Method and system for managing messages of an automation system, in particular an industrial facility
Functions trained through machine learning algorithms filter and predict important messages in industrial facilities, solve the problem of operators being flooded by messages, realize timely response to critical messages, avoid undesired reactions from the automation system, and improve production safety and efficiency.
Patent Information
- Application Number
- CN202080025915.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2019-03-29
- Filing Date
- 2020-03-12
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2040-03-12
AI Technical Summary
In complex industrial facilities, operators are flooded with a large number of messages, resulting in important alarms being ignored and timely measures cannot be taken, resulting in facility downtime and production failure. The existing systems are mainly targeting changes in alarm rules, but operators are unable to effectively interfere with the alarm system and there are security risks in public communication networks.
Functions trained using machine learning algorithms, filter and predict important messages by defining the first and second categories of messages, avoiding intervention in the reporting system, and using independent second human-machine interface to display messages to ensure safe communication separation, suitable Allocation facilities change, optimize message filtering and prediction.
Effectively focus the operator's attention on important messages, reduce false alarms and missed reports, improve the response speed to critical messages in the facility, avoid undesirable reactions to automation systems, and improve production safety and efficiency.
Smart Images

Figure CN113646720B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method and a system for managing messages of an automation system, in particular an industrial plant, according to claims 1 and 8. The invention also relates to an evaluation method and an evaluation system according to claims 6 and 13 and to a program and a computer program product according to claims 15 and 16. Background Art
[0002] Automation systems are used to monitor, control and / or regulate mobile and stationary devices and installations and to enable the operation of control devices, sensors, machines and industrial installations, which should be as autonomous and independent of human intervention as possible.
[0003] For example, automation systems are used in the fields of industrial production, energy generation and distribution, building management, logistics, drive technology, etc. The automation systems usually generate a large number of messages for, for example, the operation and maintenance personnel of the system. The messages are usually displayed to the personnel on a human-machine interface (HMI), such as an operator station, a PC or a portable device with communication capabilities, such as a tablet or a smartphone. The automation system should draw the attention of the personnel to changes in the state of the device or the facility, which may require the intervention of the personnel.
[0004] Typically, the message relates to a specific event in the automation system. A message can be, for example, an alarm message or a notification message. An alarm message is a message that an error or an abnormal event has occurred and personnel must take immediate action. For example, a notification message can indicate that maintenance is required. Event messages can include text and other information, such as the time and location at which the message was generated.
[0005] The messages are typically generated by a reporting system that is communicatively coupled to a human machine interface (HMI), such as via a communications network.
[0006] Industrial facilities are used in various industrial sectors, such as energy generation, process industries (e.g. paper, chemicals, pharmaceuticals, metals, oil and gas) and discrete manufacturing. In this case, the actual industry-specific processes, such as energy generation or production processes, are controlled and / or regulated by automation systems. Automation systems without field devices (i.e. without actuators and sensors) are often also referred to as "process control systems".
[0007] The reporting system is then usually implemented as a function in a so-called "operator system server" or "application server", in which one or more system-specific application programs are stored and executed when the installation is running.
[0008] In a complex industrial facility, such as a paper mill, the reporting system may generate hundreds of messages per minute. Many messages are absolutely important for indicating possible anomalies, but not all messages have the same impact on the operation of the process. As a result, it can happen that the operator of the facility is overwhelmed by alarm messages and is thus unable to react appropriately to truly important alarms. Sometimes, up to 80% of the messages are actually ignored by the operating / monitoring staff, but filtering out important messages that may be critical for the facility requires years of experience. Even if the operator filters out an important alarm very quickly, there is usually not enough time to take appropriate countermeasures. This leads to facility downtime and production losses.
