AI-supported automated parameterization method for field device, in particular IO link device

Abstract

The invention relates to a computer-implemented method for automatically parameterizing at least one field device communicatively connected to a communication network via a digital interface for a specified intended use of the at least one field device, in particular, the method is carried out using a configuration assistant having a large language model and based on generative artificial intelligence, the method comprises the steps of:-transmitting a configuration assistant in a natural language to the user via the computer interface to collect data relating to a specified intended use; -training a parameterized model by means of the collected data, or optimizing an existing parameterized model on the basis of the collected data; providing, via the computer interface, a pre-trained converter-based parameterized model comprising at least one converter component; -causing retraining of the pre-trained converter-based parameterized model using training data based on the collected data; issuing the trained converter-based parameterized model for automated parameterization of the at least one field device.

Description

Technical Field

[0001] This invention relates to the configuration or parameterization of field devices (particularly I / O linked devices, such as sensor arrangements for monitoring the status of technical devices connected via I / O linked communication systems). Background Technology

[0002] For example, sensor arrangements for monitoring the status of technical devices via known "IO link" communication connections are known in the prior art. Such sensor arrangements can be easily and quickly connected via "IO links" and can be integrated into existing technical devices relatively easily and reliably.

[0003] For example, the "condition monitoring sensor" developed by the applicant measures various physical variables, such as vibration, temperature, air humidity, and / or environmental pressure of technical equipment. In this way, the technical condition of, for example, machines, industrial equipment, and related components can be monitored. Therefore, the condition monitoring sensor enables the efficient and trouble-free operation of such machines or equipment, thereby significantly improving their effectiveness.

[0004] Physical variables measured by condition monitoring sensors are preprocessed and interpreted using integrated evaluation electronics, and the corresponding result data is then transmitted via I / O links to a higher-level system of the relevant machine or device.

[0005] Using standardized “IO Link” protocols, these sensors can be easily parameterized, and the evaluation of the sensors can be individually customized for the corresponding technology application or intended use.

[0006] For example, from DE 10 200 9 013 303 A1, it is known to use "IO link connection" to connect so-called "IO link devices." These "devices" are sensors and actuators that are known in themselves. Connections to other field devices, especially safety-related field devices that require parameterization, are typically performed in the prior art by directly connecting to the fieldbus or via a well-known REST API programming interface.

[0007] According to existing technology, there are two ways to parameterize the affected I / O link device. In both methods, various parameters (usually several) must be adjusted.

[0008] On one hand, parameterization can be done through the I / O Link Device Description File (IODD). The IODD format is standardized by the IO-Link Consortium and can be interpreted graphically by the corresponding software modules. A list (possibly grouped) of all parameters described in the IODD is created here. Each parameter is displayed along with its name and corresponding register. These values ​​can be changed and written to the device within permissible limits. Users must understand the meaning and effect of each parameter and relate them to their applications by referring to the provided documentation (manual).

[0009] Alternatively, parameterization can be performed without IODDs, and is actually performed via the controller or the corresponding software module. However, using this method, the registers must be manually extracted from the documentation, and the corresponding values ​​written to the sensor. This method is typically used for less complex I / O-linked sensors. Parameterization is usually performed directly by the control program provided in the programmable logic controller (PLC). Summary of the Invention

[0010] The object of this invention is to enable automated or automated parameterization of field devices, particularly those affected herein (especially I / O linked devices, including other field devices and especially safety-related field devices), in a simplified manner and with minimal technical labor or technical knowledge from the relevant user or startup engineer. In particular, it is also necessary to simplify the technical requirements for installing such devices or setting them up for changed or new applications or intended uses.

[0011] Therefore, the present invention proposes a method for simplifying or as automatically as possible configuring or parameterizing one or more field devices affected herein, particularly IO-linked devices, such as the aforementioned IO-linked sensor arrangement / device.

[0012] In this regard, it is assumed that the corresponding field device or I / O link device (e.g., the aforementioned sensor) has the capability for on-board preprocessing or further processing of the acquired data. Such further processing includes, for example, setting a switching signal in a length or position measurement system. The proposed method allows for configuration or parameterization of such devices not only on the device itself but also externally. By appropriately setting the various parameters, the on-board data preprocessing or data evaluation provided in the aforementioned sensor can be adapted to the requirements of the corresponding application or intended use.

[0013] It should be noted that configuring / parameterizing complex I / O linked devices (such as the condition monitoring sensors mentioned in the introduction) requires a considerable amount of transfer labor from the relevant user to translate application-specific knowledge into individual technical sensor parameters. To achieve this transfer labor, the user specifically needs specialized technical knowledge of the relevant application or intended use (e.g., when monitoring the condition of machine or system components through vibration analysis) and the corresponding I / O linked devices that need to be parameterized (e.g., vibration sensors). To this end, the user must identify all application-related parameters, correctly interpret their meaning and impact on the intended use or application using generally available documentation, select appropriate parameter values ​​for the intended use or application, and sequentially transmit or upload these selected parameters to the corresponding I / O linked devices (e.g., sensors) in the correct order. Dependencies between parameters must also be considered.

[0014] In the proposed method, a configuration assistant based on generative artificial intelligence (AI) automatically converts data inputs, depending on the application (e.g., a monitoring application), into a suitable set of parameters that can be uploaded to the corresponding sensor. Using the parameters uploaded in this process, corresponding to the respective application, parameters for data preprocessing on the sensor and parameters for data transmission of the preprocessed sensor data are calculated and preset.

