Modular Model Predictive Control for Industrial Plants

Abstract

A method (100) for generating a dynamic model (10) of an industrial plant (1), the method comprising: obtaining (110) a representation of the plant (1) into partitions of subunits (1a, 1b), the subunits (1a, 1b) being characterized by respective subsets (5a, 5b) of setpoints (51-57) of low-level controllers (31-37), subsets (4a, 4b) of current values ​​of process variables (41-49) and subsets (4a', 4b') of future predicted values ​​(41'-49') of the process variables (41-49); obtaining (120) for each subunit (1a, 1b) Dynamic sub-models (13a, 13b), wherein the inputs include a subset (5a, 5b) of set points (51-57) and a subset (4a, 4b) of current values ​​of process variables (41-49), and the outputs include a subset (4a', 4b') of future predicted values ​​(41'-49'); correlations (14) between the sub-units (1a, 1b) are obtained (130) from the layout of the plant (1) and / or from the composition of the plant from physical process modules; and the sub-models (13a, 13b) are combined (140) into a model (10) of the plant (1) based on the correlations (14).

Description

[0001] The present invention relates to the field of automated control of physical processes in industrial plants. Background Art

[0002] In many complex industrial plants, the final result of a production process or other industrial process is obtained through a series of individual physical processes, which are related because the result of at least one first process is fed into at least one second process. The automatic control of many such plants is organized in a hierarchy including at least two levels. There are low-level controllers acting on individual physical processes so that at least one process variable ideally matches the set point of the low-level controller. In order to determine the set point of the low-level controller, a model predictive controller (sometimes also called a high-level predictive controller) is used. Starting from the current state of the process, such a controller determines which future state the current state of the industrial process as a whole will evolve to when certain sets of set points are applied to the low-level controller based on the (usually) time-discrete dynamic model of the plant. These future states are evaluated according to one or more optimal criteria (e.g., low energy consumption, throughput maximization), and it is usually also evaluated whether the future state meets certain constraints of the industrial process or the plant (e.g., the maximum temperature of certain sub-processes). The set of set points that produce the best future state according to the optimal criterion within the limits set by the constraint is calculated and applied to the low-level controller. In the next discrete time step, the process is repeated, wherein the current state of the industrial process is determined by the actual measured values ​​of the state variables of the industrial process. An exemplary embodiment of such a control scheme is disclosed in US 10,082,772 B2.

[0003] Purpose of the Invention

[0004] It is an object of the present invention to facilitate the acquisition of a dynamic model required for predicting future states in a model predictive controller.

[0005] This object is achieved by a method for generating a dynamic model according to the independent claim and a dynamic model according to another independent claim. Another independent claim relates to a method for operating an industrial plant. The method and the model can be embodied in a computer program, a storage medium and / or a download product, a computer and / or a control system. Further advantageous embodiments are detailed in the dependent claims. Summary of the invention

[0006] The inventors have developed a method for generating a dynamic model of an industrial plant. For example, the model can be a time-discrete model. If such a model is not time-discrete, it can be discretized during the operation of the plant.

[0007] The plant comprises a plurality of physical processes which are related in that the result of at least one first process is fed into at least one second process. For example, there may be a first container in which a mixture of two substances is stirred at a certain speed and heated to a certain temperature, and then the mixture is transferred to another container in which a third substance is added at a certain mass flow rate and the mixture is irradiated with UV light of a certain intensity to polymerize it. Throughout this disclosure, the term "physical" is to be understood as a general term which explicitly also includes chemical processes in which educts are converted into products by chemical reactions, and biological processes in which such conversions are carried out with the help of living organisms.

[0008] The plant includes a plurality of low-level controllers. Each such controller acts on at least one physical process so that at least one process variable of the process is controlled to match a set point of the low-level controller. For example, such a low-level controller may control the current of a heater of a first container so that the temperature inside the container matches a set point, or it may control the current of a motor that stirs a mixture so that the motor rotates at a certain speed.

[0009] The plant also has a plurality of sensors. Each such sensor measures at least one process variable of one of the physical processes and / or the plant as a whole. For example, there may be a sensor measuring the temperature in a container, and / or a sensor for the turbidity of the mixture inside the container, which is a measure of the homogeneity of the mixture. Two examples of sensors measuring process variables of the plant as a whole are: a sensor measuring the mass flow of the final product from the plant, and a sensor measuring the total current consumed by the plant. That is, the process variables may include state variables of the physical process and / or the plant as a whole, but more generally may also include any other measured quantity related to the physical process and / or the plant as a whole.