[0009] In order to cope with the large number of messages, some solutions are already known. Thus, for example, EP 3 187 950 A1 discloses a method for managing alarms in a control system for an industrial process. The control system comprises a reporting system having a notification server generating alarm messages based on preset notification generation criteria, an operator station communicating with the server and a database with historical alarm data. Here, the historical alarm data is received in the notification server, a first group of alarms is defined as "important alarms" based on a predetermined rule, and a second group of alarms is defined as "minor alarms". A plurality of historical data segments are identified in the historical alarm data, wherein each of the data segments contains at least one alarm message of the first group of alarms. In addition, a plurality of correlation scores for the plurality of alarms of the second group are determined, and finally the notification generation criteria preset for the alarms are modified based on the plurality of correlation scores. By modifying the notification generation, the minor alarms of the second group are prioritized or suppressed. In addition, based on the correlation scores of the minor alarms of the second group, the important alarms of the first group and the estimated time period of their occurrence can be predicted.
[0010] Furthermore, it is already known to use neural networks to predict alarms or impending failures of components.
[0011] Known systems are mainly aimed at changes in the rules that generate alarms, i.e. changes in the reporting system itself. However, the operator of the automation system is usually completely unable to intervene in the alarm system or its rules. Even if the operator has such intervention possibilities, such intervention is complex, error-prone, and risks (possibly safety-related) backlash to the rest of the automation system. This also applies in particular when the notification server is remote from the facility on the cloud and is connected to the automation system via a public communication network. Summary of the invention
[0012] It is therefore an object of the present invention to provide a method and a system for managing messages of an automation system, in particular of an industrial plant, with which the above-mentioned disadvantages and problems can be avoided.
[0013] This object is achieved by a method and a system for managing messages of an automation system, in particular an industrial plant, according to claims 1 and 8. An evaluation method and an evaluation system therefor are the subject matter of claims 6 and 13. A program and a computer program product are the subject matter of claims 15 and 16. Advantageous embodiments are respectively the subject matter of the dependent claims.
[0014] The method according to the invention is based on an automation system which
[0015] - includes a reporting system for generating messages, and
[0016] - at least one first human-machine interface, on which the message is displayed, wherein
[0017] - the reporting system is communicatively connected to the human-machine interface, in particular via a communication network of the automation system,
[0018] at least a first category and a second category of messages are defined, and the messages of the automation system are respectively associated with at least one of the two categories,
[0019] - applying a function at least to input data, which represents a message which is associated with the second category (optionally also additionally associated with the first category) and which is displayed on the first human-machine interface, wherein the function is trained by an algorithm from the field of machine learning, and wherein output data are generated thereby, which represent a message of the automation system which is associated with the first category,
[0020] - The message represented by said output data is displayed on a second human-machine interface, wherein the second human-machine interface is communicatively separate from the reporting system.
[0021] In a first step, the invention is based on the knowledge that machine learning can be used, on the one hand, to advantageously filter messages of a first category (i.e., at least a set of messages of the second category and optionally also messages of the first category) from a large number of messages. In this case, the output data represent one of the messages that has also been represented by the input data within a defined time period. On the other hand, machine learning can be used to predict messages of the first category based on at least the set of messages of the second category and optionally additionally also messages of the first category. In this case, the output data typically represent messages that have not yet been represented by the input data within a defined time period. However, the output messages can also represent messages that have been represented by the input data within a defined time period. With the help of the filtered or predicted messages, the attention of personnel can be diverted to messages that are important under defined or definable criteria, or even their occurrence can be predicted.
[0022] Preferably, in the case of predicting a message, the output data includes a time point or a time period when the message appears and a probability of the message appearing at the time point or in the time period.
[0023] Different categories can preferably express: the criticality of the corresponding class of messages to the failure safety of the facility. However, different categories may also involve other aspects, such as production cost, production volume, production quality, raw material consumption, electricity consumption, emissions, etc.
[0024] In this case, the function can be continuously machine trained, taking into account the current state of the facility. As a result, the function and thus the filtering and / or prediction of messages can be continuously adapted to changes in the facility (e.g. due to wear or performance changes of facility components, changes in the quality / quantity of raw materials, changes in the operating mode of the facility), thus achieving high-quality filtering and / or prediction.