[0015] According to a first aspect of the computer-implemented method of the present invention, the method is for the automated parameterization of at least one field device communicatively connected to a communication network via a digital interface for a predetermined intended use of said at least one field device. In particular, the method is implemented by means of a configuration assistant having a large language model (LLM) and based on generative artificial intelligence (GenAI), wherein the method includes the following steps:

[0016] Train a transformer-based parametric model using operation-related and / or product-related data from at least one field device, or optimize an existing transformer-based parametric model using such data.

[0017] A trained or optimized converter-based parameterized model is provided via a computer interface, the parameterized model including at least one converter component;

[0018] The configuration assistant enables the user to communicate with it in natural language via a computer interface to collect data related to the specified intended use of at least one field device;

[0019] This enables trained converter-based parameterized models to output optimized parameterized data using data collected for a specific intended use.

[0020] Publish optimized parameterized data output from a trained converter-based parameterized model for automated parameterization of at least one field device.

[0021] Operating instructions, data sheets, or similar product-related documents can be used as operating-related and / or product-related data for at least one field device.

[0022] The above training steps are preferably performed in advance, that is, before performing the automated parameterization of at least one field device.

[0023] At least one field device can be an I / O link device that is communicatively connected to the I / O link host via an I / O link communication system. Using such a standard communication system, according to the method of the present invention, the I / O link device connected thereto can be externally and automatically parameterized, particularly optimized for specific applications.

[0024] This method offers the advantage of enabling AI-guided configuration not only for simple I / O link sensors (so-called "click-clack" I / O link sensors) but also for more complex sensors or field devices that may or may not be suitable for communication via I / O link systems, such as the condition monitoring sensors mentioned in the introduction. These more complex sensors or field devices can be used, for example, to record operational data such as vibration, temperature, installation location, operating time, and similar parameters. According to the invention, suitable or optimized parameter values ​​can be found for the corresponding intended use or application of such simpler or more complex field devices. In this way, in particular, complex dependencies between parameters can also be automatically considered. This may involve additional technical requirements or conditions related to the intended use, or information related to parameterization, such as, in this case, the boundary conditions of the operation of the affected field device.

[0025] Furthermore, the method according to the invention provides increased benefits to users of sensor systems such as those discussed herein, as it ensures shorter installation time, fewer sources of error during installation, and immediate operational readiness.

[0026] According to another aspect of the method of the present invention, deep learning and / or reinforcement learning techniques can be used to train parameterized models. Using these learning methods has the advantage of being suitable for more complex tasks where the AI-based computer system must understand unstructured data, compared to simple (or non-generative) machine learning methods best suited for well-defined tasks with structured data. Thus, in deep learning, the function of the human brain is mimicked in processing data and creating corresponding patterns, whereby patterns and relationships in the dataset are determined from human-created content. Based on patterns learned in this way, new content is generated through generative AI. Supervised learning methods are typically used to train generative AI models. Reinforcement learning techniques mimic the human learning process of solving problems or achieving goals through trial and error.

[0027] According to another aspect of the method of the invention, the generation of the trained parameterized model can be additionally performed by sampling from the learned probability distribution of the parameter values. The advantage of this approach is that it can integrate previously learned "prior knowledge." Artificial intelligence (AI) typically learns blindly from large datasets, but in current application scenarios, these large datasets are unavailable, despite the existence of substantial theoretical knowledge about parameterization. Therefore, this theoretical information can be integrated into the AI ​​before training the parameterized model, enabling better interpretation of available data, even if it is scarce. Bayesian statistics can also be used to determine how the relevant prior knowledge can be optimally utilized, allowing the parameterized model in question to be trained efficiently with significantly less data. Such sampling can be performed using systematic stochastic methods, such as random number generators.

[0028] According to another aspect of the method of the present invention, raw data based on a predetermined parameter set can be preprocessed using a preprocessing device to provide training data based on the parameter set. To ensure the quality of the parameter data generated using GenAI, it is advantageous to obtain and process suitable and correct data for parameterization, as the quality of this data is also important for the quality of parameterization. When preprocessing this data, it is advantageous to prepare the data in a manner that facilitates evaluation for training the GenAI system.

[0029] According to another aspect of the method of the invention, the input request (so-called "hint") given to the user may include one or more contexts relating to further information about the intended use. This is because the results of generative AI are only as good as those prompted by GenAI user input. In this case, the results can be significantly improved, particularly by the contextual information about the corresponding intended use.

[0030] According to another aspect of the method of the invention, the method may include updating training data based on a parameter set and initiating retraining of the trained converter-based parameterized model based on the updated parameter set. By repeatedly retraining the trained parameterized model based on the current parameter data, the quality of the parameter set determined using GenAI can be significantly improved. It may also be provided that the retraining of the already trained converter-based parameterized model is planned retraining, continuous retraining, or trigger-based retraining. Furthermore, it may be provided that the trigger for trigger-based retraining is based on thresholds and ratings related to the configuration or parameterization instructions provided by the trained converter-based parameterized model.

