Method for acquiring AI agent, method for using AI agent, control device, automation system, computer readable medium and computer program product

Abstract

The present disclosure relates to a method for acquiring an AI agent, a method of using the AI agent, a control device, an automation system, a computer readable medium and a computer program product. The invention relates to a method for obtaining an artificial intelligence (AI) agent suitable for driving an application. The method comprises obtaining (S310) training data indicative of predetermined drive operation data associated with a predetermined drive application and / or predetermined drive parameters associated with the predetermined drive application. The method further includes training (S320) a large language model (LLM)-based generative AI agent using the acquired training data to identify at least one of: first driving parameters at least partially related to the driving application, and a relationship between the first driving parameters.

Description

Technical Field

[0001] The present disclosure relates to a method for acquiring an artificial intelligence (AI) agent. The present disclosure also relates to a method for using the AI ​​agent. The present disclosure also relates to a control device, an automation system, a computer-readable medium, and a computer program product. Background Art

[0002] The execution and deployment of machine learning (ML) algorithms relies on data engineers and data scientists. Therefore, these roles require not only ML / analytics experience but also deep system and domain knowledge. Consequently, data engineers and data scientists face increasingly complex tasks and have become crucial for successfully driving analytical applications. However, the availability and performance of data engineers and data scientists can be limited, for example. Therefore, there is room for improvement. Summary of the Invention

[0003] To date, numerous developments and various products and services have been developed to support data analytics operations and algorithms for sports applications. Manipulating drive data and its relationships, the ML algorithms based on this data, and the data engineers and data scientists who develop these ML algorithms and applications are key elements for successful drive analytics applications. Among these key elements, data engineers and data scientists are crucial, as they must have a deep understanding of drive setup, drive parameters, operational scenarios, application usage, and the generated drive data. However, data engineers and data scientists have physical limitations, as they may not be available every day or at all times. For example, when available, they may not be fully focused and may not achieve optimal ML algorithm execution. These limitations are reflected in the drive analytics applications and their impact on sports applications. Consequently, data analytics applications may not be fully utilized to improve sports operations. Furthermore, if a well-trained data engineer or data scientist leaves an application or project, new data engineers or data scientists face a high learning curve to reach the same performance level as the departing data engineer or scientist. Training new data engineers or data scientists in this way is a time-consuming and resource-intensive task. Similarly, debugging drive applications and identifying optimal drive parameters for such applications also require expert knowledge. These processes are also time-consuming and resource-intensive. Furthermore, handling future errors after the driver application starts running may also require human expertise.

[0004] In view of the foregoing, and to address one or more of these issues, in a first aspect, a method for acquiring an artificial intelligence (AI) agent suitable for a driving application is provided. The method includes acquiring training data indicating predetermined driving operation data related to a predetermined driving application and / or predetermined driving parameters related to the predetermined driving application. The method also includes training a generative AI agent based on a large language model (LLM) using the acquired training data to identify first driving parameters at least partially related to the driving application and / or identify relationships between the first driving parameters.

[0005] It should be noted that the method may be a method for obtaining an AI agent suitable for a drive application in an industrial drive application system. The AI ​​agent may be an AI agent in the industrial drive application system and / or an AI agent related to the industrial drive application system.

[0006] It should be noted that the term "drive application" used throughout this application also encompasses a drive device and / or a drive (control) system. The drive device may be, for example, an electric motor, wherein the drive (control) system may be, for example, a system comprising the electric motor and another drive device.

[0007] It should also be noted that the expressions “predetermined driving operation data”, “predetermined driving application” and “predetermined driving parameters” can be understood to mean that the driving operation data, driving application and driving parameters are predetermined, preselected, prepared and / or pre-adjusted for training, i.e., so that suitable, applicable and / or appropriate training data can be obtained.

[0008] Regarding training data, it should be noted that the training data can be obtained from actual driver applications and / or can be synthetically developed. For example, the training data can indicate historical operations of the driver application, where the corresponding historical operation data is obtained from a database or historical database. Additionally or alternatively, the synthetically developed data can represent data generated by another AI, tool, and / or human. For example, the synthetically developed data can indicate simulated data based on which the operation of the driver application can be simulated.

[0009] It should be noted that the first driving parameter may be understood to mean a set of driving parameters.

[0010] With respect to the relationship between the first drive parameters, the relationship may indicate, for example, how one of the first drive parameters affects another of the first drive parameters. For example, the larger the value of one drive parameter, the larger (or smaller) the value of the other drive parameter. Thus, for example, the relationship may indicate any proportion, dependency and / or correlation, or any combination thereof. For example, the value of drive parameter "A" may be proportional to the value of drive parameter "B". Thus, increasing the value of drive parameter "A" may also cause the value of drive parameter "B" to increase. Furthermore, for example, if the value of drive parameter "B" may exceed a predetermined threshold, an upper limit value is provided to drive parameter "C". For example, where drive parameter "A" may indicate a torque associated with a motor, drive parameter "B" may indicate a speed associated with a motor, and drive parameter "C" may indicate a power supply associated with a motor.

[0011] The AI ​​agent obtained according to the method of the first aspect is advantageous because it can complete tasks, generate new tasks based on the obtained results, and prioritize tasks in real time. This AI agent further demonstrates the potential of AI-driven language models to autonomously perform tasks within various constraints and contexts. This AI agent can further open up a whole new world of applications, which are currently primarily performed by humans with the assistance of various analytical applications. Human expertise is not always required to drive analytical operations, and therefore human expertise can be used for other tasks. Therefore, human expertise can be more efficiently utilized as a resource. For example, human expertise can be used to design ML models and evaluate models and deploy LLM-based generative AI agents. Furthermore, human error can be reduced, optimization can be increased, and drivers and gateways can operate autonomously. New employee onboarding can be improved by providing more appropriate training. Furthermore, health monitoring, condition monitoring, and predictive analysis, to name a few examples, can be improved by being more supported, automated, and autonomous. This AI agent can obtain synthetic but highly accurate driver parameters, which can be used to train additional driver ML models. Assistance can be provided to monitor drive status and operating parameters using ML models from an ML model pool or repository. Furthermore, such AI agents can reduce operating and monitoring costs, reduce overall debugging time, identify optimal drive parameters in a reduced time, and / or support and guide customers in handling any future drive errors.

[0012] According to several examples of the present disclosure, the acquired training data may include historical operating data and / or simulated operating data. In addition, the acquired training data may also indicate a driving technical manual, which indicates a second driving parameter related to the driving application. The training may also include training to learn at least one of the following: a first relationship indicating a relationship between at least a portion of the first driving parameters based on the driving technical manual, and a second relationship indicating a relationship between at least a portion of the second driving parameters based on the driving technical manual. The training may also include training to develop a method for identifying a driving parameter at least partially related to the driving application based on at least one of the first relationship and the second relationship.

[0013] It should be noted that the second drive parameters can be understood to represent a set of drive parameters. The second drive parameters may be different from the first drive parameters, may have a certain overlap of at least one identical drive parameter with the first drive parameters; or may be the same as the first drive parameters. For example, for the purpose of explanation, it can be assumed here that the first drive parameters include drive parameters "A", "B" and "C". In addition, from the drive technology manual, the second drive parameters can be derived, which may include drive parameters "B", "D" and "E". In addition, training can also be understood to include learning the relationship between the first drive parameters and the second drive parameters. For example, how the drive parameters "A" and "C" are related to the drive parameters "D" and "E".

[0014] The identification of potentially ideal or potentially optimal drive parameters is thus further improved. In particular, since the drive technical manual is used to obtain the corresponding drive parameters, the training can even be specified in more detail on a certain drive application (ie the drive application to which the drive technical manual belongs).

[0015] In addition, in a second aspect, a method for achieving a driving application goal by using an AI agent is provided. The method includes using a generative AI agent based on a large language model (LLM) obtained according to the method of the first aspect. The method also includes prompting the generative AI agent based on the LLM with a goal to be achieved for the driving application. The method also includes providing the generative AI agent based on the LLM with access to driving parameters at least partially related to the driving application and / or access to operating data at least partially related to the driving application. The method also includes providing the generative AI agent based on the LLM with access to predetermined tools and / or predetermined structured representations for analyzing driving parameters. The method also includes receiving a first output from the generative AI agent based on the LLM, wherein the first output indicates at least one solution for achieving the goal and / or at least indicates information indicating that the goal of the driving application is achieved.

[0016] It should be noted that the method can be a method for achieving the goal of a drive application by using an AI agent in an industrial drive application system. The AI ​​agent can be an AI agent in the industrial drive application system and / or an AI agent related to the industrial drive application system.