[0025] In this case, there is no intervention in the reporting system and no need for intervention, but only the messages represented by the output data are displayed in addition to the messages of the reporting system. By displaying the messages on a separate, additional second human-machine interface and decoupling the second human-machine interface from the reporting system or the automation system communication, undesired safety-related reactions to the automation system can be avoided.
[0026] The communication separation can be realized particularly easily in that the input data of the function are received from a sender and the output data of the function are sent to a receiver which is different from the sender.
[0027] Functions trained by algorithms from the field of machine learning are provided by:
[0028] - receiving input training data representing messages for the automation system,
[0029] - receiving output training data representing a message for the automation system, wherein the output training data is associated with the input training data,
[0030] - Train a function based on input training data and output training data.
[0031] By training the function based on training data, the function can adapt to new situations and identify and infer patterns in messages. The parameters of the function can be adapted by training. Here, for example, supervised training, semi-supervised training, unsupervised training, reinforcement learning and / or active training can be used. In addition, representation learning (an alternative expression is "feature learning") can be used. In particular, the parameters of the function can be iteratively adapted by multiple training steps.
[0032] The function is preferably based on mutual correlation of messages, neural networks (preferably long / short term memory - LSTM) or association rules. Preferably, a combination of the methods is used, wherein the result is obtained, for example, by the sum of the results of all methods, by the majority results, by individual results, etc.
[0033] According to a further advantageous embodiment, information about the quality of the message represented by the output data is detected by the second human-machine interface and used for training the function. As a result, the function can be optimized particularly quickly.
[0034] Furthermore, a selection from a plurality of predefined categorizations can also be provided to the personnel of the device or facility via the second human-machine interface, and information about the selection can then be detected via the second human-machine interface. For example, possible categorizations may be, for example, failure safety, emissions, electricity consumption, production costs, etc.
[0035] It is advantageous to arrange the function remotely from the reporting system, in particular on a cloud-based computer system. There, training data representing messages from other automation systems can then also be detected and used to train the function. This can improve the quality of the function and thus the quality of the output data.
[0036] The system according to the invention is for managing messages of an automation system, in particular an industrial installation, wherein the automation system
[0037] - includes a reporting system for generating messages, and
[0038] - at least one first human-machine interface for displaying messages,
[0039] - wherein the reporting system is communicatively connected to the human-machine interface, in particular via a communication network of the automation system,
[0040] and wherein at least a first category and a second category of messages are defined, and wherein the messages of the automation system are respectively associated with at least one of the two categories, the system comprising
[0041] an evaluation unit for applying a function at least to input data which are displayed on a first human-machine interface and which represent messages associated with the second category (optionally also with the first category in addition), wherein the function is trained by an algorithm from the field of machine learning, and wherein output data are generated thereby, which represent messages of the automation system associated with the first category,
[0042] - a second human-machine interface for displaying the message represented by the output data, wherein the second human-machine interface is communicatively separate from the reporting system.
[0043] The system according to the invention advantageously comprises
[0044] - a first interface for receiving input data from a transmitter,
[0045] - a second interface for providing the output data to a receiver, the receiver being different from the sender.
[0046] According to another advantageous embodiment, the system according to the invention comprises a training unit for providing a trained function, the training unit comprising
[0047] - a first training interface, said first training structure being adapted to receive input training data representing a message for an automation system,
[0048] - a second training interface for receiving output training data representing a message for the automation system, wherein the output training data is associated with the input training data,
[0049] - a training computation unit for training a function based on input training data and output training data.
[0050] - A third training interface, the third training interface being used to provide a trained function.
[0051] According to a further advantageous embodiment, the second human-machine interface is used to detect information about the quality of the message represented by the output data and to transmit this information to the training unit.