[0031] The present invention also relates to a computer-implemented method for generating a parameter set suitable for a given application using a trained converter-based parameterization model as described above, the parameter set being used for the operation of at least one field device communicatively connected via a digital interface of a communication network, wherein the method includes the following steps:

[0032] Provides users with access to trained, converter-based parameterized models via computer processors;

[0033] Receive data input for a given intended purpose via a computer interface;

[0034] It enables the trained converter-based parametric model to analyze input data and provides instructions for generating new or modified parameter sets based on the trained parametric model.

[0035] According to another aspect of the method of the present invention, it can be provided that at least one field device is automatically configured for a specified intended use using the generated new or modified parameter set. Furthermore, regarding this use of the corresponding parameterized model, the at least one field device can be an I / O link device communicatively connected to an I / O link host via an I / O link communication system.

[0036] According to another aspect of the method of the present invention, a user input request (“prompt”) can be provided to prompt the user to request a trained converter-based parametric model to provide machine-readable instructions based on a query submitted by the user to the parametric model via a computer interface.

[0037] According to another aspect of the method based on the invention, the input request may include one or more operational instructions relating to one or more operational procedures of an I / O link device for a specified intended use. It may also be provided that the operational instructions include machine-readable instructions for controlling and / or monitoring the operation of the I / O link device.

[0038] The present invention also relates to a computer program product including computer-readable instructions that, when executed on a computer, cause the computer to perform the above-described method steps.

[0039] Furthermore, the present invention relates to a computer product including a computer-readable token for accessing training data for the above-described training of a parameterized model and / or for accessing a trained or pre-trained model.

[0040] The process implemented in the configuration assistant for the corresponding intended use (i.e., application-specific configuration of the I / O linked device) enables the aforementioned transfer of specified parameter sets to be performed as automatically as possible. Users only need to provide the expert system with knowledge of the corresponding application through an intuitively guided process.

[0041] Therefore, users only need to specifically address the corresponding application or intended use and provide the corresponding input data. The expert system asks users basic, guided questions that are generally easy for them to answer. In the method or apparatus according to the invention, this data input is automatically enriched with the applied expert knowledge and transformed into a set of technical parameters.

[0042] In this process, users are systematically and intuitively guided through specific dialogues, where they can combine their application knowledge regarding the technical / mechanical aspects of the relevant application. Due to the high degree of automation in expert systems, users need to perform little or no intellectual transfer labor.

[0043] Users only need to interact with the configuration assistant and provide their input. In particular, no labor or expert knowledge needs to be transferred to the user side.

[0044] The optimized set of parameters, tailored to a given intended use, can then be uploaded to one or more affected field devices, such as one or more I / O linked devices, within a single routine. This approach significantly improves the effectiveness and efficiency of configuring or parameterizing complex I / O linked devices. Attached Figure Description

[0045] Exemplary embodiments of the invention are illustrated in the accompanying drawings, and are explained in more detail in the following description.

[0046] Figure 1 A parameterized sensor arrangement for condition monitoring of system components, according to the prior art, is shown.

[0047] Figure 2 The following reference is shown. Figure 3 An excerpt of the parameter list describing the I / O-link-based sensor deployment;

[0048] Figure 3An I / O-link-based sensor arrangement for condition monitoring of system components is shown;

[0049] Figure 4 The invention is shown Figure 3 The parameterization of the IO-link-based sensor layout shown uses a configuration assistant for parameterizing the sensor layout, employs a large language model and is based on generative artificial intelligence, specifically using a combined flowchart / block diagram.

[0050] Figure 5 A parameterized implementation of an I / O-link-based sensor arrangement based on generative artificial intelligence according to the present invention is shown for monitoring the condition of a system or system component.

[0051] Figures 6A-6C The structure of a converter encoder / decoder architecture for generating parameter data is shown, particularly based on the converter-based parameterization model described above. Detailed Implementation

[0052] Figure 1 A sensor arrangement according to the prior art, such as an arrangement of two vibration sensors 100, 105 and an inductive velocity sensor 110, is shown, along with known devices for its configuration and parameterization. The sensor arrangement 100-110 shown is used for vibration monitoring of a system or machine (not shown here) and is designed to trigger an alarm when a predetermined vibration intensity is exceeded.

[0053] In the exemplary embodiment shown, three sensors 100, 105, and 110 transmit high-frequency analog signals or values ​​to an electronic diagnostic unit 130, such as an industrial PC, via corresponding analog communication connections 115, 120, and 125. The diagnostic unit 130 further processes the raw data 115, 120, and 125 provided by the sensors 100, 105, and 110 to determine if the aforementioned alarm conditions may exist. The diagnostic unit 130 transmits the corresponding aggregated data to a higher-level system via corresponding Ethernet or Internet Protocol (IP) based communication connections 135 and 140; in this embodiment, this data is transmitted to the IT network of a "cloud computing" platform 145, and also to a programmable logic controller (PLC) 150.

[0054] Configuration software 160 installed on computer 155 is used to parameterize diagnostic unit 130 and sensors 100, 105, and 110. However, in the monitoring scenario assumed here, Figure 1 The sensors 100, 105, and 110 shown do not have the capability for further onboard processing of this parameterized data.

[0055] Parameters can be read into the computer 155 and / or written to the corresponding sensors 100, 105, 110 via a graphical user interface installed in the computer 155.

[0056] However, in order to specifically configure even more complex sensors (such as the aforementioned “condition monitoring sensors” (hereinafter referred to as “CM sensors”) developed and distributed by the applicant) for their respective applications or intended uses, a large number of parameters must be manually set by the user continuously.