[0010] The model takes as input the current setpoint of the low-level controller and the current value of the process variable measured by the sensor. Based on this, the model outputs the predicted future values ​​of the process variable that would result from currently applying the above setpoint to the low-level controller.

[0011] For example, the model may predict that when starting with a specific mass flow rate of final product and a set point temperature of 40°C inside the vessel, increasing that temperature to 45°C will cause the chemical reaction inside the vessel to speed up, resulting in a 15% increase in the flow rate of the final product.

[0012] In the course of the method, a representation of the plant is obtained into sub-units. The term "representation" means that the segmentation does not physically cut off the actual components of the plant and their interconnections, but rather the engineering model of the plant is segmented. In other words, the representation can be considered a "digital twin" of the plant.

[0013] This segmentation may be obtained from any source. If the plant is a modular industrial plant, an already existing segmentation may be used which becomes the physical process modules that make up the plant.

[0014] For example, a pharmaceutical company may need to frequently reconfigure a plant to produce new products because these products are so concentrated in terms of end effect per unit mass or volume that a year's supply can be produced in just a few weeks. The company may have a certain set of physical process modules and arrange corresponding subsets of the set in corresponding configurations, for example, to make product A first, then product B, then product C, and finally back to product A. In each configuration, the physical process modules are linked together in a temporary manner during the use of that configuration.

[0015] If there are no partitions that become physical process modules, a suitable sub-unit can be actively searched for in the layout of the plant. In other words, the representation of the plant can be actively partitioned into sub-units. Several strategies for performing this operation (which can be used individually or in any combination) are detailed below. But even in the case where there are partitions that become physical process modules, this is not necessarily the final partition obtained. For example, there may be a situation where two or more physical process modules are combined into one sub-unit.

[0016] The input-output behavior of each subunit is characterized by a corresponding subset of setpoints, current values ​​of process variables, and future values ​​of process variables, and the setpoints required to achieve them. In the above toy example, the first subunit may be a first container in which the mixture is heated and stirred, and the second subunit may be a second container in which the mixture is polymerized.

[0017] For each subunit, a dynamic submodel is obtained, wherein the input of the submodel includes the above subset of set points and the above subset of current values ​​of the process variables, and the output of the submodel includes the future values ​​of the process variables. Taking the toy as an example, the first submodel can simulate how the mass flow rate of the mixture leaving the first container and the uniformity of the mixture will be affected by changes in the set points of temperature and motor speed. The second submodel can simulate how the mass flow rate of the final product leaving the second container and some measure of the quality of the polymerization will be affected by changes in the mass flow rate of the third component added in the second container and the set point for the UV light intensity.

[0018] The submodels may be obtained from any suitable source. For example, if an abstract module type package MTP is associated with a physical process module, the package may already contain a submodel corresponding to the physical process module. If a subunit in the partition corresponds to the physical process module, the submodel from MTP can be used immediately. But the submodels may also be obtained from a library or directory that associates certain subunits with submodels. In particular, the strategy for partitioning to obtain a representation of a plant may be to cover as much of the plant layout as possible with subunits for which corresponding submodels are available.

[0019] The dependencies between the sub-units are taken from the engineering layout of the plant and / or from the existing composition of the plant from the physical process modules. In the toy example, the mass flow of the mixture of the first substance and the second substance entering the second container is equal to the mass flow of the mixture leaving the first container if no other source can provide the mixture. The sub-models are combined into a model of the plant based on these dependencies. If several physical process modules are connected to each other, the mutual dependencies between these physical process modules are set by the physical interconnections between these physical process modules (e.g., by pipes, cables or hoses).

[0020] The inventors have discovered that this method for generating the dynamic model required for model predictive control of an entire plant makes the creation of the model more efficient and more transparent for engineers. Previously, when creating a holistic model for an entire plant, it was more of an art than a task for automation engineers. Process experts had to manually decide whether to use theoretical modeling based on physical laws and simplifying assumptions, experimental modeling, or a hybrid of the two. The creation of the model requires multiple iterations and testing, and is difficult to automate. For the purpose of model predictive control, the model is often so highly specialized that the average plant engineer cannot interpret the reasoning behind the model. Therefore, when it is actually used in the plant, it often generates set points that the average operator cannot interpret.

