Improvement of Control Strategy for Distributed Control System Based on Operator Action

Abstract

A computer-implemented method (100a) for modifying and / or extending an engineering tool (2) configured to generate application code (3) that, when executed on one or more controllers (4) in a distributed control system (5) of an industrial plant (1), causes the industrial plant (1) to be controlled according to a control strategy implemented in the application code (3), the method comprising: - obtaining (110) state variables (6) characterizing an operating state (1a) of at least one industrial plant (1); - obtaining (120) a set of interaction events (7) of at least one plant operator interacting with the distributed control system (5) of the industrial plant (1) via a human-machine interface; - receiving as input data the interaction events (7), the state variables (6) and optionally the distributed control system (5); ), determining (130) whether the one or more interaction events (7) indicate that the plant operator is performing a task that is not adequately covered by the current design of the distributed control system (5); and if this determination is affirmative, mapping (140) the input data to modifications and / or extensions (2a) for the engineering tool (2) that generated the application code (3) for the distributed control system (5) such that when the application code (3) is regenerated by the modified and / or extended engineering tool (2) and executed in the distributed control system (5), the plant operator manually interacts with the distributed control system less frequently and / or for a shorter period of time.

Description

Technical Field

[0001] The present invention relates to the automatic control of industrial plants, in particular plants having a distributed control system.

Background Art

[0002] The automatic control of industrial plants is usually carried out according to an engineered control strategy designed to operate the plant in an efficient manner. However, such an engineered control strategy cannot predict all situations that may occur during the operation of the plant. In such situations, the plant operator can intervene manually and take control of the plant or a part thereof, invalidating the designed control strategy.

[0003] WO2012 / 142 353A1 discloses a method for monitoring a process control system. The method includes visualizing key performance indicators (KPIs) of a plant that are unsatisfactory.

[0004] Object of the Invention The object of the present invention is to reduce the need for manual intervention during the operation of an industrial plant and / or the amount of time that the plant operator has to spend on these interventions.

[0005] This object is achieved by two computer-implemented methods for modifying and / or extending an engineering tool for a distributed control system as described in the independent claims, and by another computer-implemented method for training a machine learning model as described in another independent claim.

[0006] Disclosure of the Invention The present invention provides two computer-implemented methods for modifying and / or extending an engineering tool. The engineering tool is configured to generate application code. When this application code is executed on one or more controllers within a distributed control system of an industrial plant, the industrial plant is controlled according to a control strategy implemented within the application code.

[0007] Both methods generate a modification and / or extension of an engineering tool that generated application code for a distributed control system. When the application code is regenerated by the modified and / or extended engineering tool and executed within the distributed control system, the effect of the modification is likely that the plant operator will interact less frequently and / or for less time manually with the distributed control system.

[0008] The first method obtains this modification and / or extension based on past interaction events of a plant operator who interacts with the distributed control system of the plant via a human-machine interface. The second method obtains the modification and / or extension based on a prediction of interaction events that a plant operator is likely to initiate based on the current state of the plant. This prediction is obtained from a trained machine learning model. The first method specifically facilitates obtaining the modification and / or extension from a specific known behavior pattern of the plant operator, whereas the second method specifically facilitates obtaining the modification and / or extension from scratch without prior knowledge of the behavior pattern.

[0009] In the process of the first method, state variables characterizing the operating state of at least one industrial plant are obtained. Also, a set of interaction events of at least one plant operator who interacts with the distributed control system of the industrial plant via a human-machine interface is obtained.

[0010] As input data, based at least in part on dialogue events, state variables, and optionally predetermined design information of a distributed control system, it is determined whether one or more dialogue events indicate that the plant operator is performing tasks that are not adequately covered by the design of the distributed control system. If this determination is affirmative, the input data is mapped to the required modifications and / or extensions requested for the engineering tool that generated the application code for the distributed control system.