[0052] It is particularly advantageous if the evaluation unit and preferably also the training unit are arranged remotely from the reporting system, in particular on a cloud-based computer system.
[0053] The effects and advantages mentioned for the method according to the invention and its advantageous embodiments apply correspondingly to the system according to the invention and its advantageous embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0054] The invention and the advantageous design solutions of the invention according to the features of the dependent claims are explained in more detail below based on the embodiments in the drawings. Here, the components corresponding to each other are respectively provided with the same reference numerals. The drawings show:
[0055] Figure 1 A simplified diagram showing the principle structure of the system according to the invention,
[0056] Figure 2 Show Figure 1 A detailed view of the cloud-based assessment system,
[0057] Figure 3 The method flow according to the present invention is shown. DETAILED DESCRIPTION
[0058] Figure 1 A simplified diagram of an industrial facility 1 with an automation system 2 is shown. Such facilities are used in various industrial sectors, for example in energy generation, process industries (e.g. paper, chemicals, pharmaceuticals, metals, oil and gas) and discrete manufacturing. The actual industry-specific processes 3, for example energy generation or production processes, are controlled and / or regulated by the automation system 2. For this purpose, the automation system 2 comprises one or more industrial control devices, here control devices 4, and at least one automation server 5.
[0059] Each control device 4 then controls the operation of each sub-area of the process 3 according to the operating state of the process. For this purpose, the process 3 includes actuators 6 that can be controlled by the control device. Here, they can be individual actuators (e.g. motors, pumps, valves, switches) or groups of these actuators or all sections of a facility. The process also includes sensors 7, which provide the control device 4 with actual values of process variables (e.g. temperature, pressure, speed).
[0060] The communication network of the facility 1 comprises, at a higher level, a facility network 11 and a control device network 12 (often also referred to as a “system network”), wherein the automation server 5 communicates with the human-machine interface (HMI) 10 via the facility network, and the control devices 4 are communicatively connected to each other and to the automation server 5 via the control device network. The connection of the control device 4 to the actuator 6 and the sensor 7 can be via discrete signal lines 13 or via a fieldbus connection. The human-machine interface (HMI) 10 is usually designed as an operating and observation station and is arranged in a control room 13 of the facility 1.
[0061] The automation server 5 can be, for example, a so-called "operating system server" or "application server" in which one or more plant-specific applications 8 are stored and executed during operation of the plant 1. The applications are used, for example, to configure the control devices 4 in the plant 1, to detect and execute operator activities at the human-machine interface (HMI) 10 (e.g. setting or changing target values of process variables) or to generate messages for the plant personnel and display them on the human-machine interface (HMI) 10.
[0062] An automation system 2 without field devices, ie without actuators 5 and sensors 6 , is often referred to as a “process control system”.
[0063] In order to generate the messages M, the automation server 5 comprises a reporting system 9 which is likewise implemented here by the application 8 .
[0064] The message M usually relates to a specific event in the automation system 2. For example, a message can be an alarm message or a notification message. An alarm message is a message that an error or an abnormal event has occurred and personnel must take immediate action. For example, a notification message can indicate that maintenance is required. In this case, an event message can include text and other information, such as the time and location of the message generation.
[0065] In order to direct the attention of the personnel to messages or their predictions that are important under defined or definable criteria, the automation system additionally comprises a connection server 14, an evaluation system 20 and an additional human-machine interface 30, which is preferably designed as an operator station. Alternatively or in addition, a mobile human-machine interface 31, such as a smartphone, a laptop or a tablet, can also be present. The human-machine interfaces 30, 31 are separated from the communication network 11, i.e., communication from the communication network 11 to the human-machine interfaces 30, 31 or vice versa is not possible. In particular, the human-machine interface 30 can communicate with the evaluation system 20 via the office network 15.
[0066] Firewalls 16 , 17 may also be provided between the connection server 14 and the evaluation system 20 , and between the evaluation system 20 and the human-machine interfaces 30 , 31 .
