Method and system for auxiliary control of machine production line

Abstract

One embodiment of the present invention relates to a method and system for controlling a machine production line, in particular in a machine production line for filling and packaging food and / or beverages. The machine production line comprises a plurality of machines. The embodiments include implementing a Large Language Model (LLM) on a server and setting a control instruction context and a particular context for the LLM. The control instruction context includes a set of control instructions and grammar thereof that can be used for the machine production line. The particular context includes constraints for an output of the LLM. After state data of the machine production line is input into the LLM, a user request for outputting a control instruction of the machine production line can be acquired via a user interface. The LLM processes the user request and creates an answer based on the set control instruction context, the specific context, and the state data. The answer includes control instructions generated by the LLM and executable by the machine line.

Description

[0001] This invention relates to a method and system for auxiliary control of a machine production line, particularly for auxiliary control of a machine production line for filling and packaging food and / or beverages, and a machine production line.

[0002] Today, filling and packaging equipment in the liquid food industry is highly optimized and can process up to 120,000 units per hour. Typical filling and packaging equipment usually consists of multiple different machines and modules connected by conveyor belts. Here, these units (such as bottles, cans, pallets, bundles, etc.) are transported from one machine to another, and pass through the machines in the production line in a predetermined sequence.

[0003] Numerous assessments have shown that once factory acceptance testing is completed at the customer's site and installers leave the site after installation of the machine on the production line (e.g., beverage filling equipment), the customer's total equipment efficiency (OEE) typically declines significantly. The reasons for this are varied. It could be due to user lack of knowledge, improper machine operation, inefficient troubleshooting, and many other factors.

[0004] For these reasons, digital solutions are typically offered that integrate data from beverage filling equipment at a higher level and provide a production overview. Customers can then view various production metrics, such as "number of failures per machine," "number of containers produced," "number of containers rejected," and "OEE," to conduct more detailed analysis and derive measures to ensure that problems / failures do not recur in the future or to quickly resolve problems / failures when they do occur.

[0005] In the digital age, the automation of complex machinery plays a crucial role. Traditional control systems rely on pre-programmed algorithms and fixed process schemes to operate machines. While these systems are mature and efficient, they often lack the self-learning and adaptive capabilities needed to cope with unexpected situations or changes in the production environment.

[0006] Against this backdrop, the application of artificial intelligence (AI) has become increasingly widespread in recent years. AI-based systems can learn from data, recognize patterns, and make decisions accordingly. They can continuously learn from their own performance and improve over time, making them ideal for controlling complex machinery. However, existing AI-based control systems are still imperfect.

[0007] Artificial intelligence training for controlling machinery relies on the availability and quality of data. AI models typically require large amounts of high-quality training data to make accurate predictions and effective decisions. This data usually originates from the machinery's operational history and experiments under different operating conditions.

[0008] However, significant challenges exist with highly specialized machines. For example, the amount of available training data may be limited. This is because specialized machines are typically designed for specific applications and therefore cannot withstand a wide range of operating conditions. Furthermore, such machines are often expensive and difficult to operate, making it challenging to conduct experiments to collect data.

[0009] Moreover, the quality of available data can often be low. Highly specialized machinery, such as that in modern beverage filling equipment, is typically complex and has many variable parameters, making it difficult to collect accurate and complete data.

[0010] These limitations make training AI models for controlling highly specialized machinery a major challenge. Traditional AI models, which rely on learning from large amounts of training data, often fail to operate effectively in such environments. They may struggle to make accurate predictions and effective control decisions, resulting in poor performance and inefficiency.

[0011] Therefore, there is a need to provide a more efficient way to achieve computer-aided control of complex machine production lines.

[0012] Therefore, there is a need to improve the systems and methods used for computer-aided control of machine production lines.

[0013] According to the present invention, this objective is achieved by the method according to claim 1 and the system according to claim 7. Embodiments and improvements are included in the dependent claims.

[0014] One embodiment of the present invention relates to a method for controlling a machine production line, particularly in a machine production line for filling and packaging food and / or beverages. The machine production line includes multiple machines. The method includes implementing a Large Language Model (LLM) on a server and setting control instruction contexts and specific contexts for the LLM.