[0011] The control system is designed based on requirements specifications during the design phase. However, in this design phase, it is not possible to predict all situations that occur during the operation of the plant. The human plant operator remains within the control loop as an interactive part to fill any gaps in the designed control strategy. If rare situations that are not foreseen in the control strategy and have a very low probability of recurrence are addressed by the intervention of the plant operator, this is a better way than modifying the control strategy to address this situation as well. However, if the situation is likely to occur again, it is better to modify the design of the plant so as to free the plant operator from having to perform the same or substantially similar interventions multiple times. If it is modified in an automated design way, this has several advantages: · In all situations that can be automatically addressed, the correct operation no longer depends on the plant operator making the correct intervention at the right time, so the operation of the plant becomes safer. · Freeing the plant operator from routine interventions frees up the time that would otherwise have been spent on these routine interventions for problem-solving tasks that cannot be processed by machines. · The knowledge of the plant operator is captured and documented in a form that can also be transferred to similar plants. · It is possible to quantitatively measure how well the design of the plant covers the actual operating situations that occur within the plant.

[0012] Tasks that are not adequately covered by the current design of the distributed control system can be specified as patterns in any suitable way. In particularly advantageous embodiments, tasks that are not adequately covered by the current design of the distributed control system specifically include the following: · Manually performing solutions to operating problems not covered by the current design of the distributed control system, and / or · Repeatedly performing one or more actions starting from equal or substantially similar operating states, and / or · Accessing at least one function that requires at least a threshold number of steps to access at least at a threshold frequency.

[0013] For example, in a waste incineration plant, the combustion process is highly dependent on the composition of the waste. The plant is designed for a specific range of compositions, but a significant change in composition beyond this design may occur suddenly. For example, paper, plastic, or other materials may be normal components of household waste, so the plant design may assume that this component is always present. However, updated environmental regulations may suddenly stipulate that paper or plastic should be collected in a separate bin for recycling, and suddenly this component will disappear from household waste. Then, the plant operator may notice that the combustion has suddenly deteriorated and can find out how to improve the combustion by finely adjusting the air flow in the furnace and the agitation of the waste. If this solution is incorporated into updated engineering tools and the plant design is updated accordingly, this solution can be reused whenever waste of a similar composition is supplied to the plant in the future.

[0014] Situations where plant operators manually execute solutions to operating problems may be detected, for example, by pattern recognition in state variables and interaction event data. For example, · The state variables indicate a problematic or near-optimal state of the plant, ·A sequence of dialogue events is continuously detected, ·Accordingly, if there is a pattern in which the state variable indicates an improvement in the state of the plant, It can be inferred that the intervention by the plant operator has solved the operation problem in the plant.

[0015] When one or more actions start from an equal or substantially similar operating state and are repeatedly executed, in response to the same or substantially similar operating state occurring again, the engineering tool may be modified so that one or more actions will be automatically executed in the future. This liberates the plant operator from repetitive manual work, similar to the macro recorder in a word processing program.

[0016] If at least some functions require at least a threshold number of steps to be accessed at least at a threshold frequency, the engineering tool may be modified so that the functions can be accessed with a smaller number of steps. For example, the arrangement of controls and displays in a human-machine interface may be so large that it may not fit on the screen and may therefore be split into several pages. The human-machine interface may be on the first page by default, and in order to access a function on the fifth page, the plant operator may have to flip through the second, third, and fourth pages and then the fifth page. The initial assignment of controls and displays to different pages may be motivated by the estimated probability that the plant operator may need to access each control or display. However, in some cases of the plant, some controls and displays may become more important than in other cases of the plant. This may even change during execution. For example, if multiple residential neighbors of the plant complain to the local authority about the noise, smell, or other nuisance emitted from the plant, the plant operator may be forced to reduce this nuisance. Each sensor reading may have to be monitored more frequently and it may therefore be appropriate to move them from page 5 to page 1, where they are visible on the screen most of the time.

[0017] In these examples, the situations in which the modification of the engineering tool is appropriate can be detected by searching for or monitoring specific patterns in the data from the plant. However, there may be more cases where it becomes apparent that there are "gaps" in the previous design of the plant that must be filled by the intervention of the plant operator. Accordingly, the present invention also provides a second computer-implemented method for modifying and / or extending an engineering tool.