[0021] In contrast, the present method provides a direct approach based on building blocks (i.e., sub-models) that are transparent to the average plant operator. Therefore, the set points generated by the model are more meaningful to the average operator. Therefore, the average operator has the opportunity to supervise the overall automation control of the plant and pay attention to whether there are any abnormal situations. This provides an additional layer of safety in case of any unexpected failures, such as sensor failure or pipe blockage.

[0022] Furthermore, the model may be more adaptable to any changes in the factory. For example, the trend is to move away from monolithic factories to modular factories, where modules can be added and removed, or brought online and offline, as needed. Whenever this type of change occurs, the overall model of the factory can be adapted in a straightforward way by simply adding and removing identical copies of the same sub-model at the right place. Using the toy example, there may be two identical copies of the first container. If the demand for the final product is high, both copies may be used in order to produce enough raw material for the second container, where polymerization is to take place. If the demand is high and using one copy of the first container is sufficient, the other copy may be taken offline to save electricity for heating and stirring.

[0023] Furthermore, if any physical changes are made within one subunit, only one submodel of that subunit needs to be changed. In the toy example, the second container may need to be replaced with a new one with a different geometry, so the local distribution of the UV light intensity and the polymerization inside the second container will be different. On the other hand, this affects the relationship between the set point of the UV lamp intensity and the mass flow and quality of the final product.

[0024] The assembly of the sub-models into a final model of the entire plant is a process that can be easily automated, or at least assisted by the use of computers. Even if the task of creating individual sub-models for the sub-units is left to a human expert, the overall saving in manpower is still considerable, because it is much easier to produce several fairly simple sub-models than to produce one more complex overall model. This is especially true if a particular human expert is to be taken as a given. If the overall model is too complex for this particular expert, the expert will never be able to produce it. But if the task can be broken down into the creation of several simple sub-models, each of which is within the expert's range, the expert can produce the sub-models one by one.

[0025] Furthermore, the present approach also makes it feasible to later refine individual sub-models after the initial creation of a model for the entire plant. For example, a plant may initially be started up using fairly simple sub-models at specific locations in order to get started quickly. Further work by the plant operators may result in a new refined sub-model that can simply be snapped in place of the previous simple model. Since the sub-model is always the same for all instances of the sub-unit in the plant, there is also a market for improved sub-models. For example, if a sub-unit coincides with a unit that is sold as a package (e.g. a first container with a heater and agitator already built in), the manufacturer of the package, or even a third party, may later publish a more refined model.

[0026] The advantages described do not come "for free". The calculation of predictions based on a model generated based on the present method involves sub-calculations based on many sub-models, and the results of these sub-calculations need to be aggregated. This may take longer than calculating predictions based on a specially made highly complex overall model. However, this is a small price that can be paid in a straightforward way by adding more computing resources. In contrast, if an industrial process is too complex for a human modeling expert to understand, this process will have no model and no model predictive control at all.

[0027] In a particularly advantageous embodiment, at least one subunit is selected to correspond to a module of a modular automation solution enabled in place at the plant. When changes are made to the layout of such a plant, they are likely to be additions and removals of modules. Therefore, if a submodel is available for each such module, the overall dynamic model of the plant can be easily recreated in response to such changes simply by "recompiling" the components of the submodel according to the current layout of the plant.

[0028] Specifically, in another particularly advantageous embodiment, a module can provide its own dynamic model, and this can be immediately used as a submodel of the corresponding subunit. For example, a module can have a module type package MTP as its "digital twin". The MTP can contain a dynamic model of the subunit corresponding to the module, and / or information about the inputs and outputs of the subunit, and / or information about the operation, interface, simulation, automation and parameters of the subunit. The MTP provides an abstraction that greatly facilitates the combination of modules from different suppliers into a plant, including the generation of a dynamic model for the plant as a whole.

[0029] However, using existing subdivisions that become modules in a factory is not the only way that the overall model can be automatically generated. There are other ways to generate models at least partially with the help of computers.

[0030] One such approach is to search the factory layout for subunits that occur multiple times in the factory, where each such subunit includes at least two low-level controllers and at least two sensors. Since the submodel for each instance of the subunit is identical, the submodel only needs to be created once and can then be reused for all other instances.