[0015] Here, the server can be designed as or include a server, industrial PC, or edge device. The specific design depends on, for example, the requirements of the LLM model used and / or the type of communication. For instance, the LLM model can be hosted externally, i.e., at a service provider (such as a server farm operator). It can then communicate with the machine production line via an API. However, the LLM model can also be hosted internally at the machine production line operator.

[0016] The control instruction context includes a set of control instructions and their syntax that can be used on the machine production line. The specific context includes constraints for the LLM output. After the status data of the machine production line is input into the LLM, a user request for outputting control instructions for the machine production line can be obtained via a user interface. The LLM processes the user request and creates a response based on the set control instruction context, the specific context, and the status data. The response includes control instructions generated by the LLM and executable by the machine production line.

[0017] One embodiment of the present invention relates to a system having an LLM, a server, and a user interface.

[0018] Exemplary aspects of the invention are illustrated in the accompanying drawings. In the drawings: Figure 1 A schematic diagram is shown, illustrating an overview of the basic elements and structure of the invention. Figure 2 This diagram illustrates the phased adjustment of a large language model for application in a machine production line. Figure 3 An exemplary flowchart is shown for implementing a large language model and its operation in a machine production line; Figure 4 An exemplary device configuration for PET containers and adhesive strapping is shown; Figure 5 An exemplary equipment configuration for PET containers and shrink packaging machines is shown; Figure 6 An exemplary device configuration for jars or glass bottles is shown; and Figure 7 An exemplary device configuration for a can is shown.

[0019] Figure 1 For illustrative purposes only, this diagram illustrates an overview of the basic elements and structure of the present invention. According to one embodiment, these basic elements may be a machine production line 100, a cloud or computer server 110, and a user interface 120 through which a user 130 can interact with the server 110 and, where appropriate, with the machine production line 100.

[0020] Machine production line 100 may be, for example, a machine production line for filling and packaging food and / or beverages, but is not specifically limited to such a production line. Different exemplary embodiments of such machine production line 100 are described in... Figures 4 to 7 The text is shown and explained.

[0021] Computer server 110 is not necessarily a single server, but may consist of one or more computer devices that are directly or indirectly connected to each other. As described in more detail below, server 110 may include a Large Language Model (LLM) and provide an interface between the LLM and user interface 120. Server 110 may be directly or indirectly connected to machine production line 100. According to embodiments, server 110 and machine production line 100 are interconnected via an Internet connection, or they may have a direct communication connection. That is, the LLM on server 110 and machine production line 100 can communicate directly with each other. Through user interface 120, user 130 can monitor or control the communication between server 110, LLM, and machine production line 100.

[0022] User interface 120 allows user 130 to interact with server 110 or at least with specific services on server 110. Here, the user interface can take any form suitable for enabling the user to interact with the computer via input. The user interface is also known as a "human-machine interface" (MMS) or "human-machine interface" (HMI) and, in some cases, allows user 130 to observe equipment status and intervene in processes outside of operating machine production line 100.

[0023] Here, the user interface can be a simple computer, mobile phone, tablet, dedicated device of machine production line 100, voice assistant, computer application, computer terminal or other computer-aided device.

[0024] As previously mentioned, a so-called Large Language Model (LLM) can be implemented on server 110. These large language models, also known as language models, are a type of artificial intelligence (AI) trained to understand and generate human language. They learn by analyzing large amounts of text data and identifying patterns within it. This involves not only individual words and sentences but also the context in which they are used. A specialized language model called "Transformer" can recognize more complex contexts and learn to generate answers to questions, translate text, and even generate graphics (such as data graphs and diagrams). The known architecture of "Transformer" is based on a method called "Attention," which helps the language model understand the context of a word relative to all other words in a text paragraph.

[0025] These neural Transformers consist of multiple layers, each containing a large number of neurons. During training, each layer and each neuron is adjusted to better respond to the training data. The model's "depth" (number of layers) and "width" (number of neurons per layer) determine the complexity of the patterns the model can recognize.