[0018] Similar to the first method, in the process of this second method, state variables characterizing the operating state of at least one industrial plant are obtained. Then, using at least one trained machine learning model, based on the state variables, predict one or more interaction events that are likely to be initiated by at least one plant operator on the distributed control system via the human-machine interface in response to the operating state. The one or more predicted interaction events are mapped to modifications and / or extensions for an engineering tool that generated the application code for the distributed control system. Similar to the first method, the modifications are regenerated by the engineering tool with the modified and / or extended application code and, when executed within the distributed control system, are configured such that the frequency with which the plant operator manually interacts with the distributed control system is reduced and / or the time spent interacting with the distributed control system is shortened. The mapping may be performed, for example, by a machine learning model that also predicts interaction events, by another machine learning model, or by any other suitable method.

[0019] Generating such modifications and / or extensions for the engineering tool does not require specific patterns for the operating situation and / or for operator intervention to be known in advance. Rather, the engineering tool can learn dynamically from the plant operator and can adapt to new classes of situations that could not have been foreseen at the initial design time.

[0020] For example, a waste incineration plant that previously operated optimally with a specific design may be later connected to a district heating network so that the heat generated by the plant can be used for other purposes. Subsequently, in order to maintain the reliability of district heating, it may be necessary to keep the amount and temperature of the heat delivered to the district heating network within a predetermined range. These new goals can at least partially replace the previous goals. For example, if the calorific value of the waste decreases, it may be necessary to ignite a fuel burner to supplement the combustion of the waste in order to maintain heat supply. According to the conventional design, the burner is not ignited because fuel increases the operating cost unless it is necessary for the continuous operation of the waste incineration plant. The training of the machine learning model can capture the fact that manual intervention is required in the increasing situation after the connection to the district heating network. After training, the machine learning model can predict in which operating situations manual intervention is likely to occur. Then, the engineering tool may be modified according to the newly generated application code so that the intervention to maintain heat delivery may be automatically started in the future.

[0021] Both methods may be effective simultaneously in one and the same plant. That is, if it is detected that the engineering tool requires modification and / or extension at a certain location according to a known pattern, this modification and / or extension may be determined and implemented. In addition, the machine learning model may be used to generate modifications and / or extensions from interventions that do not conform to previously known patterns.

[0022] The input data used in either method can further include, for example, one or more of the following: · Alarms and events reported by the distributed control system, · The topology model of the industrial plant, · The layout of the human-machine interface of the distributed control system, and · The control logic of the distributed control system. If the input data is more detailed, a more sophisticated pattern for detecting "gaps" in the current design can be used and / or a machine learning model can more accurately predict plant operator intervention.

[0023] In a further particularly advantageous embodiment, when input data and / or interaction events are mapped to a modification and / or extension of an engineering tool, a function within a given control library may be determined to achieve a result substantially similar to the result of the detected and / or predicted action or sequence of actions. In this case, the detected and / or predicted action or sequence of actions may be replaced with a call to the determined function within the control library when generating the modification and / or extension for the engineering tool. In this way, the knowledge already condensed in the control library can be used in appropriate situations. For example, a plant operator can follow a specific manual protocol for setting the temperature inside a container to a new target temperature without knowing that an automation protocol is already defined in the control library for such a standard action.

[0024] In a further particularly advantageous embodiment, the modification and / or extension is such that, when the application code is regenerated by the modified engineering tool and executed in a distributed control system, in the human-machine interface of the distributed control system, · a new control element appears such that a series of actions previously continuously repeated by a plant operator are executed when this new control element is actuated, and / or, · a control element that previously required a first number of steps to access is configured to move within the human-machine interface so as to require a second, fewer number of steps to access. is configured to result in. This enables a plant operator to achieve the same intervention with less interaction between the plant operator and the human-machine interface of the distributed control system.

[0025] As described above, in a further particularly advantageous embodiment, the modifications and / or extensions for the engineering tool extension are regenerated by the engineering tool with modified application code and, when executed within the distributed control system, one or more actions previously performed by the plant operator are repeatedly initiated from an equal or substantially similar operating state and automatically executed in response to the occurrence of a particular operating state. This can free the plant operator from routine interventions and enable the plant operator to instead focus on problem-solving tasks.