[0031] In an exemplary embodiment, the search can be focused on sub-units of a particular size, and if no or no more sub-units of that size are found, the search can be continued at a reduced size. This provides a direct method to detect repeating units in a factory layout, even if the composition of these units is unknown.

[0032] Another possible strategy for partitioning the plant representation is to make the size of the subunits dependent on the accuracy requirements of the corresponding submodels. If lower accuracy is required, relatively large subunits can be abstracted by relatively simple submodels. Therefore, in another particularly advantageous embodiment, in response to determining that the accuracy requirements for the submodel of the second subunit are lower than the accuracy requirements for the submodel of the first subunit, the second subunit is selected to be larger than the first subunit.

[0033] In another advantageous embodiment, at least one boundary of the subunit is chosen to coincide with a boundary between different areas of responsibility within the plant. In this way, the acquisition of the submodel is facilitated, since fewer people need to be asked for relevant information when making the submodel.

[0034] The overall dynamic model of a plant does not need to be generated on-site at the plant itself. Rather, it can also be generated by an entity other than the plant owner or operator. For example, the automatic generation of an industrial plant model based on a computerized representation of the plant layout can be provided as a service. The finished model is thus a marketable product with its own value. Therefore, the present invention also relates to a time-discrete dynamic model for an industrial plant generated by the above method.

[0035] The finished model generated by the above method has an internal structure that is different from the overall model. Assembling the model from sub-modules leaves a characteristic "fingerprint" on the model, which can be discovered when the model is inspected. Therefore, the present invention also relates to a dynamic prediction model for an industrial plant including multiple sub-models, wherein the input of the model is distributed across the inputs of the sub-models, the output of the model is compiled from the outputs of the sub-models, and at least one output of one sub-model is processed into at least one input of another sub-model.

[0036] By means of the model, the present invention also provides a method for operating an industrial plant, the method comprising:

[0037] a plurality of physical processes which are related in that the result of at least one first process is fed into at least one second process,

[0038] a plurality of low-level controllers, each such controller acting on at least one physical process such that at least one process variable of the process is controlled to match a set point of the low-level controller, and

[0039] • A plurality of sensors, each sensor measuring at least one process variable of one of the physical processes and / or the plant as a whole.

[0040] In the course of the method, a set of future values ​​of a process variable of the plant is predicted for a plurality of candidate sets of set points for the low-level controller using a dynamic model as described above. From the plurality of candidate sets, a corresponding set of predicted future values ​​of the process variable that is optimal according to at least one predetermined optimality criterion is determined as an optimal set of set points.

[0041] There is complete freedom regarding the optimality criteria. Exemplary criteria relevant to industrial processes are: minimum energy consumption, minimum material use, maximum throughput, minimum waste, and / or minimum total cost to produce a given amount of final product; minimum level of noise or other pollutants emitted from the plant; minimum wear and tear on the plant; or minimum downtime of the plant.

[0042] An optimal set of set points is applied to low-level controllers. At least one of the low-level controllers applies at least one actuation output to at least one physical actor. The physical actor in turn physically changes at least one aspect of the behavior of at least one physical process contained in the overall industrial process.

[0043] In another particularly advantageous embodiment, the determination of the optimal set of setpoints is limited by at least one constraint on at least one setpoint, at least one process variable or any combination thereof.The possibility of explicit model constraints is one of the main advantages of model predictive control.

[0044] As described above, these methods can be implemented in whole or in part by a computer. Specifically, they can be implemented in software that produces the above advantages when executed by one or more computers and / or control systems. Therefore, the present invention also relates to a computer program comprising machine-readable instructions that, when executed on one or more computers and / or control systems, cause the computers and / or control systems to perform one or both of the methods described above.

[0045] The present invention also relates to a non-transitory machine-readable storage medium and / or a download product having a computer program as described above and / or having a dynamic model as described above. The sub-models stored in the module type package corresponding to the physical process module can be called by the dynamic model as subroutines, thereby merging these sub-models into the dynamic model.