[0057] For example, the latest generation (“V2”) of CM sensors has a total of approximately 700 individual presettable parameters. In order to parameterize the CM sensor, well-known and standardized “IODD” files (i.e., machine-readable text files) are processed to generate a list of all parameters required or expected for parameterization of such IO-linked devices.

[0058] Furthermore, there are often complex dependencies between the various parameters. In order to establish... Figure 1 The new application of the sensor arrangement shown requires manual adjustment of at least 20-30 different parameters with high precision.

[0059] Parameterization can be achieved using the aforementioned "IODD" file ("IODD" = IO link device description). Figure 1 Taking the two vibration sensors 100 and 105 shown as examples, Figure 2 The excerpt of the parameter list shown illustrates the very high complexity of configuring the parameters for the IO link devices involved here.

[0060] The parameters for the following two areas are shown here only as examples (or excerpts):

[0061] 1. "Advanced Vibration Configuration" and

[0062] 2. "Vibration speed alarm configuration".

[0063] Part 1.1. On the one hand, this relates to the configuration of time windows associated with vibration detection. The specifications of appropriate time values ​​for these time windows are crucial to the quality of vibration detection. The appropriate time value depends largely on the specific application, i.e., the system component or machine to be measured. Their vibration behavior, especially the possible vibration frequencies, is critical to the size of the time window set for vibration detection in each case. Therefore, a preset time window that is too small results in potentially detected vibrations being considered only partially because the complete phase distribution is unavailable. On the other hand, a time window that is too large may lead to the detection of vibrations or corresponding oscillations relatively late or even too late because too much of the complete phase distribution must be evaluated.

[0064] Secondly, Part 1. Presets for response latency (“Event Response Latency” value) and bandwidth value. The latter is used to limit the lower bandwidth and upper bandwidth (“Upper Bandwidth Limit” and “Lower Bandwidth Limit” values).

[0065] The quality of vibration detection can also be significantly improved by appropriately adapting it to the presets of the application. For example, the bandwidth values ​​mentioned above can be adapted to the expected vibration behavior in the application to ensure that the specified bandwidth of the vibration sensor available to us is sufficient to detect and process the detected vibrations, i.e., their vibration frequencies.

[0066] Part 2.2. Presets for various parameters that affect the behavior of pre-alarms and main alarms, which are automatically triggered when certain pre-set detected vibration behaviors are detected. This is based on the so-called "RMS" value of the vibration velocity (vibration speed), where "RMS" corresponds to the root mean square (RMS) based on the detected vibration data.

[0067] Therefore, these parameters related to the alarm behavior of the monitoring device in question must be configured very precisely to ensure that alarms are actually and reliably triggered in the typical alarm events for the corresponding application scenario.

[0068] Figure 3 An I / O-link-based sensor arrangement for monitoring the condition of system components to be monitored is shown, wherein the parameter configuration of the I / O-link devices 300, 305, 310 involved herein can be implemented according to the present invention.

[0069] I / O Link is a standardized I / O technology (IEC 61131-9) for communicating with sensors and actuators. Robust point-to-point communication is based on the well-known three-wire sensor and actuator connection principle, which places no additional requirements on cable materials. Therefore, I / O Link is not a fieldbus, but rather a further development of existing and well-proven connection technologies for sensors and actuators.

[0070] The I / O link is functional and enables enhanced diagnostics of sensors and actuators, as well as simple and fast parameterization, through bidirectional communication. It supports fast communication at three speeds: 4.8 kB, 38.4 kB, and 230.4 kB. It can also be implemented in a very small structure, enabling the miniaturization of "smart" sensors and actuators.

[0071] Figure 3 The three IO link devices 300-310 shown each represent a second-generation CM sensor (“V2”). This has an M12 connector connection that provides or supports the IO link communication protocol, and in particular, has the capability for further “onboard” processing of parameterized data.

[0072] CM sensors 300-310 are each connected to the I / O link host 330 via bidirectional I / O link connections 315, 320, and 325. The I / O link host 330 is also connected to a higher-level IT network 340, such as a cloud computing platform, via bidirectional I / O link connection 335 using data and communication technologies. Additionally, the I / O link host 330 is connected to a programmable logic controller (PLC) 350 via Ethernet or (Internet Protocol (IP) based) communication connection 345.

[0073] Within the IT network 340, sensor data provided by CM sensors 300-310 can be further processed and evaluated for the corresponding monitoring task (or application scenario). Therefore, sensor data can be compared to empirically specified thresholds to issue, for example, a warning message, or evaluated accordingly using machine learning methods with the aid of an artificial neural network. For example, such a neural network can be trained in a conventional manner using previously generated sensor data.

[0074] Then, the warning message provided by the IT network can be converted by the programmable logic controller 350 into a corresponding control intervention for the monitored system or machine (not shown here), and the existing alarm can be canceled through this control intervention.

[0075] According to the present invention, the IO link host 330 is further connected to a computer 355, on which a generative AI-based configuration assistant 360 is specifically installed. The configuration assistant 360 is used to parameterize the CM sensors 300-310. Figure 5 The diagram schematically illustrates the structure and functionality of Configuration Assistant 360, which is based on an optimization approach involving generative AI and a corresponding large language model.