[0026] The training process requires significant computational power and time, but once the model is trained, it can be used to generate or interpret text based on the patterns it has learned. These types of LLMs are offered pre-trained, fully-trained versions by various vendors, allowing end users to use "off-the-shelf" LLMs.

[0027] Embodiments of the present invention relate to integrating such an LLM into a user platform to provide computer-aided control or at least computer-assisted control of a machine production line 100. According to embodiments of the invention, the LLM is specifically tailored to provide requested data about the machine production line 100 to the user 130. This requested data may be processed information about the machine production line 100, but may also be control instructions executable by the machine production line 100. Safety considerations and data transparency are taken into account when tailoring the LLM to ensure that erroneous information is not provided to the user and that control instructions do not initiate unexpected processes in the machine production line that could potentially damage the machine production line 100.

[0028] The LLM adjustment plan is in Figure 2 This is shown in more detail below. Figure 2 The basic architecture in Figure 1 The architecture is the same, and it includes machine production line 100, server 110, and user interface 120. Furthermore, Figure 2 An adjustment to the LLM 200, which is implemented exemplary on server 110, is also shown, as well as the output of control commands from the LLM 200 to machine production line 100.

[0029] Here, LLM 200 can be a pre-trained general LLM that may not initially possess extended knowledge about machine production line 100 or the data generated by the machine production line. Therefore, context needs to be provided for the LLM first.

[0030] The fundamental idea behind applying LLM 200 to the control of machine production line 100 is to describe the control rules and principles of the equipment in a formal language and then impart this description to the LLM. LLM 200 no longer requires specific training data for machine production line 100 representing various operating conditions of the machine equipment; instead, it learns how to control machine production line 100 from these formal rules (i.e., from control instructions and their syntax) and operational history (e.g., from state data and / or the changes in state data and the corresponding control parameters).

[0031] This offers numerous advantages. LLM 200 does not require specific training data from machine production line 100, thus circumventing the limited data availability issues inherent in highly specialized machinery. Furthermore, because LLM 200 is text-based, it can be easily adapted and updated by adding new rules, parameters, syntax, or principles. Additionally, LLM 200 is capable of recognizing and learning complex relationships and patterns within these rules and associations, enabling it to make effective control decisions even in complex and dynamic operating environments.

[0032] In addition, LLM 200 offers the advantages of transparency and traceability. Its decisions are based on clear and explicit rules and learned syntax, which are easy to understand and verify. This is especially important in safety-critical applications where understanding and monitoring the behavior of control systems is essential.

[0033] Therefore, the use of LLM to control or assist in the control of machine production line 100 described in this paper provides an innovative and promising solution that overcomes the challenges of traditional AI models and paves the way for more efficient, reliable, intuitive and adaptable control systems.

[0034] However, to implement LLM 200, it must first be prepared for the specific application of (auxiliary) control.

[0035] like Figure 2 As shown, the “General LLM” 200 thus becomes a “Contextual LLM” 200. This context can consist of a specific context and a control instruction context, as further described below.

[0036] A specific context can be tailored to a completely specific equipment configuration for machine production line 100, or it can be tailored to a specific user group. A specific context sets constraints on the LLM output. Such constraints can relate to, for example, the form, structure, information sources used, or level of detail of the LLM response. Furthermore, a specific context may include specific instructions or descriptions for calculating specific metrics or information.

[0037] To precisely define the specific context of the LLM, appropriate specific text instructions can be set. These instructions can also be user-specific (e.g., what exactly customer plant managers and installation personnel can view). The expected form and structure of the LLM response can also be defined here.

[0038] In order to find the cause when calculated indicators, such as abnormally low values, can be found, the LLM can also be provided with facts about the calculation of the indicators and the causal relationship.

[0039] Since a pre-trained LLM can "understand" and accept natural language as input, in one implementation, a specific context can also be expressed as such text input.