[0046] The invention also relates to one or more computers and / or a control system having a dynamic model as described above, having a computer program as described above, having a storage medium or a download product as described above, and / or being specially adapted in any other way to perform one or both of the methods as described above. For example, such a special adaptation may include at least a partial implementation of the method in an application specific integrated circuit ASIC. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In the following, the present invention is further described using the accompanying drawings, without any intention to limit the scope of the present invention. The accompanying drawings show:

[0048] Figure 1 : An exemplary embodiment of method 100;

[0049] Figure 2 : A toy example of a factory 1 for illustrating the method 100;

[0050] Figure 3 :Use method 100 Figure 1 An exemplary model 10 generated for the plant 1 shown;

[0051] Figure 4 : An exemplary embodiment of method 200. DETAILED DESCRIPTION

[0052] Figure 1 An exemplary embodiment of a method 100 is shown. In step 110, a representation of a plant 1 is split into subunits 1a and 1b. This also means that the process variables 41-49 are split into respective sets 4a, 4b belonging to subunits 1a and 1b, the future values ​​41'-49' of the process variables 41-49 are split into respective sets 4a', 4b' belonging to subunits 1a and 1b, and the set points 51-57 for the low-level controllers 31-37 are split into respective sets 5a, 5b belonging to subunits 1a and 1b.

[0053] Some examples of how segmentation 110 may be performed are shown within block 110 .

[0054] According to block 111 , at least one module enabling a modular automation solution may be selected as a subunit 1 a , 1 b .

[0055] According to block 112, a search may be performed for subunits 1a, 1b that occur in multiple instances within the plant 1. According to subblock 113, the search may be focused on subunits 1a, 1b of a particular size. According to subblock 114, the search may then move to reduced sizes.

[0056] According to block 115, it may be determined that for the second subunit 1b, the accuracy requirement of the corresponding submodel 13b is lower than the submodel 13a corresponding to the first subunit 1a. In response to this determination, according to block 116, the second subunit 1b may be made larger than the first subunit 1a.

[0057] According to box 117 , at least one boundary of the subunit 1 a , 1 b may be selected to coincide with a boundary between different areas of responsibility in the plant 1 .

[0058] In step 120 , dynamic sub-models 13a , 13b may be generated for the sub-units 1a , 1b . Two exemplary methods are shown in block 120 .

[0059] According to block 121 , the dynamic model provided by the module selected as the subunit 1 a , 1 b may be used directly as the submodel 13 a , 13 b of this subunit 1 a , 1 b .

[0060] According to box 122, information about the model and other meta information can be obtained from the module type package MTP as the digital twin of the subunit 1a, 1b.

[0061] In step 130 , the dependencies 14 between the subunits 1 a , 1 b are obtained from the layout of the plant 1 and / or from the existing composition of the plant from the physical process module.

[0062] In step 140 , the sub-models 13 a , 13 b are combined according to the dependencies 14 to form the model 10 of the plant 1 .

[0063] Figure 2 A toy example of a plant 1 to which the method 100 may be applied is shown. For reasons of readability, Figure 2 In the figure, the low-level controllers 31-37 are not shown separately from the actors they act on. Instead, the actors are drawn and annotated with the reference numerals 31-37 corresponding to the low-level controllers.

[0064] The plant 1 comprises a first vessel V1 and a second vessel V2.

[0065] In a first container V1, in a physical process 2a, two substances A and B are stirred and heated to produce a mixture M. To this end,

[0066] The controller 31 controls the mass flow 41 of substance A measured by the sensor 61 to match the set point 51 .

[0067] The controller 32 controls the mass flow rate 42 of substance B measured by the sensor 62 to match the set point 52;

[0068] The controller 33 controls the temperature 43 inside the first container V1 measured by the sensor 63 via the heater H to match the set point 53 .

[0069] The controller 34 controls the rotation speed 44 of the stirrer S measured by the sensor 64 to match the set point 54 .

[0070] Furthermore, the sensor 68 measures the turbidity 48 of the mixture M inside the first container V1 .

[0071] The mixture M enters the second container V2, where, in the second physical process 2b, a third substance C is further added and UV polymerization is carried out to form the product P. To this end,

[0072] The controller 35 controls the mass flow 45 of substance C measured by the sensor 65 to match the set point 55 .

[0073] Controller 36 controls the intensity 46 of the UV radiation measured by sensor 66 to match set point 56; and

[0074] The controller 37 controls the mass flow 47 of the final product P measured by the sensor 67 to match the set point 57 .

[0075] Furthermore, the sensor 69 measures the viscosity 49 of the product P, which is a measure of the quality of the polymerization.

