Method for providing a list of equipment elements in an industrial plant

Abstract

The embodiments of this disclosure generally relate to methods for providing a list of equipment elements in an industrial facility. The present invention relates to the field of industrial facilities and processes, and in particular to methods for maintaining and / or improving facilities and processes. The method provides an impact list of equipment elements that affect an industrial subprocess. The method includes the steps of: selecting a subprocess in a topology model, wherein the subprocess is a equipment element that is part of an industrial facility or process, and wherein the topology model is a graph whose nodes represent equipment elements and whose edges represent interconnections between equipment elements; traversing the nodes of the topology model, wherein the traversal begins with the selected subprocess and uses a traversal strategy; and for each of at least one equipment element, adding the equipment element to the impact list of equipment elements if the degree of influence of the equipment element on the industrial subprocess is greater than a first predetermined degree of influence.

Description

Technical Field

[0001] This invention relates to the field of industrial facilities and processes, and in particular to methods for maintaining and / or improving facilities and processes. The invention also relates to computer-based systems, program units, and computer-readable storage media. Background Technology

[0002] Industrial facilities and processes are not only complex networks of elements connected by flows of materials, energy, and / or information, but also, in some cases, even for very distant elements, often possess domain process rules that define the logical interrelationships between elements and / or how they affect each other's states. To maintain and / or improve facilities and processes, a deep understanding of the functional interrelationships of selected equipment elements can be helpful. Summary of the Invention

[0003] One object of the present invention is to provide a method for maintaining and / or improving facilities and processes and / or their equipment elements. This object is achieved by the subject matter of the independent claims. Further embodiments will be apparent from the dependent claims and the following description.

[0004] One aspect relates to a method for providing a list of influences of equipment elements affecting an industrial subprocess, wherein the influences of the industrial subprocess may (or may not) include flows of materials, energy, information, and / or other aspects that control or influence the subprocess. The method includes the following steps:

[0005] Select subprocesses in the topology model, where a subprocess is a device element that is part of an industrial facility or process, and where the topology model is a graph, where nodes of the graph represent device elements and edges of the graph represent interconnections between device elements;

[0006] Traverse the nodes of the topology model, where nodes represent device elements, with traversal starting from the selected subprocess and using a traversal strategy; and

[0007] For each of the at least one device element, if the influence of the device element on the industrial subprocess is greater than a first predetermined influence level, then the device element is added to the influence list of the device elements.

[0008] Methods can be used, for example, to quickly correct errors in industrial subprocesses, such as identifying erroneous behaviors in industrial subprocesses and / or the elements included in them. An impact list is a list of equipment elements that affect an industrial subprocess. An impact list can include 0, 1, ..., n elements. The impact list can be used, for example, for training of service personnel, for constructive changes to the facility or process under consideration, and / or for further equipment and / or procedures, such as procedures performing machine learning (ML). Effects on an industrial subprocess can include the flow of materials, energy, information, and / or other aspects that control or influence the subprocess. Effects are not necessarily related to flow direction, etc. For example, a container may be affected by a first valve filling the container and a second valve emptying the container; both valves affect the container, but their flow directions differ. A subprocess is an equipment element (or multiple equipment elements) that is part of an industrial facility or process. Simple examples of a subprocess can be a container, a valve, or a valve's control element and / or combination entity, such as a valve plus its control element, or even a larger combination of equipment elements. At least in some cases, logically speaking, if an industrial subprocess "looks" like one or more elements, it may be indistinguishable. For example, an electric motor that includes its control elements in a housing can be considered as an element, i.e., "a single electric motor", or as a combination of elements, i.e., "an electric motor having a control loop that includes multiple control elements and a backup power supply".