[0026] In a further advantageous embodiment, an engineering tool configured to assemble a distributed control system from building blocks within a predefined catalog is selected. At least one such building block is a programmable logic controller (PLC). The engineering tool is configured to generate application code including control code for this PLC. When such an engineering tool is modified and / or extended, the assembly of the plant from the building blocks may remain unchanged, but the control code for the PLC may be recompiled, thereby upgrading it with new functionality.

[0027] Accordingly, either way may further include regenerating application code for the distributed control system by the modified and / or extended engineering tool. The regenerated application code may then be executed in the distributed control system. In this way, the industrial plant is controlled according to the modified and / or extended control strategy implemented in the regenerated application code.

[0028] Optionally, prior to regenerating the application code, the control engineer may be prompted to approve modifications and / or extensions to the engineering tool. In this way, the control engineer knows what will change when the application code is next regenerated. Also, the control engineer can check whether the proposed changes violate other constraints.

[0029] The present invention also provides a computer-implemented method for training at least one machine learning model for use in the second method described above.

[0030] The method begins with a record of training input data having state variables characterizing the operating state of at least one industrial plant. In response to the operating state, a label is provided regarding which interaction events at least one plant operator initiated in a distributed control system. The machine learning model to be trained maps the record of training input data to predictions of one or more interaction events that at least one plant operator initiates in response to the operating state within the training input data.

[0031] By means of a predefined cost function, how well the predictions by the machine learning model correspond to the labels of each record of the training input data is evaluated. The parameters characterizing the behavior of the machine learning model are optimized towards the goal that this results in a better evaluation by the cost function when further records of the training input data are processed by the machine learning model. In this way, the prediction of plant operator intervention becomes increasingly accurate as the training progresses.

[0032] For labeling the operating state by the dialogue events, the actual dialogue events collected from the plant may be filtered, for example, according to whether these dialogue events have led to the success or failure of the intervention by the plant operator. For example, the dialogue events may be graded according to any suitable metric regarding how beneficial they have proven to be for a given purpose. The actual dialogue events may be weighted, for example, according to these grades. The dialogue events may be excluded from being included in the label, for example, if their grades fall below a certain threshold.

[0033] In a particularly advantageous embodiment, the records of the training input data are collected from a plurality of industrial plants. In this way, the operator knowledge may be transferred at least among similar plants. In many types of plants, such as waste incineration plants, the examples of these plants differ to some extent from each other, but the basic functions are the same in all examples. Different examples of waste incineration plants may be supplied with waste of different compositions. One waste incineration plant may be connected to a district heating network, while the other may not. However, all examples have in common that there is a specific type of furnace for the combustion of waste, and this combustion is controlled by operating a specific set of parameters.

[0034] In any given case of an industrial plant, the training does not have to start from scratch. Rather, the machine learning model may first receive general training and then be further trained in a more specific way for a particular example of the plant. In this way, when a large number of different examples of the plant are deployed, there is no need to repeat the training time and the general part of the time again.

[0035] The computer implementation of the above method means that the method may be embodied in a computer program. Accordingly, the present invention also provides a computer program having machine-readable instructions that cause one or more computers to execute one of the above methods when executed by the one or more computers. The present invention also provides a non-transitory machine-readable storage medium and / or a download product having the computer program. The download product is a product that may be sold in an online store for immediate fulfillment by download. The present invention also provides one or more computers having the computer program and / or having the non-transitory machine-readable storage medium and / or the download product.

[0036] Hereinafter, the present invention will be illustrated with reference to the drawings, but it does not limit the scope of the present invention.

Brief Description of the Drawings

[0037]

Figure 1

Figure 2a

Figure 2b

Figure 3

Modes for Carrying Out the Invention

[0038] FIG. 1 is a schematic flowchart of an embodiment of methods 100a, 100b for modifying and / or extending engineering tool 2. Both methods 100a, 100b start by obtaining state variable 6 that characterizes the operating state 1a of at least one industrial plant 1 in step 110.

[0039] In the process of method 110a, at step 120, a set of interaction events 7 of at least one plant operator who interacts with the distributed control system 5 of industrial plant 1 via a human-machine interface is obtained. At step 130, as input data, based at least in part on the interaction events 7, the state variables 6, and optionally predetermined design information of the distributed control system 5, it is determined whether one or more interaction events 7 indicate that the plant operator is performing tasks that are not adequately covered by the current design of the distributed control system 5. If so (truth value 1), at step 140, the input data is mapped to a modification and / or extension 2a for the engineering tool 2 that generated the application code 3 for the distributed control system 5.