[0076] In this toy example, it is easy to split the representation of the plant 1 into subunits 1a, 1b: What is measured in or acts on the content of the first container V1 is assigned to subunit 1a. What is measured in or acts on the content of the second container V2 is assigned to subunit 1b. Correspondingly, process variables 41-49 are assigned to subsets 4a and 4b, and setpoints 51-57 are assigned to subsets 5a and 5b.

[0077] Figure 3 The model 10 generated in this way is shown. The model 10 consists of two submodules 13a and 13b corresponding to the subunits 1a and 1b. The first submodule 13a takes as input the current values ​​of the process variables 41-44 and 48 related to the container V1, and the set points 51-54. Based on these inputs, the submodel 13a predicts how the process variables 41-44 and 48 will evolve into future values ​​41'-44' and 48'.

[0078] These future values ​​41'-44' and 48' of the process variables 41-44 and 48 are fed as input into the second sub-model 13b. In addition, the second sub-model 13b takes as input the current values ​​of the process variables 45-47 and 49 related to the vessel V2, and the set points 55-57. Figure 3 In the example shown, the second submodel 13b does not take further action on the future rotation speed 44' of the stirrer S and the future turbidity 48' of the mixture M, since these quantities do not directly affect the polymerization result. However, the sum of the future mass flows 41' and 42' of substances A and B is the mass flow of the mixture M, which is relevant to the stoichiometry of the product P. Furthermore, the polymerization in the container V2 is temperature-dependent, so the future temperature 43 of the mixture M in the container V1 is also relevant to the second submodel 13b.

[0079] The second submodel 13b predicts future values ​​45'-47' and 49' of the process variables 45-47 and 49 based on their current values, the setpoints 55-57 acting directly on them and the future values ​​41'-43' of the relevant process variables 41-43 obtained from the first submodel 13a.

[0080] exist Figure 3 In the example shown, only the future values ​​41'-44' and 48' from the first submodel 13a are fed into the second submodel 13b, but not vice versa. However, in actual applications, one or more of the future values ​​45'-47' and 49' from the second submodel 13b may also be associated with the first submodel 13a.

[0081] In summary, all current values ​​of the process variables 41 - 49 and all setpoints 51 - 57 of the low-level controllers 31 - 37 form the input 11 of the new model 10 of the plant 1 , while all future values ​​41 ′- 49 ′ of the process variables 41 - 49 form the output 12 of the new model 10 .

[0082] Figure 4 An exemplary embodiment of a method 200 for operating a plant 1 is shown. In step 210, the dynamic model 10 is used to predict, based on the current values ​​of the process variables 41-49, sets 8a-8f of future values ​​41'-49' of these process variables 41-49 that will result when different candidate sets 7a-7f of setpoints 51-57 are applied to the low-level controllers 31-37.

[0083] In step 220, corresponding sets 8a-8f of future values ​​41'-49' are determined as optimal set 7 of set points 51-57 according to the optimality criteria 91a-91c. According to box 221, the optimization may be constrained according to constraints 92a-92c.

[0084] In step 230, the optimal set 7 of set points 51-57 is applied to the low-level controllers 31-37. In step 240, the low-level controllers 31-37 apply actuation outputs 77 to physical actors that physically change aspects of the behavior of the physical processes 2a, 2b.

[0085] Reference numerals list

[0086] 1: Industrial Factory

[0087] 1a, 1b: Subunits in plant 1

[0088] 2a, 2b: Physical processes in plant 1

[0089] 10: Dynamic model of factory 1

[0090] 11: Input of Model 10

[0091] 12: Output of Model 10

[0092] 13a, 13b: Submodels in model 10

[0093] 14: Correlation between subunits 1a and 1b

[0094] 31: Low-level controller for mass flow 41

[0095] 32: Low level controller for mass flow 42

[0096] 33: Low level controller for temperature 43

[0097] 34: Low-level controller for speed 44

[0098] 35: Low level controller for mass flow 45

[0099] 36: Low level controller for UV intensity 46

[0100] 37: Low-level controller for mass flow 47

[0101] 4a, 4b: Subset of current values ​​of process variables 41-49

[0102] 4a', 4b': Subset of future predicted values ​​41'-49' of process variables 41-49