[0009] Topology models can be bidirectional graphs. Elements of a topology model can represent physical entities, such as components within a facility, like containers, pipes, sensors, actuators, and / or connections. They can represent the flow of materials, energy, information, and / or other aspects of the facility. Therefore, evaluating the results of a topology model can help correct errors and / or otherwise optimize a facility or parts thereof. Graphs can be represented graphically, as matrices, as XML (Extensible Markup Language) lists, and / or in other ways. For example, a topology model can be, for instance, a pipe and instrumentation diagram (P&ID) in a topology editor and / or process graphics tool. For example, a topology model can be generated automatically from one or more P&IDs, for example, by using a conversion tool or formatting a smart, standardized P&ID. Equipment elements represented by a topology model can be interconnected, for example, mechanically, by material and / or energy flows, and / or by logical causal relationships; that is, they may, for example, affect each other's state. Therefore, nodes in a topology model can represent equipment elements, including sub-processes, and edges in a topology model can represent interconnections between equipment elements. Topology models, and the control system configurations they represent, can be created using different tools in different formats. Especially in large facilities where the connected elements may be very distant and could expand to multiple pages on a graphical tool, at least some topology models can be challenging to display graphically. When considering a topology model, it can be difficult to determine whether a subprocess or device element is affected by another device element.

[0010] According to its representation, selecting a subprocess of interest in the topology model can include "clicking" a graphical element or otherwise marking it (e.g., by selecting an XML entry that represents its subprocess or aspect (e.g., property or attribute)).

[0011] The traversal begins with the selected subprocess and employs a traversal strategy, such as traversing all elements over the vocabulary using breadth-first search, depth-first search, a combination of these, and / or other strategies, such as heuristics or rule-based methods. This can include determining the direction of analysis for influencing features (i.e., forward, backward, bidirectional) to find influencing elements and signals. Other strategies (sometimes referred to as "causal analysis strategies") can be based on statistical methods, such as correlation, transfer entropy, or machine learning, such as pattern recognition and causal inference algorithms.

[0012] For each of at least one device element, check whether the device element affects the industrial subprocess with an influence level greater than a first predetermined influence level. The influence level can be a quantitative value, such as a maximum value (e.g., the maximum flow rate through a valve) or the maximum heating capacity of a heater, or a range or "fuzzy" value. The influence level can take into account whether the process is operating in "normal mode," "emergency mode," or another mode. For example, a safety valve might not be considered for one influence list describing "normal mode," but for another influence list describing "emergency mode." If the influence level is greater than the first predetermined influence level, the device element is added to the influence list affecting the device element. Depending on the subprocess, the first predetermined influence level can be, for example, zero or a small value, or, for example, a considerably high value for an emergency function. If no node or device element with an influence level greater than the first predetermined influence level is found during the iteration, the resulting influence list is empty.

[0013] Advantageously, this method ensures that only equipment elements that actually affect the subprocess are added to the impact list; less relevant values ​​are ignored. This can facilitate a quick understanding of certain parts of a large facility and / or help to quickly correct errors. The impact list can include unique identifiers for facility / process instruments, each of which can sequentially identify a set of signals. Once the impact list is constructed from the topology model, their corresponding signals in the control builder can be sent to data scientists, for example, to assess the situation and / or train machine learning models. Furthermore, the method can provide data scientists with a list of elements for training ML models against selected elements. As mentioned above, these can be based on their physical interconnections in the facility or production process (as found in the "topology model" on the path) and logical relationships (e.g., defined as "domain process rules" between elements). Using, for example, the domain process rules, the traversal algorithm can also stop at defined element (e.g., interconnection) properties or attributes, and thus can include "only relevant (or highly relevant)" elements in the impact list. Additionally, the method can search for signals in the control system that are associated with the properties or attributes of the topology elements.

[0014] Therefore, influence lists can lead to better training of ML models based on relevant equipment elements (or topological elements in the topology model) and / or their properties or attributes; thus, ML performance results can be enhanced. Additionally, influence lists can enhance the “interpretability” of the topology model by using process-related language that operators can understand, which may lead to a better understanding of facility behavior and improved training. Furthermore, this enables the capture of process expert knowledge, such as domain rules, as a single instance and reused in cross-facility analyses. Reuse can include “stopping indicators” in the traversal of the topology model and / or “filtering indicators” when collecting elements on the traversal of materials and information flows. This can also include approximate or linguistic quantification of influence relationships between process elements to capture process expert knowledge and / or automatic derivation of influence relationships using historical or simulation data.