[0017] It should be noted that the expression “providing access” to driving parameters, operating data, predetermined tools and / or predetermined structured representations can be understood as the LLM-based generative AI agent being able to obtain these data, information and / or tools from a database or corresponding repository.

[0018] A predetermined tool and / or a predetermined structured representation may be understood to mean a predetermined, preselected, prepared and / or pre-adjusted tool and / or structured representation.

[0019] For example, a tool may refer to computer software or a sensor, such as computer software for determining or predicting the load on a motor due to a specific torque and / or a sensor for measuring the load.

[0020] For example, a structured representation may be understood as representing any type of written or textual information, graphical or pictorial information, or a combination thereof.

[0021] The advantages of this approach are its ability to complete tasks, generate new tasks based on the acquired results, and prioritize tasks in real time. This approach further demonstrates the potential of AI-powered language models to autonomously execute tasks within various constraints and contexts. This approach can further open up a whole new world of applications, currently primarily performed by humans with the assistance of various analytical applications. Driver analysis operations do not always require human expertise, and human expertise can therefore be leveraged for other tasks. Thus, human expertise can be more efficiently utilized as a resource. For example, human expertise can be used to design ML models and evaluate models and deploy LLM-based generative AI agents. Furthermore, human error can be reduced, optimization can be increased, and drivers and gateways can operate autonomously. New employee onboarding can be improved by providing more appropriate training. Furthermore, health monitoring, condition monitoring, and predictive analytics, to name a few, can be improved by becoming more supported, automated, and autonomous. This approach can obtain synthetic yet highly approximate driver parameters, which can be used to train additional driver ML models. This can also facilitate monitoring driver status and operating parameters using corresponding ML models from an ML model pool or repository. Furthermore, the method can reduce operating and monitoring costs, can reduce overall commissioning time, can identify optimal drive parameters in a reduced time, and / or can support and guide the customer in handling any future drive errors.

[0022] According to several examples of the present disclosure, the method may further include providing feedback to the LLM-based generative AI agent regarding the received first output. The method may further include receiving a second output from the LLM-based generative AI agent based on the feedback, wherein the second output indicates information indicating that a goal of the driving application has been achieved.

[0023] For example, feedback can be understood as representing drive feedback and / or drive operation feedback in terms of operating data and / or drive parameters. In addition, it should be noted that such feedback from the drive, drive application and / or human expert to the LLM-based generative AI agent can be a continuous learning until the output of the LLM-based generative AI agent is received via a prompt to provide a target goal. Based on monitoring the drive parameters and / or drive application operation data related to the monitored drive application, it can be known whether such a guide goal has been achieved. Therefore, the second output can be understood as representing the drive parameters and / or drive application operation data for being monitored.

[0024] Using feedback can help further improve goal achievement.

[0025] According to several examples of the present disclosure, the method may further include providing the LLM-based generative AI agent with direct access to the driving application and / or indirect access to the driving application.

[0026] Therefore, LLM-based generative AI agents can be used in conjunction with cloud applications. This further expands the variety of use cases for LLM-based generative AI agents. Direct access further supports personal use of LLM-based generative AI agents.

[0027] According to several examples of the present disclosure, prompting may include prompting an LLM-based generative AI agent as a goal to support debugging of a driver application. Additionally or alternatively, the goal of the prompt is to identify optimal driver parameters for the driver application. Additionally or alternatively, the goal of the prompt is to support diagnosis of driver errors in operation of the driver application. Additionally or alternatively, the goal of the prompt is to identify an optimal machine learning (ML) model to apply to the driver application.

[0028] Therefore, LLM-based generative AI agents can achieve several different goals and can achieve these goals more efficiently.

[0029] According to several examples of the present disclosure, prompting may include prompting an LLM-based generative AI agent in an automated manner via a human-machine interface and / or via a prompt wizard interface and / or via LLM programming.

[0030] Thus, interaction with LLM-based generative AI agents is provided in a simple way.

[0031] According to several examples of the present disclosure, providing access to a predetermined tool and / or a predetermined structured representation may include providing access to at least one of the following:

[0032] at least one predetermined ML model and / or at least one predetermined motion analysis ML model,

[0033] Gateway and / or driver,

[0034] Environmental status,

[0035] Human-machine interface,

[0036] database,

[0037] Prompt template,

[0038] Technical Manual,

[0039] Previous and / or historical driving operation data and corresponding previous and / or historical driving parameters,

[0040] values ​​provided by sensors and / or actuators associated with the drive application and / or the drive system,

[0041] Driver log,

[0042] at least one of performance, parameters and operating instructions associated with the predetermined tool and / or individual tools of the predetermined tool,

[0043] Tools for handling noisy and / or incomplete data, and

[0044] Tools from the Motion Analysis app.

[0045] Thus, LLM-based generative AI agents can use a variety of different inputs. Thus, LLM-based generative AI agents are able to identify several solutions to achieve a specific goal with higher reliability and / or higher diversity.

[0046] According to several examples of the present disclosure, providing access to a predetermined tool may include providing access to at least one of the following of a tool in the predetermined tool:

[0047] The name of the tool,

[0048] The performance of the tool,

[0049] How to access the tool,

[0050] Constraints in using tools,

[0051] The input and / or output parameters associated with the tool and their syntax, and

[0052] ML model repository, including ML models in markup language format.

[0053] Therefore, the utilization rate of the reservation tool is improved.

[0054] In addition, in a third aspect, a method for supporting the achievement of a goal of a driving application by using an artificial intelligence (AI) agent obtained according to the method of the first aspect is provided. The method includes obtaining a prompt for a goal to be achieved for the driving application. The method also includes obtaining access to driving parameters generated by the driving application. The method also includes obtaining access to a predetermined tool and / or a predetermined structured representation for analyzing the driving parameters. The method also includes accessing a tool and / or a structured representation in a predetermined tool and / or a predetermined structured representation based on the obtained prompt and the obtained access. The method also includes creating a plan for identifying at least one solution for achieving the goal based on the access. The method also includes identifying at least one solution. The method also includes providing an output indicating at least one solution and / or executing a solution in at least one solution based on the results of the identification.

[0055] It should be noted that the method may be a method for supporting the achievement of a goal of a drive application in an industrial drive application system.

[0056] The advantages of this approach are its ability to complete tasks, generate new tasks based on the acquired results, and prioritize tasks in real time. This approach further demonstrates the potential of AI-driven language models to autonomously execute tasks within various constraints and contexts. This approach can further open up a whole new world of applications currently performed by specialized analytics applications. Human expertise is not always required to drive analytics operations, and therefore human expertise can be leveraged for other tasks. Thus, human expertise can be more efficiently utilized as a resource. For example, human expertise can be used to design ML models and evaluate models and deploy LLM-based generative AI agents. Furthermore, human error can be reduced, optimization can be increased, and drivers and gateways can operate autonomously. New employee onboarding can be improved by providing more appropriate training. Furthermore, health monitoring, condition monitoring, and predictive analytics, to name a few, can be improved by becoming more supported, automated, and autonomous. This approach can obtain synthetic yet highly approximate driver parameters, which can be used to train additional driver ML models. This can help monitor optimal driver parameters using corresponding ML models from an ML model pool or repository. Furthermore, the method can reduce operating and monitoring costs, can reduce overall commissioning time, can identify optimal drive parameters in a reduced time, and / or can support and guide the customer in handling any future drive errors.

[0057] According to several examples of the present disclosure, execution may include deploying a machine learning (ML) algorithm selected by the AI ​​agent during creation and / or identification. Additionally or alternatively, execution may include deploying driver parameters selected by the AI ​​agent during creation and / or identification on a driver application and / or in debugging of the driver application. Additionally or alternatively, execution may include resolving driver errors associated with operation and / or debugging of the driver application. Additionally or alternatively, execution may include managing the driver application in an offline mode for a predetermined time.

[0058] Thus, a higher efficiency and / or a higher reliability is provided for several tasks to be performed.

[0059] According to a fourth aspect, a control device for driving an application is provided, wherein the control device is configured to execute the above-mentioned method of the first aspect, the second aspect and / or the third aspect.

