configuration of a software system. The software system may be a Manufacturing Execution System (MES) for manufacturing control. The software system or MES can be configured in multiple ways to adapt it to the specific needs of a manufacturing enterprise and its production processes. Configuration may include technical and organizational aspects. It is conceivable to configure a production model for the MES. The production model may define the production structure, production resources (machines, workstations, personnel), and production processes (production lines, operations, material flows). One or more data structures may be configured. The data structures, for example, specify how production data (e.g., number of pieces, production time, machine data, quality data) is formatted and how it is processed when necessary. Furthermore, connections to other IT systems, such as ERP systems, PLM systems, or quality management systems, can be configured. That is, software interfaces with other software systems or machines can be configured. The user interface of the MES can be configured, particularly by adapting the user interface to the needs of different user groups and the data and / or data formats to be output. Furthermore, notification and alarm management of the MES can be configured, particularly by defining notification and alarm rules for critical events occurring in production according to different process steps. You can also consider configuring MES reporting and / or analysis, especially by configuring reports and assessments for monitoring and analyzing production.

[0008] The method of the first aspect of the invention may begin with an optional step. In this case, a global configuration model of the software system, defining configuration parameters and their dependencies, is provided. Subsequently, a configuration request is received, which characterizes or describes a desired change or creation of the configuration of the software system. The configuration request is preferably a textual description of the configuration request, i.e., text written in a human-understandable language.

[0009] The configuration request is then processed by multiple cooperating software agents, each using a large language model (LLM). Each agent contributes to fulfilling the configuration request based on its role and expertise. Roles and expertise are preferably represented by textual descriptions. These textual descriptions are then used to configure the LLM for the corresponding agent in terms of its role and expertise.

[0010] It should be noted that alternative text descriptions can be provided via voice input.

[0011] These intelligent agents have the following roles: - At least one domain expert who possesses expertise in the application domain of the software system and is able to interpret configuration requests within the context of a specific request.

[0012] - At least one application expert who possesses detailed knowledge of the software system and its configuration possibilities, and who can translate configuration requests into specific configuration parameters.

[0013] - At least one configuration expert, who is specifically responsible for creating configurations according to a specified format of the global configuration model, optionally verifying the configurations, and optionally logging the configurations.

[0014] After the agent processes the configuration request, a new or adapted software system configuration is generated through agent interaction. Finally, this configuration can be used to set up and, if necessary, run the software system.

[0015] It is proposed that agents can interact in the following ways: exchanging information and knowledge about configuration requests, application domains, and software systems, especially via a shared communication space or communication channel of the agents; discussing and negotiating configuration options based on the expertise of each agent, especially in the aforementioned communication space; and optionally, validating the proposed configuration parameters according to the rules and constraints of a global configuration model.

[0016] Furthermore, it is proposed to specialize the agent by enabling it to access a knowledge base containing relevant documents, guidelines, archives, and historical configuration data of the software system. Preferably, the agent uses Retrieval-Augmented Generation (RAG) technology to efficiently search for configuration-related information in the knowledge base and anchor the LLM output based on the knowledge base. Preferably, the agent is configured to communicate with external systems and APIs to obtain additional information or perform actions within the software system.

[0017] In a second aspect of the invention, a configuration model is created based on the configuration of a software system, and the configuration model is output via a graphical user interface. This user interface allows visualization of the current configuration of the software system, enables manual configuration adaptation by the user when necessary, and optionally allows for inspection and verification of configurations generated by the agent and / or comparison of different configuration versions.

[0018] In another aspect of the invention, it is proposed that a configured software system be operated according to the first aspect of the invention, wherein the software system determines an output variable based on sensor variables or other input variables collected by one or more sensors (especially sensors from the manufacturing industry), and outputs a control variable based on the output variable.

[0019] This control variable can be used to control the actuators of a technical system. This technical system can be, for example, at least partially autonomous machines, at least partially autonomous vehicles, robots, tools, manufacturing or processing machine tools, or aircraft such as drones.

[0020] In other respects, the present invention relates to an apparatus and a computer program configured to perform the above-described methods, and to a machine-readable storage medium storing the computer program. Attached Figure Description

[0021] The embodiments of the present invention will now be explained in more detail with reference to the accompanying drawings. In the drawings: Figure 1 A schematic diagram of the overall architecture according to an embodiment of the present invention is shown; Figure 2 A schematic diagram of a portion of the overall architecture is shown; Figure 3 A schematic diagram of a method, apparatus, and storage medium according to one embodiment of the present invention is shown. Detailed Implementation

[0022] In one embodiment of the invention, Figure 1 The image shows a low-code MES application based on GenAI. This application typically has two areas: global configuration management 1a and GenAI-related components 1b.