[0076] See below for reference Figure 4 A more detailed description of the use of neural networks for parameterized GenAI systems that can be automated or supported by GenAI can be provided as a GenAI approach.

[0077] Figure 4 An exemplary embodiment of the method steps of the configuration assistant 400 according to the present invention is, in this case, based on an optimization or corresponding learning method 405 based on GenAI, by which the sensor arrangement 415 described herein by way of example for the field device involved (see...) Figure 3For a given application scenario or intended use 410, the three CM vibration sensors 300-310 are used to determine parameters 420 related to parameterization, for example, through the sampling described above. Furthermore, by means of data preprocessing, a suitable initial training field device 425 is created for the parameters 420 thus determined for application scenario 410. In this embodiment, the CM vibration sensors 300-310 represent IO link devices or field devices connected to an IO link communication system, but they can also be connected to another (industrial-grade) communication system.

[0078] Specifically, the GenAI-based computer interface 430 is used for initial recording and temporary storage of the corresponding sensor arrangement 415 (e.g., Figure 3 The three I / O linked devices (300-310) shown herein require technical information or boundary conditions 435 for automatic configuration or parameterization of the intended application. Configuration or parameterization is performed, in particular with reference to a specific application scenario or intended use 410, and especially through interaction 440 with a user or operator 445 of the configuration assistant 400, conducted in natural language via a corresponding computer interface. The intended use is entered by issuing a corresponding input request (“prompt”) to the user, wherein the input includes one or more contexts indicating further technical information related to the specified intended use. Data techniques based on associated context / metadata can be used to process the context related to the specified intended use.

[0079] Based on data 435 obtained by user or operator 445 through input 440, configuration assistant 400 trains a converter-based parameterization model 450 existing or predefined in a database or data store for the corresponding field device 415 (e.g., the previously described CM sensors 300-310) and corresponding intended use. This pre-trained parameterization model 450 (which includes or incorporates at least one converter component described below) is available for user or operator 445 to interact with 440 via computer interface 430. The existing parameterization model 450 is retrained using second training data generated based on an updated parameter set. This retrained converter-based parameterization model is now published to the corresponding user 445 via computer interface 430 for automated parameterization of the existing field device 415 and corresponding intended use.

[0080] The parametric model thus trained can now be used to generate a set of parameters suitable for the operation of the existing field device 415 for its existing intended use. This is performed through the interaction of a configuration assistant 400 with or in collaboration with a GenAI system. The user 445 accesses the currently trained converter-based parametric model via a computer processor. Data input for the specified intended use is recorded via a computer interface 430. The currently available parametric model is now initiated to analyze the input data and provide instructions for generating a new or modified set of parameters 455 based on the current parametric model 460. The field device 415 can now be automatically configured using the generated new or modified set of parameters for the specified intended use. For this purpose, the field device can be prompted by appropriate prompts to request machine-readable instructions from the parametric model based on queries submitted by the user to the parametric model via the computer interface. Prompts may include one or more operational instructions related to one or more operational procedures of the field device 415 corresponding to the intended use. Operational instructions may include machine-readable instructions for controlling and / or monitoring the operation of the field device 415.

[0081] Advantageously, user 445 does not need to process the various technical parameters and their meanings as is common in current technologies. Therefore, the configuration with various parameters as used in the prior art is not necessary.

[0082] For the aforementioned CM sensors 300-310, parameter sets created in the manner described for their respective intended uses can be directly applied because these sensors have integrated parameter management. Wired interfaces with bidirectional communication (e.g., IO links or Modbus) and wireless interfaces (e.g., Bluetooth or long-range radio) can be considered as communication paths for parameter set applications.

[0083] The described method for automatic parameterization of the field device 415 under discussion (e.g., the aforementioned IO-linked device) also significantly improves the effectiveness and efficiency of configuration or parameterization, especially for complex field devices.

[0084] In the following text Figure 5 The illustrated implementation describes in more detail the precise configuration process using an expert system with a corresponding configuration assistant based on the generative AI described herein. In this implementation, the I / O link device to be parameterized relates to a monitoring device intended for the purpose of "monitoring the operational status of technical equipment".

[0085] Following the start 500 of the configuration sequence or process shown, the underlying machine type intended for parameterized IO link devices is first detected 502. Specifically, in this example, this is a monitoring device used to monitor the operational status of a technical device via IO link operation. It is also assumed that the device has components that move along various spatial axes, for example, sensor technologies developed or sold by the applicant, such as corresponding tilt sensors with multiple measurement axes.

[0086] Machine type detection 502 is preferably performed via input from a user or operator, and particularly in this embodiment, it is performed via a configuration assistant implemented by interactive function blocks 700-735. For this purpose, bidirectional data exchange 504 is performed with the first function block 700. Possible machine types include, for example, machine categories such as motors, pumps, fans, and / or compressors.

[0087] Then, the monitoring functions of the IO-linked device to be parameterized are obtained at 506, specifically via user input through the second function block 705 (connected via bidirectional data exchange 508). Possible monitoring functions include, for example, the category of mechanical vibration of the machine measured on non-rotating parts (e.g., centrifugal pumps) according to ISO 10816.7, and / or the temperature value to be monitored, and / or the user-specific signal peak value to be monitored, and / or the user-related root mean square (RMS) value to be monitored. This can be based on the root mean square value of a time-varying physical quantity such as alternating current or alternating voltage.