[0040] A simplified, exemplary specific context might look like this: You are a name Chat2Act The intelligent assistant helps users answer questions about the data generated by their machine production line and can output control instructions for the machine production line. Only use facts listed in the provided source list to answer questions. Only create control instructions based on these facts. If the information is insufficient, say you don't know. Do not generate answers or control instructions that are irrelevant to the provided sources or the context of the control instructions. If it would be helpful to ask the user clarifying questions, please do so. This specific context is merely an example and may be more detailed and restrictive in actual implementation. This specific context may, for example, be provided to LLM 200' by the manufacturer of machine production line 100. The specific context can be entered into LLM 200' in such a way that subsequent user 130 interacting with the LLM cannot modify the specific context, and therefore cannot remove the constraints set by the specific context again.

[0041] As mentioned above, a control instruction context is also provided to the LLM 200, which describes the output rules of the control instructions in more detail. The control instruction context may include a set of control instructions that can be used on the machine production line and their syntax. For example, the control instruction context may specify one or more parameter value ranges for a set of control instructions, within which parameters can be changed via the control instructions. The control instruction context may also prohibit specific parameter ranges or parameter changes, for example, for safety reasons, because such parameter ranges or parameter changes may damage the machine production line 100.

[0042] However, the context need not be strictly divided into a specific context and a control instruction context. The two contexts can also be combined into a single context.

[0043] However, it is still not possible to obtain specific control instructions for machine production line 100 using "LLM with context" 200, because data or status data of machine production line 100 must first be entered into the LLM (e.g., as a source) in a further step.

[0044] Data 230 can be loaded directly from machine production line 100 to cloud 110, for example, or it can be transmitted to cloud 110 under the control of a third party. This data 230 constitutes a database for responding to user requests and creating control instructions for machine production line 100. In addition to control instructions that can be sent directly to machine production line 100, the responses may also include metrics, sensor data, and / or machine data directly collected or calculated from machine production line 100 or individual machines in machine production line 100, as described below. For this purpose, an existing database can be used, which also serves as a source for existing services.

[0045] Depending on the implementation, it may be necessary to extract appropriate data from database 231, such as data from a specific production line during one or more time periods.

[0046] Depending on the implementation, it may also be necessary to convert the data 231 extracted from the database into a form suitable for use in an LLM so that the LLM can process it. This processing step is shown within a data frame 230, which illustrates the raw data 231, which is transformed into processed data 232 by extracting and processing appropriate content from the data 231. This may include converting numerical values ​​to appropriate units (e.g., timestamps, time differences, percentages), calculating ratios, and reconstructing larger cubes. The processing may also include adding data context to the data, such as labels about the data content and meaning, so that the LLM can correctly interpret the data.

[0047] As shown in data frame 230, this data preprocessing can be performed at fixed time intervals, with pre-computation and persistent storage periodically for specific time periods to minimize the latency of LLM processing user requests. In other words, processing data over longer periods (weeks, months, quarters) may take a long time. Therefore, for these cases, the data should be pre-computed and persistently stored to minimize the waiting time for user 130.

[0048] After completing the input of state data into the LLM with context, an "LLM with context and state data" 200'' is created, which is ready to be used in the cloud 110.

[0049] LLM 200'' can be invoked by user 130 via user interface 120 and can be used as a digital assistant to adjust the data and information output of the machine production line, and can also be used to directly control the machine production line 100 according to the present invention.

[0050] User 130 can now interact with LLM 220'' and send requests to LLM 200'' via user interface 120 to obtain requested data. This data may be information about the capacity of machine production line 100, or it may be control instructions that should be output by LLM 200''. These control instructions can be directly communicated to machine production line 100 and executed there (e.g., with user permission).

[0051] The structured output / response of LLM 200 can be used to provide data to third-party systems that do not understand natural language. For example, this can be used to create operating instructions for users in machine production line 100, or to store structured information for documentation purposes. Furthermore, this structured data can be used to control machine production line 100 with appropriate parameters.

[0052] This is Figure 2 As shown in the diagram, data packet 240 is sent directly from LLM 200'' to machine production line 100 and includes one or more instructions or control commands for setting up machine production line 100. Here, these instructions or control commands are in a machine-readable file format. The machine-readable file format can be, for example, JSON or XML.