[0015] In various embodiments, traversal strategies include breadth-first search, depth-first search, a combination thereof, and / or other strategies, such as heuristic and / or rule-based methods. Other strategies (sometimes referred to as "causal analysis strategies") may be based on statistical methods, such as correlation, transfer entropy, or machine learning, such as pattern recognition and causal inference algorithms.

[0016] In various embodiments, traversal is restricted to a selectable number of hops. A “hop count” is a measure of the “distance” from the currently traversed device element to the subprocess of interest, such as the number of edges between the currently traversed device element and the subprocess of interest. For example, if traversal is restricted to N = 2 hops, the topology model only considers device elements whose maximum distance to the currently traversed device element is 2 edges from the subprocess of interest. This “stopping metric” can be combined with one or more other “stopping metrics.” This advantageously provides an easy-to-apply stopping metric for methods with fast runtime.

[0017] In various embodiments, traversal stops for each branch at influencing device elements whose impact on the industrial subprocess is less than a second predetermined level. Note that the influencing device element is already part of the influence list. This advantageously addresses the problem that highly complex facilities may be affected by many device elements, but sometimes only to a negligible degree. This "stop index" can be combined with one or more other "stop indices".

[0018] In various embodiments, a selectable number of properties or attributes are selected, limited to the subprocess. These properties or attributes can be, for example, temperature, energy flow or material flow, material density, pressure in the container, pressure in the equipment, material level, flow rate, and / or other attributes. For example, a heater might affect the temperature of a container but not its material flow. As another example, a valve might affect the material flow of a container but not its temperature. Limiting itself to a single property or attribute can further reduce complexity. Additionally, several lists of the effects of different attributes can be combined in a targeted and purposeful manner.

[0019] In various embodiments, the degree of influence is quantified by values, value ranges, fuzzy values, lower and / or upper bound approximations, particularly using rough sets and / or other quantization values. An example of a fuzzy value could be a set of values ​​{"high"; "medium"; "low"} for container temperature. At least some fuzzy values ​​can be mapped to value ranges.

[0020] In each embodiment, the degree of influence is defined by rules. Examples of such rules could be expressed as follows:

[0021] If the inlet valve V of separator S is open and each outlet valve is closed

[0022] Then the voltage level Ls in the separator rises.

[0023] Rules can be created from historical data patterns, machine learning programs, ML, and / or experts. For example, rules from different sources can be combined. Rules can also be enriched by quantifying relationships using fuzzy logic, where linguistic values ​​are used to represent quantities instead of precise numerical values. Rule specifications can be specified using a domain-specific language or can be supported by tools that allow for direct visualization on the topology model.

[0024] The method for automating rule specification can be based on deriving rules from historical data patterns and allowing (multiple) experts to review, accept, edit, and / or reject (multiple) rules. Additionally or alternatively, the automatically derived rules can be used and combined with expert feedback to enhance the quality of the automatically derived rules.

[0025] In some embodiments, the degree of influence is defined by relevance. Examples may include filtering techniques, such as filtering based on statistical indicators, such as filtering by the relevance of a property or attribute to the target variable.

[0026] In some embodiments, the degree of influence is defined by the ML wrapper method. The wrapper method solves ML problems, for example, by using different sets of properties and searching for the optimal set of properties.

[0027] In some embodiments, the degree of influence is defined by an embedded ML method, particularly by regularization and / or decision trees. In the embedded method, the ML algorithm itself identifies the relevant features. Examples may include decision tree learning algorithms such as ID3 or regression methods such as Lasso.

[0028] In various embodiments, the method further includes the step of constructing an influence graph from an influence list, wherein the influence graph includes nodes representing influencing equipment elements and weighted edges representing the degree to which the influencing equipment elements affect industrial subprocesses. Influence graphs can advantageously simplify the assessment of functional interrelationships within a facility. However, they may omit more informational details, such as the topology model, including the location of equipment elements, their dimensions, their form, etc. Influence graphs and topology models can also be combined, for example, by "overlay," where the influence graph and topology model are displayed at different levels and / or semi-transparently.

[0029] One aspect relates to a computer program product comprising instructions that, when executed by a computer, cause the computer to perform the methods described above and / or below.