[0040] In the process of method 110b, at step 150, based on the state variables 6, one or more interaction events 7 that at least one plant operator is likely to initiate on the distributed control system 5 via a human-machine interface in response to the operating state 1a of plant 1 are predicted using at least one machine learning model 8. At step 160, the one or more predicted interaction events 7 are mapped to a modification and / or extension 2a for the engineering tool 2 that generated the application code 3 for the distributed control system 5.

[0041] During the mapping 140, 160 performed in either method 110a or 100b, according to block 141 (161 respectively), a function within a predetermined control library that achieves a result substantially similar to the result of the detected and / or predicted action or sequence of actions can be determined. According to each of blocks 142, 162, this action or sequence of actions may be replaced with a call to the determined function within the control library. In this way, if the plant operator manually performs something for which a library function is already available, the tested library function is used.

[0042] According to block 105, for both methods 100a and 100b, engineering tool 2 may be selected, and engineering tool 2 · Assemble a distributed control system (5) from building blocks within a predetermined catalog, where at least one such building block is a programmable logic controller (PLC), · Is configured to generate application code including control code for this PLC.

[0043] In the process of both methods 100a and 100b, a modification 2a of engineering tool 2 may be presented to a control operator for approval at step 170. If approval is given (truth value 1), engineering tool 2 can regenerate application code 3 for distributed control system 5 at step 180. At step 190, this regenerated application code 3 may be executed in distributed control system 5. As a result, industrial plant 1 is controlled according to an updated control strategy implemented in the regenerated application code 3.

[0044] This information flow is described in more detail in FIG. 2a for method 100a and in FIG. 2b for method 100b.

[0045] According to Figure 2a, state variables 6 are obtained from the controller 4 within the distributed control system 5 and from this control system 5 as a whole. Also, interaction events 7 of at least one plant operator interacting with the control system 5 are obtained. Based on the state variables 6 and one or more interaction events 7, in step 130 of method 100a, it is determined whether this indicates that the plant operator is performing a task that is not adequately covered by the current design of the distributed control system 5. In step 140 of method 100a, the state variables 6 and the interaction events 7 are mapped to the modifications and / or extensions 2a required for the engineering tool 2 using a machine learning model, a look-up table, or any other suitable means. When this modification and / or extension is implemented, the engineering tool 2 can generate new application code 3, and then the new application code 3 can be executed on the controller 4 within the distributed control system 5.

[0046] According to Figure 2b, state variables 6 are obtained from the controller 4 within the distributed control system 5 and from this control system 5 as a whole. Based on the state variables 6, at least one machine learning model 8 predicts one or more interaction events 7 that are likely to be initiated by at least one plant operator on the distributed control system 5. These predicted interaction events 7 are mapped to the modifications and / or extensions 2a required for the engineering tool 2 in step 160 of method 100b. When this modification and / or extension is implemented, the engineering tool 2 can generate new application code 3, and then the new application code 3 can be executed on the controller 4 within the distributed control system 5.

[0047] Figure 3 is a schematic flowchart of an embodiment of a method 200 for training a machine learning model 8. In step 210, a record 6a of training input data having a state variable 6 characterizing the operating state 1a of at least one industrial plant 1 is provided. In step 220, a label 9 is provided by at least one plant operator regarding which interaction event 7 was initiated in the distributed control system 5 in response to the operating state 1a.

[0048] In step 230, the record 6a of training input data is mapped by the machine learning model 8 to be trained to a prediction 7' of one or more interaction events 7 that at least one plant operator initiates in response to the operating state 1a in the training input data. In step 240, it is evaluated how well this prediction 7' corresponds to the label 9 of each record 6a of the training input data by a predefined cost function 10. In step 250, the parameter 8a characterizing the behavior of the machine learning model 8 is re-optimized towards the goal that this results in a better evaluation 10a by the cost function 10 when further records 6a of training input data are processed by the machine learning model 8. The trained state of the finally obtained parameter 8a is labeled with the reference sign 8a*.