[0023] Global configuration management 1a is designed so that users 2 can configure all aspects of the MES system through a preferred graphical abstraction of MES configuration elements (also known as the "low-code front-end") 3, which utilizes sophisticated technologies. The graphical front-end 3 is automatically generated and displayed based on configuration schemes stored for the corresponding MES components. Known methods such as 'React Flow' can be used to generate the graphical representation. In this case, icons or UI elements are used to depict production, where edges between elements represent data flows or other relationships between MES machines, stations, and software components. Preferably, the configuration scheme or configuration exists as a JSON file or in a database. It should be noted that, additionally or alternatively to the graphical abstraction, the configuration scheme can also be output via voice output or other media.

[0024] An example of a graphical representation of a configuration in an MES component configuration scheme is the "Loop Time" setting, which defines an integer in seconds, with an upper limit of "86400 seconds" and a lower limit of "10 seconds". The graphical representation can then display an input box via a low-code frontend, allowing numeric input such as 10 < input < 86400. Here, to optimally support the user, the value can be dynamically displayed in hours, minutes, and seconds.

[0025] The preferred set of UI elements is predefined and applied to the corresponding entries in the configuration scheme based on rules.

[0026] In another preferred implementation, the interface also supports graphical comparison of multiple versions of the configuration. For example, if a user plans to make configuration changes, they can graphically compare the changes with the configuration currently used in production (side-by-side or overlay comparison). Changes to individual configuration parameters can be compared step-down and corrected as necessary.

[0027] Optionally, the front end 3 can graphically display the physical and logical structure of the factory or production line in association with the MES components related to the corresponding process steps.

[0028] User 2 can also interact directly with GenAI component 1b via dialogue through interface 3. Changes made in this way are then displayed again on interface 3 in a graphical comparison manner. Preferably, user 2 then decides which parts should be changed.

[0029] The actual configuration of the MES system is centrally maintained, versioned, and stored in the "Global Configuration Model 4". This component also contains every configuration scheme required by the UI.

[0030] MES components that are not directly based on the standardized configuration scheme can be adapted using a "Legacy Adapter" 6. This adapter 6 generates component-specific configurations from a general scheme stored in the global configuration model. This adapter includes first and second modules 7 and 8, which are respectively called the "Configuration Generator" and "MES Legacy Configuration". Module 7 is configured to convert the global configuration into a format usable by the Legacy Adapter. Module 8 is configured to represent persistent legacy configurations (e.g., database entries, files on disk, environment variables, etc.).

[0031] MES Figure 1 The figure is indicated by reference numeral 5 in the attached figure.

[0032] Component 1b related to GenAI is shown in Figure 1 On the right side. This part is on Figure 2 The interface from global configuration management to GenAI-related components is shown in a magnified format. The optional "Validator" module 9 serves as the interface. Module 9 preferably contains rule-based and heuristic checking routines to prevent the application of obviously broken configurations. Alternatively, module 9 may perform validation based on GenAI technology.

[0033] Configurations and configuration changes can be created either manually by the user or fully automatically through GenAI-based components. To achieve the latter, this invention proposes the use of an orchestrated intelligent agent (“MES Configuration Crew”).

[0034] exist Figure 2 The diagram illustrates agents 1 to 3, which have access to large language models (LLMs) 14 in different configurations and are configured to interact with each other, effectively forming an autonomous system capable of understanding and performing complex tasks. These agents leverage the capabilities of LLMs to process, understand, and respond to natural language. That is, the agents can interpret natural language input and perform actions accordingly. Furthermore, these agents are designed for autonomous processing. This means they can make decisions and perform tasks without human intervention based on information and instructions obtained through natural language. Additionally, the agents can interact with users in real time. It is conceivable that through continuous training and feedback, LLMs and LLM-based agents can improve their capabilities and adapt to new tasks and contexts.

[0035] like Figure 1 and Figure 2 As shown, multiple agents are used to automatically discuss the configuration of MES 5.

[0036] In a preferred embodiment, three agents are used and initialized as described below. Initialization can be accomplished by providing corresponding text to the corresponding LLM of the agent to configure the role of the LLM.