[0030] One aspect relates to a computer-readable storage medium in which a computer program or computer program product as described above is stored.

[0031] One aspect relates to computer-based systems configured to perform the methods described above and / or below.

[0032] One aspect relates to industrial facilities or processing systems, including computer-based systems as described above and / or below.

[0033] One aspect involves using methods described above and / or below to provide a list of affected device elements.

[0034] One aspect involves using a list of affected equipment elements, as described above and / or below, or an influence graph, as described above and / or below, for training a machine learning (ML) program. Once the list of influences of the affected equipment elements affecting an industrial subprocess is available, their corresponding properties or attributes in the control builder can be sent to data scientists and / or used to train a machine learning model. Furthermore, this can lead to solving ML optimization problems, including "at which level to stop traversal, i.e., when the selected features are sufficient to train an ML model." This optimization can be done automatically (e.g., in the sense of Automated Machine Learning (AutoML)). In other words, the method can automatically find the optimal set of features by finding, for example, the best "traversal level" or "stopping criterion" applicable to the topology graph of an industrial facility. This can provide, for example, how AutoML can be implemented in the industrial domain, such as methods for automatically selecting attributes of interest. Attribute selection can, for example, be based on a wrapper approach, where the list of attributes can be expanded from a small number of attributes (e.g., based on proximity measures in the topology) to a larger number of features, thereby finding a good trade-off between the complexity, runtime, and accuracy of the elements to be considered in a targeted manner. In each expansion step, different subsets of the current candidate features can be tested. AutoML can further optimize the hyperparameters of the ML training process.

[0035] For further illustration, the present invention is described through embodiments shown in the figures. These embodiments are considered to be examples only and not limitations. Attached Figure Description

[0036] The attached image depicts:

[0037] Figure 1 An exemplary topology model of a portion of the facility is schematically shown;

[0038] Figure 2 An exemplary model of a collaborative representation of a facility is schematically shown;

[0039] Figure 3 This is a flowchart based on one embodiment. Detailed Implementation

[0040] Figure 1A (small) exemplary topology model of a portion of an industrial facility and process is schematically shown. In the illustrated topology model, container B101 is selected as the subprocess of interest. Traversal is limited to N=2 hops. Relevant edges are marked with thick lines. The selection is limited to two properties of the subprocess: material (fluid) flow and temperature. Therefore, the list of influences includes valve V105, pump P101, valve V104, valve V102, container B102 for material flow, and heat exchanger E104 for temperature. For each device, numerous measurements can be performed to collect its state. For example, container B102 may have level sensors B113 and B114, which transmit signals to the control system related to “material flow”. Therefore, the list of influences for the device elements includes the relevant “material flow” elements. These can serve as inputs for constructing an ML model related to “material flow”.

[0041] Figure 2 An exemplary model of a collaborative representation of industrial facilities and processes is schematically illustrated. For example, a person (e.g., a data scientist) might request a list of influences of equipment elements affecting an industrial subprocess. Therefore, the person can (step 1) use a topology model to select a subprocess of interest and (step 2) send a query related to it. This query can be initiated (step 3) by traversing the topology model, where the traversal begins with the subprocess of interest and uses a traversal strategy, such as a breadth-first search using the topology model. As an intermediate step (step 4), all elements of the breadth-first search are shown; this may include element IDs. Furthermore, rules can be used to define the degree of influence (step 5). Then (step 6), the selection can be restricted to one or more properties or attributes of the subprocess. This (step 7) may include identifying sensors and / or signals related to the properties or attributes. These results (step 8) can be collected and (step 9) output as an influence list and / or influence graph. After training an ML model based on the influence list (step 10), the results are reported (step 11) and shown (step 12) as a localization within the topology model. Note that this may not be possible in every embodiment. Figure 2 All steps described herein. Furthermore, some of the described steps may be used solely to demonstrate the progress of the method to humans, for example, to enable human-centered decision-making and / or to increase trust in the method.