[0053] The operability of the user platform of the machine production line 100 is thus improved, because the user 130 is supported by LLM in a question-and-answer format and can be navigated to the required information or obtain the required control commands more quickly, thereby controlling the machine production line 100 according to the user's wishes.

[0054] According to another aspect of the invention, the response of LLM 200'' may also include operating instructions for the user, for example as a supplement to or alternative to control instructions sent directly to machine production line 100.

[0055] When providing operating instructions to users, multiple operating options can be offered. This is especially important when responding to complex control situations involving events across an entire machine production line 100, where the solution is not always "unique." Often, even adjusting a single value can trigger further changes in the process.

[0056] In this regard, for models like the LLM 200'' with a wide range of operating options, alternative operating procedures can be provided. Thus, for example, in response to an impending failure identified through response analysis, maintenance can be recommended as soon as possible, while continuing operation at the same (maximum) speed. Alternatively, maintenance can be postponed and the overall production line output reduced to avoid damaging the component requiring maintenance.

[0057] These decision options are available to users through LLM 200''. However, because LLM 200'' has access to all data, in addition to operating the alternatives, it can also indicate the impact of the corresponding alternatives on the production line (e.g., OEE, delaying subsequent production, etc.). This makes it easier for users to select alternatives.

[0058] Figure 3 An exemplary flowchart is shown for implementing LLM and its operation in machine production line 100.

[0059] Method 300 begins with step S302, in which a large language model is implemented on server 110. This implementation process may include implementing a specific general language model optimized for a particular language (English, German, French, Spanish, Chinese, etc.) or mastering multiple languages.

[0060] In step S304, a context is set for the LLM. This context may consist of a specific context and a control instruction context. The specific context may set constraints on the LLM's output. The set constraints may constrain at least one of the following: the form, structure, information source used, and level of detail of the LLM's output.

[0061] The control instruction context includes a set of control instructions and their syntax that can be used on a machine production line. The LLM takes this set of control instructions and their syntax into account when creating control instructions.

[0062] Both specific contexts and control command contexts can be set and entered into the model by the manufacturer of machine production line 100, online service operator, or other organization.

[0063] In step S306, the status data of machine production line 100 is input into the LLM. The status data may include directly acquired or calculated metrics, sensor data, and / or machine data. The status data may originate from one or more specific time periods. In an optional sub-step of S306, the status data may also be preprocessed. Preprocessing of the status data of machine production line 100 includes at least one of the following operations: (i) converting one or more values ​​in the status data into one or more suitable units that can be processed by the LLM, (ii) calculating ratios, (iii) reconstructing the cube, and / or (iv) adding data context to the data.

[0064] In the next step S308, a user request for outputting machine production line control instructions is received via user interface 120. For example, user request 130 may request the LLM to output control instructions to the user or machine production line 100 to resolve previously identified problems in the current operation. This request is merely an example.

[0065] The user request is input into the LLM and processed by the LLM in step S310 based on the specific context set and the state data.

[0066] Finally, in step S312, a response to the user's request is created. This response may include control instructions generated by the LLM and executable by the machine production line. The provision process may also include forwarding these instructions to the user 130 who issued the request. The process by which the LLM provides the requested information may further include providing a description of one or more sources used by the LLM in determining the requested information. This description of one or more sources may be presented in text form and may be selected and associated with a corresponding source by the user 130. For example, the user may click on a source and, in addition to the LLM's response, access a database that forms the basis for the LLM's response.

[0067] In the final step S314, the generated control command can be output to the machine production line 100. According to one embodiment, the control command is not automatically executed by the machine production line 100, but first requires user confirmation that the control command should be executed by the machine production line.

[0068] In the following Figures 4 to 7 Various exemplary device configurations for various bottle filling devices are described herein, in which the present invention or at least parts and aspects thereof may be implemented. Figures 4 to 7 The description is intended only to provide a general overview of the machines for which state data can be collected, and on which the LLM can process user requests.

[0069] Figure 4 An exemplary device configuration 1000 for PET bottles or PET containers and adhesive strapping is shown. Figure 4 As shown, equipment configuration 1000 includes various modules that form a production line at the end of which filled PET containers are output on pallets in the form of bundles. Some of the modules and machines may be optional, and the invention is not limited to the specific form and arrangement of the equipment configuration.