[0037] Agent 1 is a manufacturing expert. Agent 1 understands the relevant production domain and possesses corresponding manufacturing knowledge. This production domain expertise can be provided to the LLM through various methods. The agent's task is to understand and optimize the production facilities and the production lines they contain.

[0038] Agent 2 is the MES Application Engineer. Agent 2 is familiar with MES products, preferably LLMs specifically designed for MES. The agent's task is to apply MES products to the appropriate application scenarios.

[0039] Agent 3 is a configuration expert. Agent 3 is an expert in writing and documenting MES product-specific configurations. Its task is to read and write the configurations in a specified format, and optionally generate human-readable documentation for those configurations.

[0040] In addition to the aforementioned role of the LLM through text input describing the agent's role and / or task (also known as contextual cues), other techniques can be used to additionally or alternatively endow the LLM with its corresponding characteristics and / or expertise.

[0041] One of these techniques is so-called fine-tuning of the corresponding LLM for the agent. Fine-tuning of an LLM refers to the process of adapting a pre-trained language model to a specific task or domain to improve its performance in those domains. In fine-tuning, the pre-trained model is further trained using a specific dataset relevant to the target task or domain. This data can come from various sources. Typically, the agent's data consists of other documents related to its task. For agent 1, the data could be, for example, data directly from a factory and generated during the production process. For agent 3, other data used for fine-tuning could be pre-created configuration schemes.

[0042] It has been demonstrated that by fine-tuning using historical production data and data from past application projects (especially for agents 1 and 2), their responses can achieve exceptionally high accuracy.

[0043] Known LLMs can be used for these agents. Preferably, the GPT model, especially GPT4o or a similar model, is used for agent 1; and the Mistral Instruct, CodeLlama or a similar model is used for agents 2 and 3.

[0044] Another technique besides fine-tuning is the so-called "Retrieval-Augmented Generation" (RAG), which is a method that combines the advantages of information retrieval systems (Retrieval) and generative language models (Generation) to improve the quality and relevance of generated answers.

[0045] Typically, RAG begins with an input request, which the retrieval module responds to by searching a large database for relevant documents or text passages. This retrieved information then serves as the context for the LLM, upon which the LLM generates coherent and well-founded responses.

[0046] Agent specialization can be achieved through Retrieval-Enhanced Generation (RAG), for example, by preparing data using a Small Language Model (SLM), allowing the LLM to respond to prompts without requiring retraining. Preferably, user manuals, integration guidelines, process descriptions, machine manufacturer documentation, and other sources from the MES system are used, and the data is vectorized using RAG and stored in database 10. Preferably, each agent has its own RAG database containing knowledge of task specialization specific to that agent.

[0047] Database 10 or more can be populated using different scrapers 11, 12, and 13. For example, website scraper 11, PDF scraper 12, and code document scraper 13 can be used.

[0048] In a preferred implementation of Agent 3, the agent is specialized using fine-tuning with RAG and LLM. For fine-tuning, multiple YAML files containing valid deployment configurations for the respective software products can be used in this case. These files describe graphs, which depict the interactions between MES modules.

[0049] Optionally, so-called tools can be predefined for the agent. These tools are programs that the agent can execute independently and autonomously decide when to execute.

[0050] It is possible to use only two agents. Preferably, only a manufacturing expert and an MES application engineer are used. In this case, the latter can jointly undertake the task of configuring the agents. However, it is also possible to use more than three agents. For this purpose, it is preferable to add another agent to the aforementioned agents, such as a "Cost Manager," whose role is to create configurations that only operate within a predefined cost framework.

[0051] The agents 1 to 3 described are each assigned a task and work together to solve that task in order to produce the best possible result. The task can be either specified by the user (e.g., through chat interaction on the front end) or a repetitive and pre-set routine task by the system operator or product manufacturer.

[0052] A task can be represented as follows: • “Create initial MES configuration for production line XY” • "Expand the currently valid configuration to reflect the packaging process described in the following documents" • “Migrate the configuration to ensure its compatibility with NEXEED IAS 2025.01.”

[0053] The agent communicates based on its task and expertise, generating a corresponding adapted configuration version. This version is created as a new configuration version in global configuration management and can optionally be submitted to the user for review.

[0054] It should be noted that the concept according to the present invention can be used alternatively to configure any complex software system, especially as long as the necessary documentation (such as a guide) for the software system exists.