[0060] The advantages of this control device are its ability to complete tasks, generate new tasks based on acquired results, and prioritize tasks in real time. The control device also demonstrates the potential of AI-driven language models to autonomously execute tasks within various constraints and contexts. This control device can further open up a whole new world of applications currently performed by specialized analytics applications. Driver analytics operations do not always require human expertise, and human expertise can therefore be leveraged for other tasks. Consequently, human expertise can be more efficiently utilized as a resource. For example, human expertise can be used to design ML models and evaluate models and deploy LLM-based generative AI agents. Furthermore, human error can be reduced, optimization can be increased, and drivers and gateways can operate autonomously. New employee onboarding can be improved by providing more appropriate training. Furthermore, health monitoring, condition monitoring, and predictive analytics, to name a few, can be improved by becoming more supported, automated, and autonomous. The control device can acquire synthetic yet approximate driver parameters, which can be used to train additional driver ML models. This can provide assistance in detecting optimal and appropriate driver parameters using corresponding ML models from an ML model pool or repository. Furthermore, the control device can reduce operating and monitoring costs, can reduce overall commissioning time, can identify optimal drive parameters in a reduced time, and / or can support and guide the customer in handling any future drive errors.

[0061] According to a fifth aspect, an automation system is provided, comprising a first control device according to the fourth aspect and a second control device according to the fourth aspect, wherein the first control device is configured to execute the method of the second aspect, the second control device is configured to execute the method of the third aspect, and wherein the first control device and the second control device are communicatively connected.

[0062] Such an automated system is advantageous because it can reduce operating and monitoring costs, reduce overall commissioning time, identify optimal drive parameters in a reduced time, and / or support and guide the customer in dealing with any future drive errors. At the very least, human error can be reduced, automation can be increased, and human expertise as a resource can be utilized more efficiently.

[0063] According to a sixth aspect, a computer-readable medium is provided, the computer-readable medium comprising instructions that, when executed by a computing system, cause the computing system to perform the method of the first, second, and / or third aspects. The computer-readable medium may be transient or non-transient, volatile or non-volatile.

[0064] Such a computer-readable medium is advantageous because it can reduce operating and monitoring costs, reduce overall commissioning time, identify optimal drive parameters within a reduced time, and / or support and guide the customer in handling any future drive errors. At a minimum, human error can be reduced, automation can be increased, and human expertise as a resource can be more efficiently utilized.

[0065] According to a seventh aspect, there is provided a computer program product comprising instructions which, when executed by a computing system, enable or cause the computing system to perform the method of the first, second and / or third aspects.

[0066] Such a computer program product is advantageous because it can reduce operating and monitoring costs, can reduce overall commissioning time, can identify optimal drive parameters in a reduced time, and / or can support and guide the customer in dealing with any future drive errors. At a minimum, human error can be reduced, automation can be increased, and human expertise as a resource can be utilized more efficiently.

[0067] According to an eighth aspect, a computing system is provided, which is configured to execute the method of the first aspect, the second aspect and / or the third aspect.

[0068] Such a computing system is advantageous because it can reduce operating and monitoring costs, reduce overall commissioning time, identify optimal drive parameters in a reduced time, and / or support and guide the customer in dealing with any future drive errors. At a minimum, human error can be reduced, automation can be increased, and human expertise as a resource can be utilized more efficiently.

[0069] The method of the first aspect, the second aspect and / or the third aspect may be implemented by a computer.

[0070] Optional features of the first, second and / or third aspects may form part of any one of the third to eighth aspects mutatis mutandis.

[0071] According to several examples, an “automation system” may also refer to an “industrial automation system”, a “drive application system” or an “industrial drive application system”, which may, for example, include and / or be associated with a ski lift system or a marine application.

[0072] As used herein, the term "acquire" may include, for example, receiving from another system, device, or process; receiving via interaction with a user; loading or retrieving from storage or memory; measuring or capturing using a sensor or other data acquisition device.

[0073] As used herein, the term "determining" encompasses a wide variety of actions and may include, for example, calculating, computing, processing, deriving, investigating, looking up (e.g., looking up in a table, database, or other data structure), ascertaining, etc. Furthermore, "determining" may include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory), etc. Furthermore, "determining" may include resolving, selecting, choosing, establishing, etc.

[0074] The indefinite article "a" or "an" does not exclude a plurality. Furthermore, as used herein, "a" and "an" should generally be construed as meaning "one or more" unless specified otherwise or clear from the context to be in the singular.

[0075] Unless otherwise specified or clear from the context, the phrases "one or more of A, B, and C," "at least one of A, B, and C," and "A, C, and / or B" as used herein are intended to refer to all possible permutations of one or more of the listed items. That is, the phrase "A and / or B" means (A), (B), or (A and B), and the phrase "A, B, and / or C" means (A), (B), (C), (A and B), (A and C), (B and C), or (A, B, and C).

[0076] The term "comprising" does not exclude other elements or steps. In addition, the terms "comprising", "including", "having" and the like can be used interchangeably in this document.

[0077] The present disclosure may include one or more aspects, examples or features, either alone or in combination, whether specifically disclosed in the combination or disclosed alone.Any optional feature or sub-aspect of one of the above aspects may be appropriately applied to any other aspect.

[0078] The foregoing aspects will become apparent and elucidated by reference to the detailed description provided hereinafter. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] A detailed description will now be given, by way of example only, with reference to the accompanying drawings in which:

[0080] Figure 1 schematically illustrates examples of use cases for applications of LLM-based generative AI agents according to several examples of the present disclosure;

[0081] Figure 2 schematically illustrates examples of use cases for applications of LLM-based generative AI agents according to several examples of the present disclosure;

[0082] Figure 3 A flowchart illustrating a method according to several examples of the present disclosure is shown;

[0083] Figure 4 A flowchart illustrating a method according to several examples of the present disclosure; and

[0084] Figure 5 Flowcharts illustrating methods according to several examples of the present disclosure are shown. DETAILED DESCRIPTION

[0085] According to several examples, data is generated by the driver. This data is then collected via various communication protocols and stored in a database, such as a cloud-based database. Data engineers and / or data scientists then use this stored data to develop various ML algorithms. Furthermore, data engineers can split the ML model into two parts, for example, where the first part is deployed and run directly on the driver, such as in a sandbox environment, and the second part is deployed and run at a gateway, such as a Next-Generation Gateway (NGGW). The driver-deployed ML model component sends the processed data to the subsequent gateway-deployed ML model component. To achieve this split, various devices, such as driver and NGGW constraints and / or driver CPU load, are used. The entire split and deployment process is performed with human assistance, such as guided workflows or wizards, and relies heavily on human expertise. For example, data generated by on-board sensors, motors, and / or drivers is collected and available as dumps, but there is still room for optimization and use. Various tool infrastructures are available, but there is significant potential to optimally integrate these different tool infrastructures with existing ML models and applications. For example, drift detection and mitigation are challenges in the operation of ML algorithms. For example, another task requiring human expertise is commissioning drive systems and identifying their potential optimal parameters. For example, the optimal parameters for the same drive can vary depending on its use case. Therefore, many factors play a crucial role in identifying the optimal parameters for a drive application. Furthermore, if a drive system error occurs, the current practice is to search for solutions online, as this is faster than reading a technical manual or contacting an expert. Therefore, identifying the optimal parameters for a drive application and / or resolving any operational errors depends heavily on human expertise.

[0086] Artificial intelligence (AI) agents that have access to previously deployed systems and are already aware of available tools (such as analytical algorithms) can support human experts in the task of debugging and monitoring applications that drive systems, or even have better historical memory than human workers. In addition, AI agents are able to combine historical events even if they have no obvious correlation.

[0087] According to the present disclosure, a generative AI agent based on LLM for sports business is provided. The execution and scheduling of existing sports data analysis solutions rely entirely on human intervention. However, during system debugging and runtime, people are not fully available 24 hours a day, nor are they available every day. Therefore, there is a need for analytical algorithms, such as those for monitoring and predictive analysis, and there is a gap in real-time monitoring of the working conditions of these analytical algorithms. It should be noted that the driver provides a large amount of data dumps, which are being analyzed by various tools and / or dashboards available in the ecosystem. However, there is still great potential in optimizing the use of these tools to gain insights, and therefore, these tools are used together with the execution and scheduling of data analysis applications to ideally obtain the best results for sports applications. Therefore, the present disclosure shows an AI-based generative AI agent that is capable of reasoning, planning and / or acting via prompts to achieve a given goal by using driver data, at least one of the various available tools and / or at least one of the various ML models.

[0088] Therefore, according to several examples of the present disclosure, a generative AI agent based on a large language model (LLM) is disclosed. The LLM-based generative AI agent further automates the splitting, optimization, and deployment of analysis and ML algorithms in resource-constrained edge and device environments or parts thereof, for example, without splitting.