[0103] 41: Mass flow rate of substance A

[0104] 42: Mass flow rate of substance B

[0105] 43: Temperature in container V1

[0106] 44: Speed ​​of stirrer S

[0107] 45: Mass flow rate of substance C

[0108] 46: UV intensity

[0109] 47: Mass flow rate of product P

[0110] 48: Turbidity of mixture M in container V1

[0111] 49: Viscosity of product P in container V2

[0112] 41'-49': Future predicted values ​​of process variables 41-49

[0113] 5a, 5b: Subset of setpoints 51-57

[0114] 51: Set point for mass flow rate 41

[0115] 52: Set point for mass flow rate 42

[0116] 53: Set point for temperature 43

[0117] 54: Set point for speed 44

[0118] 55: Set point for mass flow rate 45

[0119] 56: Set point for UV intensity 46

[0120] 57: Set point for mass flow rate 47

[0121] 61: Sensor for 41 mass flow

[0122] 62: Sensor for mass flow 42

[0123] 63: Sensor for temperature 43

[0124] 64: Sensor for speed 44

[0125] 65: Sensor for mass flow 45

[0126] 66: Sensor for UV intensity 46

[0127] 67: Sensor for mass flow 47

[0128] 68: Sensor for turbidity 48

[0129] 69: Sensor for viscosity 49

[0130] 7: Set point 51-57 optimal set

[0131] 77: Actuation output based on optimal set 7 of set points 51-57

[0132] 7a-7f: Candidate set of set points 51-57

[0133] 8a-8f: The set of future prediction values ​​41'-49' obtained for the candidate set 7a-7f

[0134] 100: Methods for generating models 10

[0135] 110: Split the representation of plant 1 into subunits 1a and 1b

[0136] 111: Select module as subunit 1a, 1b

[0137] 112: Search for repeated subunits 1a and 1b

[0138] 113: Focus the search 112 on a specific size of subunits 1a, 1b

[0139] 114: Continue searching for the subunits 1a and 1b with reduced size in 112

[0140] 115: Determine the accuracy requirements of sub-models 13a and 13b

[0141] 116: Select the size of subunits 1a and 1b according to the accuracy requirements

[0142] 117: Select boundaries based on area of ​​responsibility

[0143] 120: Obtain sub-models 13a and 13b of sub-units 1a and 1b

[0144] 121: Select sub-models 13a and 13b provided by the module

[0145] 122: Read information from module type package MTP

[0146] 130: Determine the correlation between subunits 1a and 1b

[0147] 140: Combining sub-models 13a and 13b according to correlation 14

[0148] 200: Method for controlling plant 1

[0149] 210: Set 8a-8f of the future values ​​41'-49' of the predicted candidate set 7a-7f'

[0150] 220: Determine the optimal set 7 using the optimal criteria 91a-91c

[0151] 221: Apply constraints 92a-92c to determine 220

[0152] 230: Apply optimal set 7 to low-level controller 31-37

[0153] 240: Apply execution output 77 to physical processes 2a, 2b

[0154] A, B, C: Matter

[0155] H: Heater

[0156] M: Mixture

[0157] P: Product

[0158] S:Blender

[0159] V1, V2: Container

Claims

1. A method (100) for generating a dynamic model (10) of an industrial plant (1), wherein the industrial plant (1) is a modular industrial plant divided into physical process modules and the industrial plant (1) comprises a plurality of physical processes which are related in that the result of at least one first process (2a) is fed into at least one second process (2b), a plurality of low-level controllers (31-37), each such controller (31-37) acting on at least one physical process such that at least one process variable of that process (2a-2b) is controlled to match a set point (51-57) of said low-level controller (31-37), and a plurality of sensors (61-69), each sensor (61-69) measuring at least one process variable of one of the physical processes and / or of the industrial plant (1) as a whole, wherein the set points (51-57) of the low-level controllers (31-37) and the current values ​​of the process variables measured by the sensors (61-69) are inputs (11) to the dynamic model (10), and predicted future values ​​(41'-49') of the process variables are outputs (12) of the dynamic model (10), the predicted future values ​​(41'-49') being produced by applying the set points (51-57) to the low-level controllers (31-37), the outputs (12) of the dynamic model (10) characterizing the future state to which the current state of the entire industrial process performed by the industrial plant (1) will evolve when the set points (51-57) are applied to the low-level controllers (31-37), and wherein the method (100) comprises: obtaining (110) a representation of the industrial plant (1) into a partitioning of a plurality of subunits (1a, 1b) corresponding to physical process modules of the modular industrial plant, wherein the input-output behavior of each subunit (1a, 1b) is characterized by a corresponding subset (5a, 5b) of the set points (51-57), a subset (4a, 4b) of the current values ​​of the process variables and a subset (4a', 4b') of the predicted future values ​​(41'-49') of the process variables; obtaining a dynamic sub-model (13a, 13b) of the corresponding sub-unit (1a, 1b) from a module type package MTP corresponding to each sub-unit (1a, 1b), wherein the input of the dynamic sub-model (13a, 13b) comprises the subset (5a, 5b) of the set points (51-57) and the subset (4a, 4b) of current values ​​of process variables, and the output of the dynamic sub-model (13a, 13b) comprises the subset (4a', 4b') of predicted future values ​​(41'-49') of process variables; deriving dependencies (14) between the subunits (1a, 1b) from the existing composition of the modular industrial plant from physical process modules, wherein these dependencies (14) correspond to the physical interconnections between these physical process modules; and Combining (140) the dynamic sub-models (13a, 13b) into the dynamic model (10) of the industrial plant (1) according to the correlation (14).