[0088] Then, the input data 510 required for parameterization is obtained. In this exemplary embodiment, this includes product category 512, performance category 516, and axis assignment 520 for the I / O link device used for parameterization.

[0089] To obtain the three input data 512, 516, and 520 mentioned above, bidirectional data exchange 514, 518, and 522 with the corresponding functional blocks 710, 715, and 720 is also performed here. Therefore, in this exemplary embodiment, pump categories such as "Pump Category I" and "Pump Category II" are provided in the third functional block 710. However, in the fourth functional block 715, electrical power levels are provided, particularly based on the corresponding number of rotor blades; for example, 1-200 kW for >3 rotor blades, 201-1000 kW for >=3 rotor blades, and / or >100 kW regardless of the number of rotor blades. Finally, in the fifth functional block 720, one or two shafts are provided, or can be selected, arranged perpendicular to the drive shaft of the electric actuator of such pump.

[0090] In the exemplary embodiment shown, data regarding the alarm settings 524 to be provided on the monitoring device connected via the IO link is then collected, specifically the corresponding alarm value 526 for the IO link device based on the underlying sensor data. Thus, using so-called "smart sensor" technology, an alarm can be triggered if moisture penetrates the corresponding device. This can indicate, for example, that the device is in extreme environmental conditions. In any case, the corresponding alarm level must be parameterized.

[0091] To record the aforementioned alarm value 526, a bidirectional data exchange 528 with the relevant sixth function block 725 is also performed here. In this function block 725, various risk categories are specified according to the degree of hazard. These can be based, for example, on risk analysis, through which all hazards associated with a particular machine can be identified. This risk assessment is based on a series of logical steps, for example, according to DIN EN ISO 14121, which makes it possible to systematically investigate the potential hazards posed by the corresponding machine.

[0092] Finally, in this exemplary embodiment, the data or values ​​required for data transmission 530 (e.g., wireless data transmission) are obtained. In this example, this data relates to control parameters for digital data transmission, such as the data transmission protocol to be used in each case, an appropriate transmission rate 532, and the specification of the application 536 on which the data transmission is based.

[0093] To obtain the two input data 532 and 536 mentioned above, bidirectional data exchange 534 and 538 with the corresponding function blocks 730 and 735 is also performed here. Therefore, in the seventh function block 730, based on the current existing configuration of the monitoring device used for monitoring the operating status of the technical device via IO link operation, the following data is automatically transmitted:

[0094] - Process values, such as vibration amplitude and / or vibration velocity (again, if necessary, as the corresponding RMS values).

[0095] - Status bits, such as the corresponding bits for alarm zones "Pre-alarm", "Main alarm" and / or "Danger Zone".

[0096] Finally, in the eighth function block 735, the user records the exact specification of the corresponding application or intended use. An example of this could be, for instance, "first monitoring function".

[0097] Based on all the configuration data already recorded for the parameterization of the corresponding I / O link device, this data is now processed, indicated by the dashed line on the right, to automatically create the final I / O link parameters. Data line 540 is first used to check 542 whether the corresponding monitoring function of the corresponding I / O link device should be parameterized for the first time. If so, the predefined initial I / O link parameters are retrieved via data line 544 546. In the next processing step 550, connected via additional data lines 548, 552, corresponding I / O link parameter data suitable for further processing is generated.

[0098] In parallel with the processing steps 542-550 described above, the acquired input data 510 and 524 (i.e., the aforementioned data 512, 516, 520 and alarm-related data 526) are fed to the database 560 at 558. In this embodiment, the knowledge-based database stores possible applications of the I / O linked devices (e.g., monitoring applications discussed herein) in a rule-based manner for use with such application-related data.

[0099] Based on the rules corresponding to the currently available data 512, 516, 520, and 526, the IO link parameter data corresponding to 566 is generated via data line 564. Additionally, in this exemplary embodiment, the aforementioned data transmission data 532 and 536 are also converted 572 into corresponding IO link parameter data via data line 570.

[0100] The total IO link parameter data, ultimately obtained by combining or merging data lines 554, 568, and 574 via 556, is specifically used as input data for the parameterized model of the GenAI system (which has been previously trained with the corresponding data), or as input data for the artificial neural network on which the system is based. The GenAI system then automatically delivers IO link parameter data suitable for the current application or intended use of the IO link device (in this case, a monitoring application).

[0101] It should be noted that the sequences shown here for the entire configuration 502-536 or the corresponding function blocks 700-735 are merely exemplary and may be modified if necessary.

[0102] Therefore, the final generation of all IO link parameters 578 based on the described acquired data is based on the described GenAI method, which determines a set of parameters suitable for the corresponding application or intended use based on data input by the user or operator, and in particular, takes into account the correlation between the underlying parameters.

[0103] The entire processing sequence 502-578 finally ends 576 with the parameterized result 578 mentioned above existing.

[0104] Figures 6A-6C The known structure of the converter-encoder / decoder architecture relevant here is shown for generating parameter data based on a named converter-based parameterized model.