[0089] Now refer to Figure 1 , Figure 1 Examples of use cases for applications of LLM-based generative AI agents according to several examples of the present disclosure are schematically illustrated.

[0090] LLM-based generative AI agents, or as described below and Figure 1 The DriveAIAgent 100 shown is capable of automating the splitting, optimization, and deployment of analytics and ML algorithms in resource-constrained edge and device environments or portions thereof, such as without splitting. Figure 1, which shows a specific use case, but the same DriveAIAgent100 can have many use cases. Therefore, there is a one-to-many use case mapping type setting. That is, one DriveAIAgent 100 can be mapped to several use cases. In addition, the work or tasks performed by one DriveAIAgent 100 can be divided into several DriveAIAgents 100. Several DriveAIAgents 100 can work or can be used in parallel. One DriveAIAgent 100 can be applied to one drive application. However, several DriveAIAgents 100 can also be applied to the same drive application. Therefore, there can also be a many-to-many use case mapping type setting.

[0091] The DriveAIAgent 100 disclosed in accordance with several examples of the present disclosure is based on an AI agent based on (1) an LLM, (2) a set of well-described analysis tools for MO drive system analysis, and (3) an agent memory.

[0092] The LLM can be thought of as part of the AI ​​agent's brain. The LLM is responsible for understanding the AI ​​agent's environment in order to generate responses and make decisions. The LLM is typically trained on large datasets of text and code.

[0093] Accessing the analytical tools or tools described in detail enables the AI ​​agent to access resources, which may also include external resources, resources such as databases, values ​​of sensors such as drives, and / or actuators and their added values. This allows the AI ​​agent to interact with the AI ​​agent's environment and take actions to achieve the AI ​​agent's goals.

[0094] An AI agent's memory can be understood as its storage device. This storage device stores the AI ​​agent's knowledge, experience, and goals. Memory can be short-term and / or long-term. Short-term memory is used to store the AI ​​agent's current state, while long-term memory is used to store the AI ​​agent's past experiences.

[0095] DriveAIAgent 100 fine-tunes the drive operation data based on the LLM, so that DriveAIAgent 100 knows how the drive works, the various parameters of the drive or drive parameters, and the relationship between these drive parameters. Through this solution, the operator 110 can prompt DriveAIAgent 100 in an automated manner using a wizard, a chat interface and / or via LLM programming, for example, prompting the operator 110 with a definition or predetermined goal that the operator 110 wants to achieve for a dedicated or predetermined drive application. Such an operator 110 does not need to be an expert in drive parameters and / or the interweaving of drive parameters, nor does the operator 110 need to know what tools are available and how to use these tools to achieve the predetermined goal. DriveAIAgent 100 takes this input (i.e., the predetermined goal) as a prompt from the operator 110 and creates its several tasks based on DriveAIAgent 100 reasoning and planning performance.

[0096] Based on these tasks, DriveAIAgent 100 may need to access various tools that can help DriveAIAgent 100 achieve these tasks. Figure 1 The DriveAIAgent 100 shown will traverse the tag description file, such as Figure 1 The markup description and wrapper 120 shown in FIG, which contains a list of available tools, including access to available ML models stored in the ML model repository 130, such as the performance, parameters or driving parameters 140 of these tools and operating instructions. Based on this file, the DriveAIAgent 100 can identify suitable or required tools for achieving the predetermined goal. However, it should be noted that the DriveAIAgent 100 does not necessarily have to traverse the list of tools such as Figure 1 The markup description file or markup description and wrapper 120 shown may instead be used with any means that enables the DriveAIAgent 100 to obtain such information from the markup description and wrapper 120.

[0097] Then, if the DriveAIAgent 100 knows the appropriate or required tools, the DriveAIAgent 100 will create a plan. After the plan is ready, the DriveAIAgent 100 can find one or several optimized solutions, as the DriveAIAgent 100 can propose several solutions for achieving the predetermined goal and provide them as output to the operator 110, for example in natural language.

[0098] The operator 110 can review the output and decide whether to agree with the provided solution(s). After the operator 110 agrees, the DriveAIAgent 100 can execute the agreed solution accordingly. When the predetermined goal is achieved, the DriveAIAgent 100 will notify the operator 110.

[0099] In more detail, DriveAIAgent 100 is a generative AI agent based on LLM. Existing LLM is used and fine-tuned based on drive operation data and drive parameters. Therefore, DriveAIAgent 100 can become an expert in understanding real-time drive parameters. Since DriveAIAgent 100 can monitor driving in real time, it may be necessary to provide DriveAIAgent 100 with access to generated drive parameters, including historical drive parameters and / or previously generated drive parameters. DriveAIAgent 100 may also need to access additional tools that can provide value-added insights from the generated drive parameters. Access to such tools can be provided in such a manner, for example, a markup language that describes at least one of the following from the tool:

[0100] The name of the tool;

[0101] Tool performance;

[0102] How to access the tool;

[0103] Restrictions on using tools;

[0104] Input and / or output parameters and their syntax; and

[0105] • ML model repository 130 so that the DriveAIAgent 100 along with the tools can also access a repository containing ML models, optionally in the same markup language format, that has information from several tools.

[0106] According to several examples of this disclosure, the following will be considered for deploying ML algorithms. That is, DriveAIAgent 100 has the necessary elements to analyze and autonomously split, optimize, and deploy ML algorithms for both edge and drive. Now, DriveAIAgent 100 waits for the operator 110 to input a prompt, which DriveAIAgent 100 uses as a target, and the operator 110 wishes to implement this prompt through a specific edge drive setting application.

[0107] For example, reference Figure 1, the operator 110 can prompt the DriveAIAgent 100 to deploy anomaly detection on the drive system, for example, for the motor. The DriveAIAgent 100 can then obtain relevant information and tools to achieve this goal. For example, the DriveAIAgent 100 obtains an ML model suitable for anomaly detection at the motor from the ML model repository 130. In addition, the DriveAIAgent 100 obtains a tool that is suitable for preprocessing the operating data of the drive system, for example, because such preprocessed data helps with anomaly detection. The preprocessed data can be a preprocessed torque value. In addition, the DriveAIAgent 100 obtains relevant drive parameters, such as the drive parameters indicated in the obtained ML model or the obtained tool. The DriveAIAgent 100 can also determine which drive parameters to obtain based on the history and / or current condition of the monitored motor. Then, when the DriveAIAgent 100 has "collected" all the information and tools required for deployment, the DriveAIAgent 100 can partition the ML algorithm for anomaly detection on the drive system into several (for example, three) deployable modules. At least one of these modules (for example, Figure 1 Module 1) can be forwarded to the driver, for example in or into a sandbox environment. For example, forwarding can be done via OPC UA. The remaining modules (e.g. Figure 1 Modules 2 and 3 in the diagram are still located at the edge, enabling monitoring from the edge to the drive. Figure 1 At the elements "monitoring", "2", "3" and "communication middleware" on the edge, an explanatory element for indicating "docker (container)" is provided.

[0108] According to several examples of the present disclosure, reference Figure 2 , the following is an example use case for identifying optimal drive parameters. That is, as before, DriveAIAgent 100 can access the required tools 200, such as the environmental status 210, previous operation data with its drive parameters 220, tools for processing noisy and incomplete data 230, and drive technical manuals 240. Therefore, based on these tools 200, DriveAIAgent 100 can support debugging and can identify optimal drive parameters. In addition, in the event of any drive error during operation, based on the drive log and various data associated with the drive log (which can be analyzed by ML algorithms), DriveAIAgent 100 can support the operator 110 in diagnosis.

[0109] As a reference Figure 2For example, the operator 250 prompts the DriveAIAgent 100 to identify optimal drive parameters for a drive system that is a ski lift system, where the ski lift system has a specific architecture. Figure 1 Similarly, DriveAIAgent 100 will “collect” all the information and tools needed to achieve this goal, and based on this “collection”, output the optimal driving parameter configuration and the corresponding optimal ML model, such as Figure 2 shown.

[0110] It should be noted that, in other words, according to one of several examples, a drive parameter can be understood as representing a constraint and / or limitation, such as a speed range, such as 50 to 100 rpm of the motor device. If the motor device is to be operated, the motor device will be controlled to have a speed within the given speed range, such as 70 rpm. This 70 rpm can be understood as representing operational data. Operational data can be understood as data representing the actual operation of the application.

[0111] It should also be noted that, for example, for commissioning, hundreds of drive parameters may need to be set and / or adjusted.