[0070] Equipment configuration 1000 includes an oven 1002 for preforms, a preform sorter 1004 with a feeder, and a blow molding machine 1008. Modules 1002, 1004, and 1008 typically form a stretch blow molding machine in which PET containers are made and shaped from initial materials. The produced PET containers are then conveyed to a filler 1010, where bottles are filled. The filler may optionally include a rinsing device. During storage or transportation, various particles, such as dust, cardboard, or wooden pallet residue, may accumulate in the preforms. These particles can be removed using the rinsing device. A sealing machine may be arranged at the end of the filler to seal the PET containers after filling.

[0071] Optionally, the equipment configuration 1000 may include a rotating device after the filling machine 1010 for hot filling of PET containers. The filled PET containers are conveyed to a separator 1020 and then to a drying unit 1024 via one or more conveyor belts 1016 (which may also include a buffer 1018 for intermediate loading of filled containers), where the PET containers are dried.

[0072] After drying, the PET containers are conveyed to labeling machine 1026. Labeling machine 1026 can be designed for various labeling techniques, such as hot glue, cold glue, self-adhesive labels, or sleeve labels. After printing or labeling, the PET containers are conveyed to handle applicator via second drying unit 1028, production line distributor 1030, conveyor belt 1032, adhesive bundle production device 1034, and curing path. In adhesive bundle production device 1034, PET containers are grouped together in specific bundle sizes and packaged into bundles, such as "six-packs". In handle applicator, handles are attached to the containers, making it possible to comfortably carry the bundles. The finished bundles are then correspondingly arranged into layers by robot 1042 and packed on pallets by palletizer 1044.

[0073] In equipment configuration 1000, so-called format carriages or format racks can be arranged at various modules and machines to provide quickly changeable format kits for short changeover times and automated tool switching. Examples of format carriages are format carriage 1006 for blow molding machine 1008, format carriage 1012 for filling machine 1010, format carriage 1022 for labeling machine 1026, format carriage 1038 for adhesive bundling production device 1034, and format carriage 1046 for palletizer 1044.

[0074] Figure 5 Another exemplary device configuration 1100 for PET containers and shrink packaging machines is shown. The device 1100 in Figure 10 includes components from… Figure 4 The equipment configuration includes 1000 modules and many modules and machines in the machine; however, some differences exist. Therefore, for Figure 10, the modules that have been combined are omitted. Figure 4 The description of the module.

[0075] A significant difference between the two exemplary device configurations 1000 and 1100 is that the labeling machine 1126, with labeling module 1127, can be installed after the blow molding machine 1008 and before the filling machine 1008. For this purpose, device configuration 1100 may include up to six transport tracks 1150 into which PET containers can be extruded. After the PET containers have been correspondingly extruded into one of the six tracks 1150, they are conveyed to a film wrapping module 1152 and then to a shrink tunnel 1154.

[0076] Figure 6 An exemplary device configuration 1200 for use with jars or glass bottles is shown. Figure 6 The exemplary device configuration 1200 is again with Figure 4 and Figure 5 The device configurations 1000 and 1100 have some similarities, therefore the description of the device configuration is limited to the differences in device configuration.

[0077] like Figure 6 As shown, an exemplary device configuration may include two separate feed sections. Figure 6 The first feed section on the left shows a branch for cans, or optionally a sub-branch for new reusable bottles. Here, the containers (i.e., cans or new bottles) are guided into the machine by the depalletizer 1302, where they are guided to the filler 1010 via a conveyor belt. Figure 6 The second feed section on the right shows a sub-branch for reusable bottles, which are introduced into the equipment from the reusable sorting device (not shown).

[0078] In cases where used reusable bottles are introduced into device 1200 via a sub-branch for reusable bottles, the reusable bottles first pass through a cleaning or washing machine 1304. Another possible difference in the exemplary device configuration 1200 is the addition of a transshipment packaging machine 1306 after the labeling machine 1026. The transshipment packaging machine can sort bottles or cans into cardboard clip application devices or boxes, or both.