[0067] like Figure 2 As shown in detail, the cloud-based evaluation system 20 includes a first communication interface in the form of a cloud Internet gateway 21, a data stream processing unit 22, a memory 23 for actual data (real-time data), a memory 24 for historical data, a training unit 25, an evaluation unit 26, a network application 27, a second communication interface 28 for wired communication, and another second communication interface 29 for wireless communication in addition or in addition.
[0068] At least a first category A and a second category B of messages are defined in the reporting system 9 , and the messages of the automation system are respectively assigned to at least one of the two categories A, B.
[0069] During the operation of the automation system, a plurality of class B messages (the data of the messages are represented by B1, B2, B3, etc.) and a slightly smaller number of class A messages (the data of the messages are represented by A1, A2, etc.) are generated by the reporting system 9 and displayed on the human-machine interface 10 via the communication network 11 (see Figure 3 In addition, the data representing the message are transmitted to the evaluation system 20 via the connection server 14 (see Figure 3 In step 52 of FIG. 1 ), the cloud internet gateway 21 is used to receive the data from the connection server 14 as a sender.
[0070] The gateway 21 forwards the data to a data stream processing unit 22. The data stream processing unit stores the data in a memory 23 for subsequent (almost) real-time processing and for archiving in a memory 24 (see Figure 3 53 in step 5).
[0071] The data in the memory 23 are forwarded as input data to the evaluation unit 26 without any significant delay.
[0072] The evaluation unit 26 is used to apply a function to the input data, wherein the function is trained by an algorithm from the field of machine learning and wherein output data are generated thereby, which represent messages of the automation system associated with the first category A (see Figure 3 54 in step 54).
[0073] The function trained by machine learning can be usefully used on the one hand to filter out the first category messages from the total set of all messages (i.e. messages of the first category and the second category). Here, this is exemplarily message A1. In this case, the output data represents one of the messages that has been represented by the input data within a defined time period.
[0074] On the other hand, machine learning can be used to predict messages of the first category based at least on messages of the second category and optionally additionally also on messages of the first category. In this case, the output data typically represent messages that have not been represented by the input data within a defined time period. However, the output data can also represent messages that have been represented by the input data within a defined time period. Here, this is exemplarily message A2.
[0075] In both cases, the attention of the personnel can be directed to the messages A1 and A2 which are important from a defined perspective, or their occurrence can even be predicted in order to initiate countermeasures in good time.
[0076] The output data A1 and A2 are transmitted via the network application 27 or via the communication interface 28 to the human-machine interface 30 and / or via the communication interface 29 to the human-machine interface 31 and displayed there (see Figure 3 55 in step 5). The interface 28, 29 is therefore used to provide the output data to a receiver which is different from the sender of the input data.
[0077] Thus, in addition to all other messages of the reporting system 9, the message represented by the output data A1 or A2 is also displayed in the control room 13. By displaying the message on a separate, additional second human-machine interface 30, 31, the attention of the personnel is particularly diverted to the message. The communication separation of the second human-machine interface 30, 31 from the reporting system 9 or the automation system 2 can avoid undesired, safety-related reactions to the automation system 2. In this case, no intervention in the reporting system 9 can take place and is not required.
[0078] In order to provide a trained function, the evaluation system comprises a training unit 25 comprising:
[0079] a first training interface 41 for receiving input training data ET representing messages of the automation system 2 ,
[0080] a second training interface 42 for receiving output training data AT representing messages of the automation system 2 , wherein the output training data are associated with input training data ET,
[0081] a training calculation unit 43 for training a function 45 based on input training data ET and output training data AT,
[0082] A third training interface 44 for providing a trained function 45 to the evaluation unit 26 .
[0083] In this case, the input training data ET are received by the data memory 24. The output training data AT are received by the expert. The training of the function 45 is based on the input training data ET and the output training data AT.