[0055] Figure 3 A flowchart of a computer-implemented method (100) for automating the configuration of software systems, particularly manufacturing execution systems (MES), is schematically shown. This method (100) is characterized by the use of interacting software agents, each utilizing a large language model (LLM) and contributing to the completion of configuration requests based on its role and expertise. To maximize agent performance, they can be additionally specialized for their respective roles through fine-tuning and gain access to a knowledge base containing relevant documentation, guidelines, and historical configuration data. Each agent possesses its own specialized LLM, which is trained through fine-tuning using data relevant to its role. Communication between agents occurs via a central communication channel, in which agents exchange information and intermediate results.

[0056] The method (100) begins with receiving (S21) a configuration request. This request may be transmitted to the system in different ways, such as by direct input from a user via a graphical user interface, import from a file, or by an automatic trigger (e.g., due to a corresponding change in the system). The configuration request itself may contain either a natural language description of the desired changes or a formal specification written in a language that the system can understand.

[0057] In the next step (S22), the configuration request is forwarded to the participating software agents for processing. These agents are the domain experts, application experts, and configuration experts explained above.

[0058] Each agent analyzes the configuration request from its own perspective and generates its own suggestions and solutions. These suggestions and solutions are shared and discussed with other agents via a central communication channel. Through iterative negotiation and verification of the proposed configuration parameters, the agents gradually approach the (optimal) solution.

[0059] Here, domain experts verify, for example, whether the proposed configuration is compatible with production processes and guidelines. Application experts check the technical feasibility and consistency of configuration parameters in the MES system. Configuration experts ensure that the final configuration conforms to the presets of the global configuration model and is correctly recorded.

[0060] Throughout the approach, agents can additionally access a central knowledge base to locate missing information, review best practices, or analyze historical configurations. For example, domain experts can access machine manufacturer documentation to check the compatibility of new machines with existing production systems. Application experts can refer to code examples and configuration templates to simplify the implementation of configuration changes within the MES system.

[0061] In the final step (S23) of method (100), the final configuration integrated by these agents is stored in the global configuration model of the software system. Optionally, this configuration can also be verified by the user or activated directly in the MES system.

[0062] also, Figure 3 An apparatus (300) is shown configured to perform method 100. The apparatus has a machine-readable storage medium (301) on which a computer program is stored. The computer program contains instructions that, when executed by a computer, cause the computer to perform method 100.

Claims

1. A computer-implemented method (100) for automatically configuring a software system, comprising the following steps: Receive (S21) a configuration request describing the desired change or creation of the configuration of the software system; The configuration request is processed (S22) by multiple interacting software agents, each using at least one large language model (LLM), wherein each agent contributes to completing the configuration request based on its role and expertise, and wherein the agent plays at least the following roles: The roles of the domain expert and the application expert are as follows: the domain expert possesses professional knowledge of the application domain of the software system; the application expert possesses knowledge of the software system and its configuration possibilities, and can translate the configuration request into specific configuration parameters; and the configuration expert can create configurations according to the specified format of the global configuration model. The new or adapted configuration of the software system is generated through the interaction of the intelligent agent (S23).

2. The method according to claim 1, wherein the interaction of the agent includes at least the following interaction possibilities: a) Exchange information and knowledge about the configuration request, the application domain, and the software system; b) Configuration options are discussed and negotiated based on the expertise of each agent; and optional c) Validate the proposed configuration parameters according to the rules and constraints of the global configuration model.

3. The method according to any one of the preceding claims, wherein the agent additionally has the following characteristics: a) Access a knowledge base containing relevant documentation, guidelines, best practices, and historical configuration data for the software system; b) Use Retrieval Enhancement Generation (RAG) technology to search the knowledge base; c) Communicate with external systems and APIs to obtain additional information about the software system.

4. The method according to any one of the preceding claims, wherein the graphical user interface provides the global configuration model, which implements: a) visualizing the current configuration of the software system; b) manually adapting the configuration by the user; c) checking and verifying the configuration generated by the agent; and d) comparing different configuration versions.

5. The method according to any one of the preceding claims, wherein the software system is a manufacturing execution system (MES), and the application field includes manufacturing control.

6. An apparatus (300) configured to perform the method according to any one of the preceding claims.

7. A computer program comprising instructions that, when executed by a computer, cause the computer to perform the method according to claims 1 to 5.

8. A machine-readable storage medium (201) having a computer program as claimed in claim 7 stored thereon.

9. An application of the determined configuration in a software system using the method according to any one of claims 1 to 5.

Images