[0105] Figure 6A One implementation of a converter-encoder architecture is shown. The converter encoder includes an encoder input 678, one or more encoder blocks 674, 614, and an encoder output 676. In this implementation, the encoder output 676 includes two processing stages: a linear stage 616 and a so-called "softmax" stage 618. This linear stage is known to connect each input neuron of a neural network to each output neuron. This "softmax" stage is known to be frequently used in the final stage of neural network models for classification tasks, and it converts the original output values ​​(so-called "logits") into probabilities by forming an exponential function for each output and normalizing these values ​​by dividing them by the sum of all values. In this way, a vector of K real values ​​can be converted into a vector of K real values ​​that sum to 1. The input values ​​can be positive, negative, zero, or greater than one, where the "softmax" function converts these values ​​into values ​​between 0 and 1, making them interpretable as probability values.

[0106] The converter-encoder architecture shown here can be derived from converter-encoder-decoder architectures known in the prior art (see [link]). Figure 6C The converter-encoder architecture shown can have an additional encoder output, so as... Figure 6C The decoder shown is a converter-encoder-decoder architecture where the encoder block is directly connected.

[0107] Input data is received at encoder input 678, which may apply a so-called "embedding" 602 to the input data. Applying input embedding 602 may instruct the input data to be transmitted through an embedding level. Furthermore, encoder input 678 may apply position encoding 604. Applying position encoding 604 may refer to adding position factors to the embedded input data. Preferably, the input data may specify a sequence of elements, wherein the position factors can provide an indication of the position of the elements within the sequence.

[0108] The embedded input data can be processed by one of the encoder blocks 674, 614 (optional blocks) shown. The embedded input data can be provided via a special data connection to normalization 608 (hereinafter referred to as "level normalization") performed at the appropriate level or multiple levels of the underlying software architecture. Such levels / multiple levels may involve, for example, the application level (or user interface), the level of the "large language model," and the data level intended for training a parametric model.

[0109] For such embedded input data, a so-called "multi-head self-attention" mechanism can be implemented, which allows the underlying model to focus on different parts of the input data, similar to how humans focus on specific words when understanding sentences. Through this mechanism, the model can determine which parts of the input are relevant to a specific task, making it extremely flexible and powerful. Therefore, this processing step on the transformer is used to improve the expressiveness and modeling capabilities of neural networks, which form the basis of AI systems. Thus, the transformer can capture various types of dependencies and relationships between words or elements in the input data sequence. Such a processing step can also be understood as a filter applied to the embedded input data. By applying a filter to the embedded input data, elements associated with the embedded input data that contribute to the output data to be generated can be identified for use in generating the output data. Therefore, a filter can represent the degree of contribution of elements associated with the embedded input data to the output data to be generated.

[0110] For the embedded input data, in addition to multi-head self-attention 606, a so-called feedforward stage 610 and a corresponding level normalization 612 can be provided. Through this feedforward stage, a feedforward neural network is formed in a well-known manner, consisting of an input stage, one or more hidden stages, and an output stage. Therefore, data flows in only one direction, i.e., from the input stage through the hidden stages to the output stage.

[0111] Figure 6B One implementation of a converter-decoder architecture is illustrated. The converter-decoder includes a decoder input 684, one or more decoder blocks 680, 632, and a decoder output 692. Converter-decoder architectures are known from the prior art and... Figure 6C The converter-encoder-decoder architecture is derived from the converter-encoder-decoder architecture shown. The converter-decoder architecture can correspond to the decoder architecture associated with it and is independent of receiving one or more hidden states from the encoder of the converter-encoder-decoder. Various such converter-decoder architectures are available in the prior art, such as the well-known general and trained “GPT” converter.

[0112] Decoder input 684 can apply the current embedding 620 and positional encoding 622 of the input data, similarly, input embedding 602 and positional encoding 204, as in... Figure 6A As described in the corresponding context. Decoder block 680 may include a corresponding level normalization 626, a mask “multi-head self-attention” 624, a so-called “feedforward” stage 628, and a corresponding level normalization 630, which are provided via special data connections of the corresponding level normalization 626.

[0113] Furthermore, the "multi-head self-attention" 624 of the mask can be applied to the embedded input data, where the "multi-head self-attention" 624 of the mask essentially corresponds to Figure 6A The context of "multi-head self-attention" is as follows. However, it includes an additional mask of the portion of the embedded input data associated with elements in the sequence that appear later than the element to be generated. Alternatively, the portion of the input data associated with elements in the sequence that appear later than the element to be generated may not be received and / or not transformed into the embedded input data. Thus, the transformer-encoder can be configured for classification tasks, although the transformer-decoder can only be configured for text generation.

[0114] Figure 6C The interaction or collaboration between the described converter-encoder and converter-decoder is illustrated.

[0115] The converter-encoder-decoder may include an encoder input 688, one or more encoder blocks 686, 664, a decoder input 694, a decoder block 690, and a decoder output 692. The encoder input 688 may correspond to... Figure 6A The encoder input is 278. One or more encoder blocks 686, 664 can correspond to... Figure 6A One or more encoder blocks 674, 614. Decoder input 294 can correspond to Figure 6B The decoder input is 684. Decoder block 690 may include a multi-head self-attention 670 for the mask, level normalization 672, the previously mentioned feedforward stage 638, and level normalization 640. These can be formed similarly to the multi-head self-attention 624, level normalization 626, 630, and feedforward stage 628 for the mask, as in... Figure 6B As described in the context. Decoder block 690 may also include multi-head self-attention 650 and level normalization 648.