2. The method (100) according to claim 1, further comprising: Further information about the input and output of the subunit (1a, 1b) and / or about the operation, interface, simulation, automation and parameters of the subunit (1a, 1b) is obtained from a module type package MTP corresponding to the subunit (1a, 1b).

3. A method (100) according to any one of claims 1 to 2, wherein obtaining (110) the segmentation into subunits also includes searching (112) for subunits (1a, 1b) that occur multiple times in the industrial plant (1), wherein each such subunit (1a, 1b) comprises at least two low-level controllers (31-37) and at least two sensors (61-69).

4. The method (100) according to claim 3, wherein the search is focused (113) on subunits (1a, 1b) of a certain size and, if no or no more subunits (1a, 1b) of said size are found, the search is continued (114) with a reduced size.

5. A method (100) according to any one of claims 1, 2 and 4, wherein in response to determining (115) that the accuracy requirement for the dynamic submodel (13b) of the second subunit (1b) is lower than the accuracy requirement for the dynamic submodel (13a) of the first subunit (1a), the second subunit (1b) is selected (116) to be larger than the first subunit (1a).

6. A method (100) according to any one of claims 1, 2 and 4, wherein at least one boundary of a subunit (1a, 1b) is selected (117) to coincide with a boundary between different areas of responsibility in the industrial plant (1).

7. The method according to any one of claims 1, 2 and 4, further comprising: using the dynamic model (10) to predict (210) a set (8a-8f) of future values ​​(41'-49') of a process variable of the industrial plant (1) for a plurality of candidate sets (7a-7f) of set points (51-57) of the low-level controllers (31-37); determining, from the plurality of candidate sets (7a-7f), the candidate sets (7a-7f) whose corresponding prediction sets of future values ​​(41'-49') of the process variable are optimal according to at least one predetermined optimality criterion (91a-91c), as the optimal set (7) of set points (51-57); Applying the optimal set of set points (7) to the low-level controllers (31-37); as well as Applying at least one actuation output (77) by at least one low-level controller (31-37) to at least one physical actor, which in turn physically changes at least one aspect of the behavior of at least one physical process.

8. The method according to claim 7, wherein the determination of the optimal set (7) of set points (51-57) is limited (221) by at least one constraint (92a-92c) regarding at least one set point (51-57), at least one process variable or any combination thereof.

9. A computer program product comprising a computer program, the computer program comprising machine readable instructions which, when executed on one or more computers, cause the computers to perform the method according to any one of claims 1 to 8.

10. A computer program product comprising a computer program comprising machine readable instructions which, when executed on a control system, cause the control system to perform a method according to any one of claims 1 to 8.

11. A non-transitory machine-readable storage medium comprising a computer program, the computer program comprising machine-readable instructions which, when executed on one or more computers, cause the computers to perform the method according to any one of claims 1 to 8.

12. A non-transitory machine-readable storage medium comprising a computer program, the computer program comprising machine-readable instructions which, when executed on a control system, cause the control system to perform the method according to any one of claims 1 to 8.

13. A control system having a computer program product according to claim 10 and / or having a storage medium according to claim 12.

14. A computer having a computer program product according to claim 9 and / or having a storage medium according to claim 11.

Images