[0112] According to several examples of embodiments, there are several requirements for DriveAIAgent 100 to be able to work. For example, if not all of them are available, DriveAIAgent 100 will need and most efficiently use at least one or a subset of the following:

[0113] · For fine-tuning LLMs that drive operational data,

[0114] Various tools from sports analysis applications, such as classifier models, anomaly detection models, etc.

[0115] For example,

[0116] · Access to technical manuals,

[0117] Access to previous and / or historical drive parameters based on its usage and environmental status,

[0118] Access to a repository of motion analysis ML models,

[0119] A markup language syntax through which tools and ML models can be expressed,

[0120] · Operator prompt wizard,

[0121] Various prompt templates that should be selected based on operator input goals, and

[0122] Access gateways and drivers.

[0123] In the following, according to several examples of the present disclosure, a use case is presented, which includes automatic splitting and deployment of ML algorithms for industrial projects.

[0124] In the case of industrial projects, currently the project can be completely dependent on the data engineer. Therefore, the success of the analytical application of the project developed by the data engineer depends entirely on the data engineer's knowledge and deep understanding of the driving operations, the driving parameters that drive the operations, and the relationship between these driving parameters. In addition, the data engineer also needs other tools to understand this data, such as dashboards. Using DriveAIAgent 100, the required information can be fed into the LLM. The LLM, which has storage and access to the relationships between this required information, data, and supporting tools, can be better optimized to identify the final analytical operation, such as an ML algorithm, which should be deployed on the drive, on the edge, or both, and should then be deployed.

[0125] In the following, a use case is presented that includes drive parameterization according to several examples of the present disclosure.

[0126] Currently, commissioning requires a drive expert. Even then, the entire commissioning process is time-consuming. Furthermore, there is room for improvement in identifying drive parameters. DriveAIAgent 100 will have access to tools and information such as previous and / or historical operating data including relevant drive parameters, tools for handling noisy and incomplete data, and drive technical manuals. With this, DriveAIAgent 100 can identify potentially optimal drive parameters and reduce overall commissioning time. DriveAIAgent 100 can also support operators in resolving any drive errors.

[0127] To support debugging and identifying optimal drive parameters, DriveAIAgent 100 can be fine-tuned using drive technical manuals. LLM can be used to understand and process large amounts of structured representations, such as technical manuals, texts, reports, and / or research papers. Research papers can be relevant for developing new drive parameter identification methods. This data can be used to understand the relationships between different drive parameters and develop new methods for identifying drive parameters.

[0128] Drive parameters are often nonlinear and time-varying, making them difficult to measure or estimate using traditional methods. These drive parameters are influenced by various factors, including the drive's load, environment, and / or aging. Furthermore, due to noisy and incomplete data, there are still gaps in identifying optimized drive parameters based on the available data for training.

[0129] However, with the in-depth knowledge of the drive operation and drive parameters by the DriveAIAgent 100, the data engineer can be relieved of this burden. Therefore, operating the DriveAIAgent 100 does not require experienced experts or data engineers, but the operator 110 can already handle the DriveAIAgent 100. Figure 1 As shown, data engineers 150 can focus on developing various ML models that can be deployed to the ML model repository 130, and the ML model repository 130 serves as a markup language. Figure 1 As shown, after the DriveAIAgent 100 is fine-tuned with the driving-related data, the operator 110 can begin to give prompts to identify driving parameters. When it is necessary to provide the best driving parameters and the appropriate ML model, the DriveAIAgent 100 can call on various tools as described above.

[0130] According to several examples of the present disclosure, at least one of the following advantageous features is shown:

[0131] The basic or fully autonomous implementation of a distributed on-premise analytics ecosystem as a core enabler for analytics scenarios based on it, including anomaly detection, asset health monitoring, and more;

[0132] Autonomous real-time deployment of ML models to gateways and / or drivers;

[0133] The entire system consists of the following components and functions:

[0134] ML model repository: A data analysis engineer 150 who is an expert in the field of sports can create ML models and deploy these ML models to the ML model repository 130 so that the DriveAIAgent 100 can deploy one or more of these ML models when needed.

[0135] Database: Predetermined and / or historical data available from the database can help the DriveAIAgent 100 understand the previous decisions and previous performance of the motion application.

[0136] Tool descriptions, such as markup descriptions of tools: Using tools and algorithms described via a markup language (e.g., controlled and / or natural markup language), the DriveAIAgent 100 will understand the tools, their capabilities, and how to use them when needed.

[0137] Human-machine interfaces, such as dashboards and / or alternative solutions, for example: such interfaces provide value-added results to the DriveAIAgent 100 and may represent one possible interpretation interface of the DriveAIAgent system.

[0138] QnA interface for prompting and / or controlling the human-machine interface: for the operator 110 to give the goals that the operator 110 wants to achieve from the motion application controlled by the DriveAIAgent.

[0139] Motion application based prompt templates: Based on operator input, certain prompt templates can be used so that the DriveAIAgent 100 can gain better understanding and can perform structured planning.

[0140] Access to tools: DriveAIAgent 100 can be aware of Drive-GW constraints and actual loads, which DriveAIAgent 100 can consider when deploying ML on both the gateway and the drive; and

[0141] Also suitable for non-split analytics and / or ML solutions.

[0142] According to several examples of the present disclosure, at least one of the following use cases may be considered:

[0143] “Hey AI agent! — Please do whatever it takes to monitor this drive and keep it running smoothly!” Therefore, the DriveAIAgent 100 should be powered by a set of algorithms and / or ML models.

[0144] The subset can be selected and run completely autonomously in sequence;

[0145] “Hey AI agent! — Please do whatever you can to save energy on this driver (system)!”

[0146] Therefore, the DriveAIAgent 100 should investigate which parameters can be adjusted to optimize a certain KPI, and can then do so accordingly;

[0147] In the event that the drive and its operating location cannot be reached, the DriveAIAgent 100 will

[0148] Often useful, continuous manual monitoring can be performed for days to months;

[0149] If the drive and device are not online, such as on a ship or remote for a period of time (e.g., several months), it can be prompted to handle drive operations for three months, for example, so DriveAIAgent 100

[0150] This situation can be handled in very different ways. These different ways or action strategies

[0151] It can be designed by humans;

[0152] Therefore, regarding how the DriveAIAgent 100 should behave in a normal (eg, non-isolated)

[0153] Several policies may be executable. For months of remote operation, the policy should be different from that for normal operation.

[0154] strategy for making;

[0155] DriveAIAgent 100 can be used as an agent pool or agent committee where multiple

[0156] AIAgent and participate in decision-making and action;

[0157] DriveAIAgent 100 can be used as a commissioning co-pilot to assist commissioning engineers; and

[0158] DriveAIAgent 100 can be used to identify optimal driving parameters.

[0159] Hereinafter, further examples according to the present disclosure are shown.

[0160] Now refer to Figure 3 , showing a flowchart of a method for obtaining an artificial intelligence AI agent suitable for driving an application according to several examples of the present disclosure.

[0161] The method starts in S300. The method is used to obtain an artificial intelligence AI agent suitable for driving an application, that is, the artificial intelligence AI agent described above. Figure 1 and Figure 2 Overview of DriveAIAgent 100.

[0162] The method includes, in S310 , training data indicating predetermined driving operation data associated with a predetermined driving application and / or predetermined driving parameters associated with the predetermined driving application.

[0163] Regarding training data, it should be noted that the training data can be obtained from actual driver applications and / or can be synthetically developed. For example, the training data can indicate historical operations of the driver application, where the corresponding historical operation data is obtained from a database or historical database. Additionally or alternatively, the synthetically developed data can represent data generated by another AI, tool, and / or human. For example, the synthetically developed data can indicate simulated data based on which the operation of the driver application can be simulated.

[0164] The method also includes, in S320, training a generative AI agent based on a large language model LLM using the acquired training data to identify at least one of the following: a first driving parameter at least partially related to the driving application, and a relationship between the first driving parameters. Training can be understood to include training, fine-tuning, and a combination of training and fine-tuning. For example, LLM training or fine-tuning specific to industrial automation or more precisely for industrial drive application use cases. Training can be understood to mean a continuous learning, that is, the generative AI agent based on the LLM is continuously improved. The training used throughout this disclosure can include fine-tuning of an already existing pre-trained ML model. In addition, training can be understood as updating weights.

[0165] The method ends in S330.