[0116] Similar to Figure 6B The description suggests that the context tensor can be obtained from the multi-head self-attention 670 and level normalization 672 of the mask. Level normalization 648 can be applied to the context vector obtained from the multi-head self-attention 650 and the hidden states of one or more encoder blocks 686, 664. Similar to... Figure 6B The context vector generated by level normalization 648 can be processed via feedforward stage 638 through the aforementioned level normalization 640. Similar to decoder block 690, the context vector generated by level normalization 640 can be provided to another decoder block 642. Context vectors obtained from one or more decoder blocks 690, 642 can be provided to decoder output 692. Decoder output 292 can correspond to... Figure 6B The decoder outputs 682.

[0117] Using the described architecture, the converter-encoder-decoder can receive input data at encoder input 688 and one or more encoder blocks 686, 664, and at decoder block 690 and decoder output 692, and process it as described. Based on the input data, the converter-encoder-decoder can sequentially generate output data. The sequentially generated output data can be provided to decoder input 694, one or more decoder blocks 690, 642, and decoder output 692, and / or processed by decoder input 694, one or more decoder blocks 690, 642, and decoder output 692. Preferably, a sequence can be provided to decoder input 694. After decoder input 694 has generated at least a portion of the output data, at least a portion of the elements of the generated output data can be provided to decoder input 694. Thus, by considering both the input data and the generated output data, subsequent elements of the output data can be generated with higher accuracy.

Claims

1. A computer-implemented method for automatic parameterization of at least one field device (300-310, 415) communicatively connected via a digital interface to a communication network (315-325) for a specified intended use of the at least one field device, characterized in that, The method is performed by a configuration assistant (360, 400) having a large language model (Fig. 4) and being based on generative artificial intelligence, wherein the method comprises the following steps: - training (Fig. 4) a parameterized model based on a transducer (450) by means of operation-related and / or product-related data (425) of the at least one field device, or optimizing an existing parameterized model based on a transducer based on such data; - providing the trained or optimized parameterized model based on a transducer via a computer interface (430), the parameterized model comprising at least one transducer component; - communicating (440) the configuration assistant (400) in natural language to a user via the computer interface (430) to obtain data related to the specified intended use of the at least one field device; - causing (445) the trained parameterized model based on a transducer (450) to output optimized parameterized data (455) using the data obtained for the specified intended use; - issuing (445) the optimized parameterized data (455) output by the trained parameterized model based on a transducer (450) for automated parameterization of the at least one field device (300-310, 415).

2. The method according to claim 1, wherein the at least one field device (300-310, 415) is an IO-link device communicatively connected to an IO-link host via an IO-link communication system (315-325).

3. The method according to claim 1 or 2, characterized in that, The training of the parameterized model (450) is performed using deep learning and / or reinforcement learning techniques.

4. The method of claim 3, wherein, The generation of the trained parameterized model (450) is additionally performed by sampling from a probability distribution of learned parameter values.

5. The method according to any of the preceding claims, characterized in that, Raw data based on a predetermined parameter set is pre-processed by a pre-processing device to provide the training data based on the parameter set.

6. The method according to any of the preceding claims, characterized in that, The user prompt input data comprises one or more contexts related to further information about the specified intended use.

7. The method according to any of the preceding claims, characterized in that, The method comprises updating the training data based on the parameter set and causing the trained parameterized model based on a transducer to retrain based on the updated parameter set.

8. The method of claim 7, wherein, The retraining of the already trained parameterized model based on a transducer is a predetermined retraining, a continuous retraining or a trigger-based retraining.

9. The method of claim 8, wherein, The trigger for the trigger-based retraining is based on a threshold and an evaluation related to configuration or parameterization instructions provided by the trained parameterized model based on a transducer.

10. A computer-implemented method for generating a parameter set (455) using a parameterized model based on a transducer (450) trained according to one of the preceding claims, the parameter set (455) being suitable for a specified intended use of operation of at least one field device (300-310, 415) communicatively connected via a digital interface of a communication network (315-325), wherein the method comprises the following steps: - providing, by a computer processor, a user (445) with access to a trained converter-based parameterized model (450); - receiving, via a computer interface (430), input data (435) related to the given intended use; - causing the trained converter-based parameterized model to analyze (450) the input data and provide instructions for generating a new or modified parameter set (455) based on the trained parameterized model (450).

11. The method of claim 10, wherein the at least one field device is an IO-Link device communicatively connected to an IO-Link host via an IO-Link communication system.

12. The method of claim 11, wherein, By means of the generated new or modified parameter set, the at least one field device is automatically configured for the specified intended use.

13. The method according to claim 11 or 12, characterized in that, By means of an input prompt, the user is caused to request the trained converter-based parameterized model to provide machine-readable instructions based on a query submitted by the user to the parameterized model via the computer interface.

14. The method of claim 13, wherein, The input prompt comprises one or more operational instructions related to one or more operational processes of the IO-Link device according to the specified intended use.

15. The method of claim 14, wherein, The operational instructions comprise machine-readable instructions for controlling and / or monitoring the operation of the IO-Link device.

16. A computer program product comprising computer readable instructions which, when executed on a computer, cause the computer to perform the steps of any one of claims 1 to 15.

17. A computer product comprising a computer readable token for accessing training data for training a parameterized model and / or accessing a trained or pre-trained model according to one of claims 1 to 15.

Images