[0049] List of reference signs 1 Industrial plant 1a Operating state of industrial plant 1 2 Engineering tool for the distributed control system 5 2 Modification / expansion for engineering tool 2 3 Application code for the distributed control system 5 4 Controller in the distributed control system 5 5 Distributed control system 6 State variable characterizing the operating state 1a 6a Training data record having a state variable 7 Interaction events between the plant operator and the control system 5 7’ Prediction of interaction event 7 during training of model 8 8 Machine learning model 8a Parameters characterizing the behavior of machine learning model 8 8a* Finally trained state of parameter 8a 9 Labels for training data record 6a 10 Cost function for training machine learning model 8 10a Evaluation by the cost function 100a Method for obtaining modification 2a based on a pattern 100b Method for obtaining modification 2a based on interactive learning 105 Select a specific distributed control system 5 110 Obtain state variable 6 120 Obtain interaction event 7 130 Determine the execution of uncovered tasks 140 Mapping of input data to modification / expansion 2a 141 Determine control library functions 142 Use library functions instead of manual actions 150 Predict interaction event 7 160 Map the predicted interaction event to modification / expansion 2a 161 Determine control library functions 162 Use library functions instead of manual actions 170 Submit modification 2a for approval by the control engineer 180 Regenerate application code 3 190 Execute the new application code 3 200 Method for training machine learning model 8 210 Provide a record 6a of training input data 211 Obtain record 6a from multiple plants 1 220 Provide label 9 230 Map record 6a to prediction 7’ Evaluate the prediction 7’ using the cost function 10 Optimize the parameter 8a The following matters described in the claims of the original application are appended as they are. [1] A computer-implemented method (100a) for modifying and / or extending an engineering tool (2) configured to generate application code (3) that, when executed on one or more controllers (4) within a distributed control system (5) of an industrial plant (1), causes the industrial plant (1) to be controlled according to a control strategy implemented within the application code (3), the method comprising: · obtaining (110) state variables (6) characterizing the operating state (1a) of at least one industrial plant (1); · obtaining (120) a set of interaction events (7) of at least one plant operator who interacts with the distributed control system (5) of the industrial plant (1) via a human-machine interface; · as input data, based at least in part on the interaction events (7), the state variables (6), and optionally predetermined design information of the distributed control system (5), determining (130) whether one or more of the interaction events (7) indicate that the plant operator is performing a task that is not adequately covered by the current design of the distributed control system (5); · if this determination is affirmative, regenerating the application code (3) by the modified and / or extended engineering tool (2) and mapping the input data to a modification and / or extension (2a) of the engineering tool (2) that generated the application code (3) for the distributed control system (5) such that when executed in the distributed control system (5), the frequency with which the plant operator manually interacts with the distributed control system is reduced and / or the time spent interacting with the distributed control system is shortened (140). [2] The tasks not adequately covered by the current design of the distributed control system (5) are specifically: · manually performing solutions to operating problems not covered by the current design of the distributed control system (5), and / or · Repeating the execution of one or more actions starting from an equal or substantially similar operating state (1a), and / or · The method (100) according to [1], comprising accessing at least one function that requires at least a threshold number of steps to be accessed at least at a threshold frequency. [3] A computer-implemented method (100b) for modifying and / or extending an engineering tool (2) configured to generate the application code (3) that, when executed on one or more controllers (4) within a distributed control system (5) of an industrial plant (1), causes the industrial plant (1) to be controlled according to a control strategy implemented within the application code (3), the method comprising: · Obtaining (110) a state variable (6) characterizing the operating state (1a) of at least one industrial plant (1); · Based on the state variable (6), using at least one trained machine learning model (8) to predict (150) one or more interaction events (7) that are likely to be initiated by at least one plant operator on the distributed control system (5) via a human-machine interface in response to the operating state (1a); · Mapping (160) the one or more predicted interaction events (7) to a modification and / or extension (2a) of the engineering tool (2) that generated the application code (3) of the distributed control system (5) such that when the application code (3) is regenerated by the modified and / or extended engineering tool (2) and executed within the distributed control system (5), the frequency with which the plant operator manually interacts with the distributed control system (5) is reduced and / or the time taken to interact with the distributed control system (5) is shortened. [4] The input data comprises: · Alarms and events reported by the distributed control system (5); · A topology model of the industrial plant (1); · The layout of the human-machine interface of the distributed control system (5); · The method (100a, 100b) according to any one of [1] to [3], further comprising one or more of the control logic of the distributed control system (5). [5] The mapping (140, 160) includes: · determining (141, 161) a function within a predetermined control library that achieves a result substantially similar to the result of the detected and / or predicted action or sequence of actions; and · replacing (142, 162) in the modification and / or extension (2a) for the engineering tool (2), the detected and / or predicted action or sequence of actions with a call to the determined function within the control library, the method (100a, 100b) according to any one of [1] to [4]. [6] The modification and / or extension (2a) is such that when the application code (3) is regenerated by the modified engineering tool (2) and executed in the distributed control system (5), in the human-machine interface of the distributed control system (5), · a new control element appears such that a series of actions previously continuously repeated by the plant operator are executed when this new control element is actuated, and / or · a control element that previously required a first number of steps to access within the human-machine interface is accessed so as to require a second, fewer number of steps to access, the method (100a, 100b) according to any one of [1] to [5]. [7] The modification and / or extension (2a) is configured such that when the application code (3) is regenerated by the modified engineering tool (2) and executed in the distributed control system (5), one or more actions previously repeatedly executed by the plant operator starting from an equal or substantially similar operating state (1a) are automatically executed in response to the occurrence of a specific operating state (1a), the method (100a, 100b) according to any one of [1] to [6]. [8] An engineering tool (2) is selected (105), the engineering tool (2) · assembles the distributed control system (5) from building blocks within a predetermined catalog, where at least one such building block is a programmable logic controller (PLC). · A method (100a, 100b) according to any one of [1] to [7], configured to generate application code including control code for this PLC. [9] · Regenerating (180) the application code (3) for the distributed control system (5) by the modified and / or extended engineering tool (2); · Further comprising: executing (190) the regenerated application code (3) in the distributed control system (5), thereby controlling the industrial plant (1) according to the control strategy implemented in the regenerated application code (3). A method (100a, 100b) according to any one of [1] to [8].