[0166] It should be noted that this LLM-based generative AI agent can be represented as in the above reference Figure 1 and Figure 2 The DriveAIAgent 100 described. In addition, it should be noted that the method may include training several LLM-based generative AI agents or several DriveAIAgents 100, which can process tasks in parallel and / or subsequently. The method can also be applicable to multi-agent scenarios. In addition, the LLM-based generative AI agent (i.e., DriveAIAgent 100) can have memory access, planning and reasoning performance, and can take actions accordingly. In addition, the drive application to which the LLM-based generative AI agent is applicable can be an industrial drive application.

[0167] according to Figure 3An AI agent that can be acquired using this method is advantageous because it can complete tasks, generate new tasks based on the acquired results, and prioritize tasks in real time. This AI agent further demonstrates the potential of AI-driven language models to autonomously perform tasks within various constraints and contexts. This AI agent can further open up a whole new world of applications currently performed by specialized analytics applications. Human expertise is not always required to drive analytical operations, and therefore human expertise can be leveraged for other tasks. Thus, human expertise can be more efficiently utilized as a resource. For example, human expertise can be used to design ML models and evaluate models and deploy LLM-based generative AI agents. Furthermore, human error can be reduced, optimization can be increased, and drivers and gateways can operate autonomously. New employee onboarding can be improved by providing more appropriate training. Furthermore, health monitoring, condition monitoring, and predictive analytics, to name a few, can be improved by becoming more supported, automated, and autonomous. This AI agent can acquire synthetic yet highly approximate driver parameters, which can be used to train additional driver ML models. This can help monitor optimal driver parameters using corresponding ML models from an ML model pool or repository. Furthermore, such AI agents can reduce operating and monitoring costs, can reduce overall commissioning time, can identify optimal drive parameters in a reduced time, and / or can support and guide the customer in handling any future drive errors.

[0168] According to several examples of the present disclosure, the acquired training data may further indicate a driving technical manual, the driving technical manual indicating a second driving parameter related to the driving application. The training may further include training to learn at least one of the following: a first relationship indicating a relationship between at least a portion of the first driving parameters based on the driving technical manual, and a second relationship indicating a relationship between at least a portion of the second driving parameters based on the driving technical manual. The training may further include training to develop a method for identifying a driving parameter at least partially related to the driving application based on at least one of the first relationship and the second relationship.

[0169] It should be noted that the second drive parameters can be understood to represent a set of drive parameters. The second drive parameters can be different from the first drive parameters, can have a certain overlap of at least one identical drive parameter with the first drive parameters; or can be the same as the first drive parameters. For example, for the purpose of explanation, it can be assumed here that the first drive parameters include drive parameters "A", "B" and "C". In addition, from the drive technology manual, the second drive parameters can be derived, which may include drive parameters "B", "D" and "E". In addition, training can also be understood to include learning the relationship between the first drive parameters and the second drive parameters. For example, how drive parameters "A" and "C" are related to drive parameters "D" and "E".

[0170] It should be noted that the drive parameters can be used to configure the drive for one or more specific applications. Thus, the "Configure Drive" task bridges the gap between identifying the correct drive parameters and their application to industrial applications using electric motors.

[0171] Thus, the identification of potentially ideal or potentially optimal driving parameters is further improved.

[0172] Now refer to Figure 4 , showing a flowchart of a method for achieving driving application goals by using an artificial intelligence (AI) agent according to several examples of the present disclosure.

[0173] The method starts in S400 .

[0174] In S410, the method includes using Figure 3 A generative AI agent based on a large language model (LLM) obtained by this method.

[0175] The method further includes, in S420 , prompting the LLM-based generative AI agent with goals to be achieved in order to drive the application.

[0176] The method further includes, in S430 , providing the LLM-based generative AI agent with access to driving parameters at least partially related to the driving application and / or access to operating data at least partially related to the driving application.

[0177] The method further includes, in S440 , providing the LLM-based generative AI agent with access to a predetermined tool and / or a predetermined structured representation for analyzing the driving parameters.

[0178] The method further includes, at S450, receiving a first output from the LLM-based generative AI agent, wherein the first output indicates at least one solution for achieving the goal and / or at least indicates information indicating that the goal of the driving application is achieved. For example, the output can be natural language, json, or python code.

[0179] The method ends in S460.

[0180] It should be noted that this LLM-based generative AI agent can be represented as in the above reference Figure 1 and Figure 2 The DriveAIAgent 100.

[0181] The advantages of this approach are its ability to complete tasks, generate new tasks based on the acquired results, and prioritize tasks in real time. This approach further demonstrates the potential of AI-driven language models to autonomously execute tasks within various constraints and contexts. This approach can further open up a whole new world of applications currently performed by specialized analytics applications. Human expertise is not always required to drive analytics operations, and therefore human expertise can be leveraged for other tasks. Consequently, human expertise can be more efficiently utilized as a resource. For example, human expertise can be used to design ML models and evaluate models and deploy LLM-based generative AI agents. Furthermore, human error can be reduced, optimization can be increased, and drivers and gateways can operate autonomously. New employee onboarding can be improved by providing more appropriate training. Furthermore, health monitoring, condition monitoring, and predictive analytics, to name a few, can be improved by becoming more supported, automated, and autonomous. This approach can obtain synthetic yet highly approximate driver parameters, which can be used to train additional driver ML models. This can help monitor optimal driver parameters using corresponding ML models from an ML model pool or repository. Furthermore, the method can reduce operating and monitoring costs, can reduce overall commissioning time, can identify optimal drive parameters in a reduced time, and / or can support and guide the customer in handling any future drive errors.

[0182] According to several examples of the present disclosure, the method may further include providing feedback to the LLM-based generative AI agent regarding the received first output. The method may further include receiving a second output from the LLM-based generative AI agent based on the feedback, wherein the second output indicates information indicating that a goal of the driving application has been achieved.

[0183] Using feedback can help further improve goal achievement.

[0184] According to several examples of the present disclosure, the method may further include providing the LLM-based generative AI agent with direct access to the drive application and / or indirect access to the drive application. For example, indirect access may mean via software running on a cloud server or cloud application that obtains the output of the LLM-based generative AI agent, i.e., DriveAIAgent 100, and then uses these outputs for the gateway and / or the drive.

[0185] Therefore, LLM-based generative AI agents can be used in conjunction with cloud applications. This further expands the variety of use cases for LLM-based generative AI agents. Direct access further supports personal use of LLM-based generative AI agents.

[0186] According to several examples of the present disclosure, prompting may include prompting an LLM-based generative AI agent as a goal to support debugging of a driver application. Additionally or alternatively, the goal of the prompt is to identify optimal driver parameters for the driver application. Additionally or alternatively, the goal of the prompt is to support diagnosis of driver errors in operation of the driver application. Additionally or alternatively, the goal of the prompt is to identify an optimal machine learning (ML) model to apply to the driver application.

[0187] Therefore, LLM-based generative AI agents can achieve several different goals and can achieve these goals more efficiently.

[0188] According to several examples of the present disclosure, prompting may include prompting an LLM-based generative AI agent via a human-machine interface and / or via a prompt wizard interface and / or via LLM programming, e.g., in an automated manner.

[0189] Thus, interaction with LLM-based generative AI agents is provided in a simple way.

[0190] According to several examples of the present disclosure, providing access to a predetermined tool and / or a predetermined structured representation may include providing access to at least one of the following:

[0191] at least one predetermined ML model and / or at least one predetermined motion analysis ML model,

[0192] Gateway and / or driver,

[0193] Environmental status,

[0194] Human-machine interface,

[0195] database,

[0196] Prompt Template

[0197] Technical Manual,

[0198] Previous and / or historical driving operation data and corresponding previous and / or historical driving parameters,

[0199] values ​​provided by sensors and / or actuators associated with the drive application and / or the drive system,

[0200] Driver log,

[0201] at least one of performance, parameters and operating instructions associated with the predetermined tool and / or individual tools of the predetermined tool,

[0202] Tools for handling noisy and / or incomplete data, and

[0203] Tools from the Motion Analysis app.

[0204] Thus, LLM-based generative AI agents can use a variety of different inputs. Thus, LLM-based generative AI agents are able to identify several solutions to achieve a specific goal with higher reliability and / or higher diversity.

[0205] According to several examples of the present disclosure, providing access to a predetermined tool may include providing access to at least one of the following of a tool in the predetermined tool:

[0206] The name of the tool,

[0207] The performance of the tool,

[0208] How to access the tool,

[0209] Constraints in using tools,

[0210] The input and / or output parameters associated with the tool and their syntax, and

[0211] ML model repository, including ML models in markup language format.

[0212] Therefore, the utilization rate of the reservation tool is improved.