[0084] Training data representing messages of other automation systems can also be recorded in the memory 24 and used to train the function 45. The quality of the function 45 and thus of the output data can thereby be improved.
[0085] By training the function 45 based on the training data, the function can adapt to new situations and identify and infer patterns in the message sequence. In this case, the parameters of the function can be adapted by training. In this case, it can use, for example, supervised training, semi-supervised training, unsupervised training, reinforcement learning and / or active training. In addition, representation learning can be used (an alternative expression is "feature learning"). In particular, the parameters of the function can be iteratively adapted by multiple training steps.
[0086] This function 45 is preferably based on cross-correlation of messages, neural networks (preferably long / short term memory - LSTM) or association rules. Preferably, a combination of the methods is used, wherein the result is obtained, for example, by the sum of the results of all methods, by the majority results, by individual results, etc.
[0087] From the second human-machine interface 30 or 31, the personnel of the installation can also detect information Q about the quality of the message represented by the output data and transmit it to the evaluation system 20 and use it there for training the function 45. This makes it possible to optimize the function 45 particularly quickly. For example, the information Q can relate to whether the predicted message occurred at all and, if it did, whether the time or time period specified for its occurrence was correctly predicted.
[0088] In the case of predicting a message represented by the output data A2, the output data preferably includes a point in time or a time period at which the message appears and a probability of the message appearing at the point in time or in the time period.
[0089] Different categories can preferably express the criticality of the corresponding class of messages to the failure safety of the facility. However, different categories may also involve other aspects, such as production cost, production volume, production quality, raw material consumption, electricity consumption, emissions, etc.
[0090] Advantageously, the personnel of the installation 1 are also offered a selection from a plurality of predefined categorizations via the second human-machine interface 30 , and information about the selection can then be detected via the second human-machine interface 30 .
Claims
1. A method for managing messages (M) of an automation system (2), wherein the automation system (2) comprises: - a reporting system (9) for generating messages (M), and - at least one first human-machine interface (10) on which the message (M) is displayed, wherein - the reporting system (9) is in communication connection with the first human-machine interface (10), - defining at least a first category (A) and a second category (B) of messages, and wherein the messages (2) of the automation system (2) are respectively associated with at least one of the first category (A) and the second category (B), - applying a function (45) at least to input data (B1, B2, B3, A1) representing messages (M) associated with the second category (B) and displayed on the first human-machine interface (10), wherein the function (45) is trained by an algorithm from the field of machine learning and wherein output data (A1, A2) are generated thereby, representing messages (M) of the automation system (2) associated with the first category (A), - displaying said message (M) represented by said output data (A1, A2) on a second human-machine interface (30), wherein said second human-machine interface (30) is communicatively separate from said reporting system (9), Therein, information (Q) about the quality of a message represented by the output data (A1, A2) is detected by the second human-machine interface (30) and used for training the function (45), wherein the message (M) is an alarm message or a notification message.
2. The method according to claim 1, characterized in that The method is used for managing messages of an industrial facility (1).
3. The method according to claim 1, characterized in that The reporting system (9) is communicatively connected to the first human-machine interface (10) via a communication network (11) of the automation system (2).
4. The method according to any one of claims 1 to 3, characterized in that The input data (B1, B2, B3, A1) of the function (45) are received by a transmitter (14), and the output data (A1, A2) of the function (45) are sent to a receiver (30) which is different from the transmitter (14).
5. The method according to any one of claims 1 to 3, characterized in that The function (45) trained by an algorithm from the field of machine learning is provided in the following way: - receiving input training data (ET) representing messages for said automated system (2), - receiving output training data (AT) representing a message of said automation system, wherein said output training data (AT) is related to said input training data (ET), - training said function (45) based on said input training data and said output training data.
6. The method according to any one of claims 1 to 3, characterized in that The function (45) is located remotely from the reporting system (9).