[0079] Figure 7An exemplary device configuration 1300 for cans is shown, wherein elements already described in other device configurations are not described again. In device configuration 1300, cans are introduced from a can magazine 1402 containing cans into a depalletizer 1302. After the cans have passed through a filling machine and been filled, they are sealed by means of a sealing magazine 1404, and the cans are further conveyed along device 1400 via a conveyor belt, as described above.

[0080] If not required, the optional pasteurizer 1408 can be bypassed via bypass 1412. In pasteurizer 1408, freshly filled products can be pasteurized for preservation.

[0081] Compared to device configurations 1000, 1100, and 1200, exemplary device configuration 1300 shows different tanks for corresponding consumables, such as tank 1410 having rinsing liquid and / or filling product, and tank 1406 having lubricant. These tanks may also be included in the exemplary device configurations already described above. For example, chemical product 106 conveyed from mixer 110 to the machine may be stored in tanks 1406 and 1410.

Claims

1. A method for controlling a machine line (100), in particular in a machine line for filling and packaging food and / or beverages, wherein the machine line comprises a plurality of machines, and wherein the method comprises: implementing (S302) a large language model, LLM, on a server (110); setting (S304) a control instruction context and a specific context, wherein: the control instruction context comprises a set of control instructions and their syntax that can be used for the machine line, the specific context sets constraints for the output of the LLM, and the set constraints limit at least the output format, structure and / or the used information source of the LLM’s answer; inputting (S306) state data of the machine line into the LLM; receiving (S308) a user request regarding the machine line via a user interface (120); processing (S310) the user request by the LLM based on the set control instruction context, the specific context and based on the state data; creating (S312) an answer to the user request based on the processing, wherein the answer comprises control instructions generated by the LLM and executable by the machine line; and outputting (S314) the generated control instructions by the LLM.

2. The method according to claim 1, further comprising: obtaining a user confirmation that the control instructions shall be executed by the machine line.

3. The method according to claim 1 or 2, wherein the control instruction context further specifies one or more parameter value ranges of the set of control instructions within which a parameter is allowed to be changed by the control instruction.

4. The method according to any one of claims 1 to 3, wherein the control instructions are output in a machine-readable file format, and wherein the machine-readable file format is JSON or XML.

5. The method according to any one of claims 1 to 4, wherein the state data comprises at least directly acquired or calculated indicators, sensor data and / or machine data, and wherein the state data is from one or more specific time periods.

6. The method according to any one of claims 1 to 5, further comprising: pre-processing the state data of the machine line, wherein the pre-processing of the state data comprises at least one of: converting one or more values in the state data into one or more suitable units that can be processed by the LLM, calculating ratios, reconstructing multi-dimensional data sets, aggregating data, and adding a data context to the data.

7. A system for controlling a machine line, in particular in a machine line for filling and packaging food and / or beverages, wherein the machine line comprises a plurality of machines, and wherein the system comprises: a server on which a large language model, LLM, is implemented, wherein a control instruction context and a specific context are set for the LLM, wherein: the control instruction context comprises a set of control instructions and their syntax that can be used for the machine line, the specific context sets constraints for the output of the LLM, and the set constraints limit at least the output format, structure and / or the used information source of the LLM’s answer. The specific context sets constraints for the output of the LLM, and The constraints set limit at least the output format, structure of the answer of the LLM and / or the used information sources; and a user interface for interacting with the LLM; wherein the LLM is adapted for: inputting state data of the machine production line; receiving a user request regarding the machine production line via the user interface; processing the user request based on the control instruction context set, the specific context and based on the state data; creating an answer to the user request based on the processing, wherein the answer comprises control instructions generated by the LLM and executable by the machine production line; and outputting the control instructions generated.

8. The system of claim 7, wherein the user interface is a web user interface, a voice assistant or a computer application.

9. The system of claim 7 or 8, wherein the user interface is adapted for obtaining a user confirmation that the control instructions shall be executed by the machine production line.

10. The system of any one of claims 7 to 9, wherein the control instruction context further specifies one or more parameter value ranges for the set of control instructions within which a parameter is allowed to be altered by the control instructions.

Images