[0213] Now refer to Figure 5 , which shows a flow chart of a method for supporting the use of Figure 3 The artificial intelligence (AI) agent obtained by this method can achieve the goal of driving the application.

[0214] The method starts in S500 .

[0215] The method includes, in S510 , obtaining a hint of a goal to be achieved for driving an application.

[0216] The method further comprises, in S520 , obtaining access to driving parameters generated by the driving application. The driving parameters may be driving parameters of the driving application saved in the past, ie, the driving parameters may be understood as previous and / or historical driving parameters.

[0217] The method further includes, at S530 , obtaining access to a predetermined tool and / or a predetermined structured representation for analyzing the driving parameters.

[0218] The method further comprises, in S540 , accessing a tool and / or a structured representation in a predetermined tool and / or a predetermined structured representation based on the acquired prompt and based on the acquired access.

[0219] The method further includes, in S550 , creating a plan for identifying at least one solution for achieving the goal based on the visit.

[0220] The method further includes, in S560 , identifying at least one solution.

[0221] The method further includes, in S570 , providing an output indicating at least one solution and / or executing a solution in the at least one solution based on the identified result.

[0222] The method ends in S580.

[0223] It should be noted that this AI agent can represent the above reference Figure 1 and Figure 2 Overview of DriveAIAgent 100.

[0224] The advantages of this approach are its ability to complete tasks, generate new tasks based on the acquired results, and prioritize tasks in real time. This approach further demonstrates the potential of AI-driven language models to autonomously execute tasks within various constraints and contexts. This approach can further open up a whole new world of applications currently performed by specialized analytics applications. Human expertise is not always required to drive analytics operations, and therefore human expertise can be leveraged for other tasks. Consequently, human expertise can be more efficiently utilized as a resource. For example, human expertise can be used to design ML models and evaluate models and deploy LLM-based generative AI agents. Furthermore, human error can be reduced, optimization can be increased, and drivers and gateways can operate autonomously. New employee onboarding can be improved by providing more appropriate training. Furthermore, health monitoring, condition monitoring, and predictive analytics, to name a few, can be improved by becoming more supported, automated, and autonomous. This approach can obtain synthetic yet highly approximate driver parameters, which can be used to train additional driver ML models. This can help monitor optimal driver parameters using corresponding ML models from an ML model pool or repository. Furthermore, the method can reduce operating and monitoring costs, can reduce overall commissioning time, can identify optimal drive parameters in a reduced time, and / or can support and guide the customer in handling any future drive errors.

[0225] According to several examples of the present disclosure, execution may include deploying a machine learning (ML) algorithm selected by the AI ​​agent during creation and / or identification. Additionally or alternatively, execution may include deploying driver parameters selected by the AI ​​agent during creation and / or identification on a driver application and / or in debugging of the driver application. Additionally or alternatively, execution may include resolving driver errors associated with operation and / or debugging of the driver application. Additionally or alternatively, execution may include managing the driver application in an offline mode for a predetermined time.

[0226] Thus, a higher efficiency and / or a higher reliability is provided for several tasks to be performed.

[0227] According to several examples of the present disclosure, a control device for driving an application is provided, wherein the control device is configured to execute Figure 1 、 Figure 2 or Figure 3 Any of the above methods.

[0228] In more detail, according to various examples, a control device for driving an application is disclosed, wherein the control device is configured to perform the above-mentioned Figures 3 to 5 For example, the control device may include a processor and a memory for storing instructions, which instructions, when executed by the processor, may cause the control device to perform, for example, the above-mentioned reference Figures 3 to 5 For this execution, the control device may comprise several functional parts, for example, for executing the Figure 3 The acquisition portion of the process of step 310, and the method for executing the Figure 3 Additionally and / or alternatively, the control device may comprise several functional parts, for example, for executing the training part according to step 320. Figure 4 The use portion of the process of step 410 is used to perform the Figure 4 The prompt portion of the process of step 420 is used to perform the Figure 4 The providing portion of the process of step 430 for performing the Figure 4 The process of step 440 provides a portion of the process, and the method for performing the process according to Figure 4 Additionally and / or alternatively, the control device may comprise several functional parts, for example, for executing the Figure 5 The acquisition portion of the process of step 510 is used to perform the Figure 5 The acquisition portion of the process of step 520 is used to perform the Figure 5 The acquisition portion of the process of step 530 is used to perform the Figure 5 The access portion of the process of step 540 is used to perform the Figure 5The creation portion of the process of step 550 is used to perform the Figure 5 The identification portion of the process of step 560, and the method for executing the Figure 5 The process of step 570 provides a portion. In addition, such a portion can be understood to mean a component for performing a specific function or a portion configured to perform a specific function.

[0229] The advantages of this control device are its ability to complete tasks, generate new tasks based on acquired results, and prioritize tasks in real time. The control device also demonstrates the potential of AI-driven language models to autonomously execute tasks within various constraints and contexts. This control device can further open up a whole new world of applications currently performed by specialized analytics applications. Driver analytics operations do not always require human expertise, and human expertise can therefore be leveraged for other tasks. Consequently, human expertise can be more efficiently utilized as a resource. For example, human expertise can be used to design ML models and evaluate models and deploy LLM-based generative AI agents. Furthermore, human error can be reduced, optimization can be increased, and drivers and gateways can operate autonomously. New employee onboarding can be improved by providing more appropriate training. Furthermore, health monitoring, condition monitoring, and predictive analytics, to name a few, can be improved by becoming more supported, automated, and autonomous. The control device can acquire synthetic yet approximate driver parameters, which can be used to train additional driver ML models. This can provide assistance in detecting optimal and appropriate driver parameters using corresponding ML models from an ML model pool or repository. Furthermore, the control device can reduce operating and monitoring costs, can reduce overall commissioning time, can identify optimal drive parameters in a reduced time, and / or can support and guide the customer in handling any future drive errors.

[0230] According to several examples of the present disclosure, there is provided an (industrial) automation system (or (industrial) drive application system), the system comprising the above-mentioned control device as a first control device and the above-mentioned control device as a second control device, the first control device being configured to execute according to Figure 4 The second control device is configured to execute the method according to Figure 5 The method further comprises: providing a first control device and a second control device in a communicative connection.

[0231] Such an automated system is advantageous because it can reduce operating and monitoring costs, reduce overall commissioning time, identify optimal drive parameters in a reduced time, and / or support and guide the customer in dealing with any future drive errors. At the very least, human error can be reduced, automation can be increased, and human expertise as a resource can be utilized more efficiently.

[0232] According to several examples of the present disclosure, a computer-readable medium is provided, the computer-readable medium including instructions that, when executed by a computing system, cause the computing system to execute a program according to Figure 1 、 Figure 2 and Figure 3 The computer readable medium may be transitory or non-transitory, volatile or non-volatile.

[0233] Such a computer-readable medium is advantageous because it can reduce operating and monitoring costs, reduce overall commissioning time, identify optimal drive parameters within a reduced time, and / or support and guide the customer in handling any future drive errors. At a minimum, human error can be reduced, automation can be increased, and human expertise as a resource can be more efficiently utilized.

[0234] According to several examples of the present disclosure, a computer program product is provided that includes instructions that, when executed by a computing system, enable or cause the computing system to perform a process according to Figure 1 、 Figure 2 and Figure 3 Any of the methods.

[0235] Such a computer program product is advantageous because it can reduce operating and monitoring costs, can reduce overall commissioning time, can identify optimal drive parameters in a reduced time, and / or can support and guide the customer in dealing with any future drive errors. At a minimum, human error can be reduced, automation can be increased, and human expertise as a resource can be utilized more efficiently.

[0236] According to several examples of the present disclosure, a computing system is provided, which is configured to execute Figure 1 、 Figure 2 and Figure 3 Any of the methods.

[0237] Such a computing system is advantageous because it can reduce operating and monitoring costs, reduce overall commissioning time, identify optimal drive parameters in a reduced time, and / or support and guide the customer in dealing with any future drive errors. At a minimum, human error can be reduced, automation can be increased, and human expertise as a resource can be utilized more efficiently.

[0238] References Figure 3 、 Figure 4 and Figure 5 Any of the methods outlined in can be implemented by a computer.

[0239] Any unit, module, circuit system or method described herein can be implemented using hardware, software and / or firmware configured to perform any operation described herein. Hardware may include one or more processor cores, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), etc. Software may be embodied as software packages, codes, instructions, instruction sets and / or data recorded on at least one transient or non-transient computer-readable storage medium. Firmware may be embodied as codes, instructions or instruction sets and / or data hard-coded in a memory device (e.g., a non-volatile memory device).