[10] Before regenerating (180) the application code (3), further comprising prompting (170) a control engineer to approve the modification and / or extension (2a) to the engineering tool (2). A method (100a, 100b) according to [9].

[11] A computer-implemented method (200) for training at least one machine learning model (8) for use in the method according to [3] and optionally according to any one of [4] to

[10] (100b), the method comprising: · Providing (210) a record (6a) of training input data having state variables (6) characterizing the operating state (1a) of the at least one industrial plant (1); · Providing (220) a label (9) regarding which interaction event (7) at least one plant operator initiated in the distributed control system (5) in response to the operating state (1a); · Mapping (230) the record (6a) of the training input data by the machine learning model (8) to a prediction (7’) of one or more interaction events (7) that at least one plant operator initiates in response to the operating state (1a) within the training input data; · Evaluating (240) how well the prediction (7’) by the machine learning model (8) corresponds to the label (9) of each record (6a) of the training input data by a predefined cost function (10); · Optimizing a parameter (8a) characterizing the behavior of the machine learning model (8) towards the goal that when a further record (6a) of training input data is processed by the machine learning model (8), this results in a better evaluation (10a) by the cost function (10) (250).

[12] The method (200) according to

[11] , wherein the record (6a) of the training input data is collected (211) from a plurality of industrial plants (1).

[13] A computer program comprising machine-readable instructions that, when executed by one or more computers, cause the one or more computers to execute the method (100a, 100b, 200) according to any one of [1] to

[12] .

[14] A non-transitory machine-readable storage medium and / or a download product having the computer program according to

[13] .

[15] One or more computers having the computer program according to

[13] and / or having the non-transitory machine-readable storage medium and / or the download product according to

[14] .