7. The method according to claim 6, characterized in that The function (45) is provided on a cloud-based computer system (40).
8. A method according to any one of claims 1 to 3 for managing messages (M) of an automation system (2), wherein at least a first category (A) and a second category (B) of the messages (M) are defined, and wherein the messages (M) of the automation system (2) are respectively associated with at least one of the first category (A) and the second category (B), the method comprising: - receiving input data (B1, B2, B3, A1) from a sender (14), wherein said input data represents at least a message (M) associated with said second category (A, B), - applying a function (45) trained by an algorithm from the field of machine learning to the input data (B1, B2, B3, A1), wherein output data (A1, A2) are generated thereby, the output data representing messages (M) of the automation system (2) associated with the first category (A), - providing the output data (A1, A2) to a receiver (30), wherein the receiver (30) is different from the transmitter (14).
9. The method according to claim 8, characterized in that The function (45) trained by an algorithm from the field of machine learning is provided in the following way: - receiving input training data (ET) representing messages (M) of said automation system (2), - receiving output training data (AT) representing a message (M) of said automation system, wherein said output training data (AT) is related to said input training data (ET), - training said function (45) based on said input training data (ET) and said output training data (AT).
10. A system for managing messages (M) of an automation system (2), wherein the automation system (2) comprises: - a reporting system (9) for generating a message (M) and - at least one first human-machine interface (10) for displaying said message (M), - wherein the reporting system (9) is communicatively connected to the first human-machine interface (10), and wherein at least a first category (A) and a second category (B) of messages (M) are defined, and wherein the messages (2) of the automation system (2) are respectively associated with at least one of the first category (A) and the second category (B), It is characterized by having an evaluation unit (26) for applying a function (45) at least to input data (B1, B2, B3, A1) which are displayed on the first human-machine interface (10) and which represent messages (M) associated with the second category (B), wherein the function (45) is trained by an algorithm from the field of machine learning and wherein output data (A1, A2) are generated thereby, which represent messages (M) of the automation system (2) associated with the first category (A), - a second human-machine interface (30) for displaying the message (M) represented by the output data (A1, A2), wherein the second human-machine interface (30) is communicatively separate from the reporting system (9), The second human-machine interface (30) is used to detect information (Q) about the quality of the message (M) represented by the output data (A1, A2) and transmit the information to the training unit (25), wherein the message (M) is an alarm message or a notification message.
11. The system according to claim 10, characterized in that The system is used to manage messages (M) of an industrial facility (1).
12. The system according to claim 10, characterized in that The reporting system (9) is communicatively connected to the first human-machine interface (10) via a communication network (11) of the automation system (2).
13. The system according to any one of claims 10 to 12, characterized in that include - a first interface (21) for receiving the input data (B1, B2, B3, A1) from a transmitter (14), - a second interface (28) for providing the output data (A1, A2) to a receiver (30), the receiver being different from the transmitter (14).
14. The system according to any one of claims 10 to 12, characterized in that A training unit (25) for providing a trained function (45) is provided, the training unit comprising - a first training interface (41) for receiving input training data (ET) representing messages (M) of the automation system (2), - a second training interface (42) for receiving output training data (AT) representing a message (M) of the automation system (2), wherein the output training data (AT) is related to the input training data (ET), - a training calculation unit (43) for training the function (45) based on the input training data (ET) and the output training data (AT), - a third training interface (44) for providing a trained function (45).
15. The system according to any one of claims 10 to 12, characterized in that The evaluation unit (26) and the training unit (25) are arranged remotely from the reporting system (9).
16. The system according to claim 15, characterized in that The evaluation unit (26) and the training unit (25) are arranged on a cloud-based computer system (40).
17. A computer program product comprising instructions for causing the computer to execute the method according to claim 8 or 9.
18. A computer-readable medium comprising instructions, which, when executed on a computer, cause the computer to perform the method according to claim 8 or 9.