[0240] If implemented in software, these functions can be stored as one or more instructions or codes on a computer-readable medium or transmitted via a computer-readable medium. Computer-readable media include computer-readable storage media. Computer-readable storage media can be any available storage medium that a computer can access. As an example and not limitation, such computer-readable storage media can include flash memory storage media, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer. As used herein, disks and optical disks include optical disks (CDs), laser disks, optical disks, digital versatile disks (DVDs), floppy disks, and Blu-ray disks (BDs), wherein disks typically reproduce data magnetically, while optical disks typically reproduce data optically with lasers. In addition, propagated signals can be included within the scope of computer-readable storage media. Computer-readable media also include communication media, including any media that facilitates the transmission of a computer program from one place to another. For example, a connection can be a communication medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies (such as infrared, radio, and microwave), then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies (such as infrared, radio, and microwave) are also included in the definition of communications media. Combinations of the above should also be included within the scope of computer-readable media.

[0241] Applicants hereby disclose individually each individual feature described herein, as well as any combination of two or more such features, to the extent such feature or combination can be implemented according to the common general knowledge of a person skilled in the art based on the specification as a whole, regardless of whether such feature or combination of features solves any problem disclosed herein, and without limiting the scope of the claims. Applicants indicate that aspects of the present disclosure may consist of any such individual feature or combination of features.

[0242] It should be noted that the embodiments of the present disclosure are described with reference to different categories. In particular, some examples are described with reference to methods, while other examples are described with reference to devices. However, those skilled in the art will understand from the description that, unless otherwise indicated, any combination of features belonging to one category, in addition to any combination of features belonging to different categories, is also considered to be disclosed by this application. However, all features can be combined to provide synergistic effects, rather than simply adding the features together.

[0243] Although the present disclosure has been shown and described in detail in the drawings and the foregoing description, such illustration and description should be considered illustrative rather than restrictive. The present disclosure is not limited to the disclosed embodiments. Other variations to the disclosed embodiments may be understood and implemented by those skilled in the art through a study of the drawings, the present disclosure, and the appended claims.

[0244] The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

[0245] Any reference signs in the claims should not be construed as limiting the scope.

[0246] Reference numerals:

[0247] 100: DriveAIAgent (Drive AI Agent); 110: Operator; 120: Tag description and wrapper; 130: ML model repository; 140: Drive parameters; 150: Data engineer; 200: Tools; 210: Environment state; 220: Previous operation data with its drive parameters; 230: Tools for handling noisy and incomplete data; 240: Drive technical manual.

Claims

1. A method for obtaining an artificial intelligence (AI) agent (100) suitable for driving an application, the method comprising: Acquiring ( S310 ) training data, the training data indicating predetermined driving operation data associated with a predetermined driving application and / or indicating predetermined driving parameters associated with the predetermined driving application; The obtained training data is used to train (S320) a generative AI agent based on a large language model (LLM) to identify at least one of the following: a first driving parameter at least partially related to the driving application, and The relationship between the first driving parameters.

2. The method according to claim 1, The training data obtained includes historical operation data and / or simulated operation data; and / or wherein the acquired training data further indicates a driving technical manual, wherein the driving technical manual indicates a second driving parameter related to the driving application; wherein the training further comprises training to learn at least one of: a first relationship indicating a relationship between at least a portion of the first driving parameters based on the driving technique manual, and a second relationship indicating a relationship between at least a portion of the second driving parameters based on the driving technique manual; and The training further comprises training to develop a method for identifying a driving parameter at least partially related to the driving application based on at least one of the first relationship and the second relationship.

3. A method for achieving a goal of driving an application by using an artificial intelligence (AI) agent, the method comprising: Using (S410) a generative AI agent (100) based on a large language model (LLM) obtained by the method according to claim 1 or 2; Prompting (S420) the LLM-based generative AI agent (100) with the goal to be achieved for the driving application; providing (S430) the LLM-based generative AI agent (100) with access to driving parameters at least partially related to the driving application and / or access to operating data at least partially related to the driving application; providing (S440) the LLM-based generative AI agent (100) with access to a predetermined tool and / or a predetermined structured representation for analyzing the driving parameters; as well as A first output is received (S450) from the LLM-based generative AI agent (100), wherein the first output indicates at least one solution for achieving the goal and / or at least indicates information indicating that the goal of the driving application is achieved.

4. The method according to claim 3, further comprising: providing feedback to the LLM-based generative AI agent (100) regarding the received first output; as well as Based on the feedback, a second output is received from the LLM-based generative AI agent (100), wherein the second output indicates information indicating that the goal of the driving application is achieved.

5. The method according to claim 3 or 4, further comprising: The LLM-based generative AI agent (100) is provided with direct access to the driving application and / or indirect access to the driving application.

6. The method according to any one of claims 3 to 5, wherein the prompting comprises prompting the LLM-based generative AI agent (100) with the goal as follows: Supporting debugging of the driver application; and / or identifying optimal drive parameters for the drive application; and / or supporting diagnosis of driver errors in operation of the driver application; and / or Identify the best machine learning (ML) model to apply to the driver application.

7. The method according to any one of claims 3 to 6, wherein the prompting comprises prompting the LLM-based generative AI agent (100) in an automatic manner via a human-machine interface and / or via a prompt wizard interface and / or via LLM programming.

8. The method according to any one of claims 3 to 7, wherein providing access to the predetermined tool and / or the predetermined structured representation comprises providing access to at least one of: at least one predetermined ML model and / or at least one predetermined motion analysis ML model, Gateway and / or driver, Environmental status, Human-machine interface, database, Prompt Template Technical Manual, Previous and / or historical driving operation data and corresponding previous and / or historical driving parameters, values ​​provided by sensors and / or actuators associated with the drive application and / or drive system, Driver log, at least one of performance, parameters and operating instructions associated with the predetermined tool and / or individual tools of the predetermined tool, Tools for handling noisy and / or incomplete data, and Tools from the Motion Analysis app.

9. The method of any one of claims 3 to 8, wherein providing said access to said predetermined tool comprises: The access is provided to describe at least one of the following of a tool in the predetermined tool: The name of the tool, The performance of the tool, How to access said tool, Constraints in using the tools described, The input and / or output parameters associated with the tool and their syntax, and ML model repository, including ML models in markup language format.

10. A method for supporting an artificial intelligence (AI) agent (100) obtained by using the method according to claim 1 or 2 to achieve a goal of driving an application, the method comprising: Obtaining ( S510 ) a prompt for the goal to be achieved for the driving application; obtaining ( S520 ) access to driving parameters generated by the driving application; obtaining ( S530 ) access to a predetermined tool and / or a predetermined structured representation for analyzing said drive parameters; Based on the obtained prompt and the obtained access, accessing (S540) a tool and / or a structured representation in the predetermined tool and / or the predetermined structured representation; Based on the access, creating (S550) a plan for identifying at least one solution for achieving the goal; identifying ( S560 ) the at least one solution; as well as Based on a result of the identification, an output indicating the at least one solution is provided (S570) and / or a solution in the at least one solution is executed.

11. The method of claim 10, wherein the performing comprises at least one of: deploying a machine learning (ML) algorithm selected by said AI agent (100) during said creating and / or said identifying, deploying the driving parameters selected by the AI ​​agent (100) during the creation and / or identification on the driving application and / or in the commissioning of the driving application, resolving driver errors associated with the operation of the driver application and / or the debugging of the driver application, and Manage driver applications in offline mode for a predetermined period of time.

12. A control device for a drive application, the device being configured to perform the method according to claim 1 or 2, any one of claims 3 to 9 and / or any one of claims 10 or 11.

13. An automation system, comprising a first control device according to claim 12 and / or a second control device according to claim 12 and / or a third control device according to claim 12, wherein the first control device is configured to perform the method according to any one of claims 1 or 2, the second control device is configured to perform the method according to any one of claims 3 to 9, and the third control device is configured to perform the method according to any one of claims 10 or 11, and wherein at least two of the first control device, the second control device and the third control device are directly and / or indirectly communicatively connected.

14. A computer-readable medium comprising instructions which, when executed by a computing system, cause the computing system to perform the method according to any one of claims 1 or 2, any one of claims 3 to 9, and / or any one of claims 10 or 11.

15. A computer program product comprising instructions which, when executed by a computing system, enable or cause the computing system to perform the method according to any one of claims 1 or 2, any one of claims 3 to 9 and / or any one of claims 10 or 11.

Images