Claims

1. A computer-implemented method (100) for modifying and / or extending an engineering tool (2) configured to generate application code (3) that, when executed on one or more controllers (4) within a distributed control system (5) of an industrial plant (1), causes the industrial plant (1) to be controlled according to a control strategy implemented within the application code (3), the method comprising: ・ obtaining (110) state variables (6) characterizing the operating state (1a) of at least one industrial plant (1); ・ obtaining (120) a set of interaction events (7) of at least one plant operator who interacts with the distributed control system (5) of the industrial plant (1) via a human-machine interface; ・ determining (130), as input data, based at least in part on the interaction events (7) and the state variables (6), whether one or more interaction events (7) indicate that the plant operator is performing a task that is not adequately covered by the current design of the distributed control system (5); ・ if this determination is affirmative, remapping (140) the input data to the modification and / or extension (2a) of the engineering tool (2) that generated the application code (3) for the distributed control system (5) such that, when the modified and / or extended engineering tool (2) regenerates the application code (3) and the application code (3) is executed in the distributed control system (5), the frequency with which the plant operator manually interacts with the distributed control system is likely to be reduced and / or the time taken for the plant operator to interact with the distributed control system is likely to be reduced; ・ regenerating (180) the application code (3) for the distributed control system (5) by the modified and / or extended engineering tool (2); - In the distributed control system (5), execute the regenerated application code (3) (190), thereby controlling the industrial plant (1) according to the control strategy implemented in the regenerated application code (3). The modification and / or extension (2a) When the application code (3) is regenerated by the modified engineering tool (2) and executed in the distributed control system (5), in the human-machine interface of the distributed control system (5), - A new control element appears such that a series of actions previously continuously repeated by the plant operator are executed when this new control element is actuated, and / or - A control element that previously required a first number of steps to access so as to move within the human-machine interface so as to require a second, smaller number of steps to access. A method (100) configured to result in.

2. As input data, based at least in part on predetermined design information of the distributed control system (5), determine (130) whether one or more interaction events (7) indicate that the plant operator is performing a task not sufficiently covered by the current design of the distributed control system (5). The method (100) according to claim 1.

3. The task not sufficiently covered by the current design of the distributed control system (5) specifically includes - Manually executing a solution to an operating problem not covered by the current design of the distributed control system (5), and / or - Repeatedly executing one or more actions starting from an equal or substantially similar operating state (1a), and / or - Accessing at least one function that requires at least a threshold number of steps to access at least at a threshold frequency, the method (100) according to claim 1.

4. The input data is - Alarms and events reported by the distributed control system (5), and - A topology model of the industrial plant (1), and - A layout of a human - machine interface of the distributed control system (5); and a control logic of the distributed control system (5), further including one or more of, the method (100) according to any one of claims 1 to 3.

5. The mapping (140, 160) - Determining (141, 161) a function within a predetermined control library that achieves a result substantially similar to the result of a detected and / or predicted action or sequence of actions; and - Replacing (142, 162) the detected and / or predicted action or sequence of actions with a call to the determined function within the control library in the modification and / or extension (2a) for the engineering tool (2), the method (100) according to any one of claims 1 to 4.

6. The modification and / or extension (2a) is regenerated by the engineering tool (2) in which the application code (3) is modified, and when executed in the distributed control system (5), starting from an equal or substantially similar operating state (1a), one or more actions previously repeatedly executed by the plant operator are automatically executed in response to the occurrence of a specific operating state (1a), the method (100) according to any one of claims 1 to 5.

7. An engineering tool (2) is selected (105), the engineering tool (2) - Assembling the distributed control system (5) from building blocks within a predefined catalog, where at least one such building block is a programmable logic controller (PLC), - A method (100) according to any one of claims 1 to 6, configured to generate application code including control code for the PLC.

8. - The method (100) according to claim 1, further comprising causing a computer to prompt a control engineer for approval of the modification and / or extension (2a) to the engineering tool (2) before regenerating the application code (3) (170).

9. - A computer program comprising machine-readable instructions that, when executed by one or more computers, cause the one or more computers to perform the method (100) according to any one of claims 1 to 8.

10. - A non-transitory machine-readable storage medium and / or a download product having the computer program according to claim 9.

11. - One or more computers having the computer program according to claim 9 and / or having the non-transitory machine-readable storage medium and / or download product according to claim 10.

Images