[0042] therefore, Figure 2The process initiated by a data scientist can be summarized as follows: this process requires obtaining a relevant feature set of the selected elements from the topology model from which the data scientist is interested in training and building the ML model. As a possible further step, the data scientist can, for example, obtain support for interpreting the ML results based on specified logical rules and physical interconnections. Furthermore, the mapping from topology elements to control system signals can be accomplished based on the control system module type used to implement the element; that is, according to the specific module type used to implement a particular element (e.g., a PID loop), a configured subset of signals can be selected. Additionally, the system can maintain a database of known signal selection configurations for a library of control module types and allow system users to add unknown control module types to this database when encountered.

[0043] Figure 3 A flowchart 20 according to one embodiment is shown. In step 22, a subprocess of an industrial facility or process is selected in a topology model. A subprocess may be a device element that is part of an industrial facility or process, and the topology model may be a graph whose nodes represent device elements and whose edges identify interconnections between device elements. In step 24, the topology model is traversed, where the traversal begins with the selected subprocess and uses a traversal strategy. In step 26, for each device element, the degree of influence is checked, i.e., the degree to which the device element influences the industrial subprocess is quantitatively measured. If the device element influences the industrial subprocess with an influence greater than a first predetermined degree, then in step 27, the device element is added to the influence list affecting the device elements. Otherwise, no action is taken (step 28). The influence list is a list of influencing device elements that affect the industrial subprocess of interest. The influence list may include properties, attributes, signals, and / or other entities of each device element. The influence list can be used as input for training a machine learning program.

Claims

1. A computer-implemented method for providing a list of influences of device elements affecting an industrial subprocess, wherein influencing the industrial subprocess includes controlling or otherwise affecting material flows, energy flows, information flows, and / or other aspects of the subprocess, the method comprising the steps of: Select the subprocess in the topology model, where the topology model is the Piping and Instrumentation Diagram (P&ID). The subprocess mentioned therein is an equipment element that is part of an industrial facility or process, and The topology model is a graph, where nodes represent device elements and edges represent interconnections between device elements. Traverse the nodes of the topology model, wherein the traversal begins with a selected subprocess and uses a traversal strategy; as well as For each of at least one device element, if the influence of the device element on the industrial subprocess is greater than a first predetermined influence level, then the device element is added to the influence list of the affected device elements for rapid repair of errors in the industrial subprocess.

2. The method according to claim 1, The traversal strategy mentioned above includes: Breadth-first search, depth-first search, hybrid breadth-first search and depth-first search, heuristic, rule-based and / or statistical methods.

3. The method according to any one of claims 1 to 2, The traversal is restricted to a selectable number of hops.

4. The method according to any one of claims 1 to 2, The traversal stops at each branch where the influence of the industrial subprocess on the affected device element is less than a second predetermined degree.

5. The method according to any one of claims 1 to 2, The choice is limited by the nature of the number of choices available for the subprocess.

6. The method according to any one of claims 1 to 2, The degree of influence is quantified by values, value ranges, fuzzy values, lower and / or upper bound approximations, and in particular by using rough sets and / or by other quantization values.

7. The method according to any one of claims 1 to 2, The degree of influence mentioned above is defined by rules.

8. The method according to claim 7, The rules mentioned therein are created by machine learning programs (ML) and / or experts based on historical data patterns.

9. The method according to any one of claims 1 to 2, The degree of influence mentioned above is defined by correlation.

10. The method according to any one of claims 1 to 2, The degree of influence is defined by the ML wrapper method.

11. The method according to any one of claims 1 to 2, The degree of influence is defined by ML embedding methods, particularly by regularization and / or decision trees.

12. The method according to any one of claims 1 to 2, further comprising the following step: Construct an impact graph from the impact list. The influence diagram includes nodes representing the influencing device elements, and The influence diagram includes weighted edges, which represent the degree to which the influencing device element affects the industrial subprocess.

13. A computer program product comprising instructions that, when executed by a computer, implement the method according to any one of claims 1 to 12.

14. A computer-readable storage medium having instructions stored thereon that, when executed by a processor, implement the method according to any one of claims 1 to 12.

15. An industrial facility or processing system comprising a computer-based system configured to perform the method according to any one of claims 1 to 12.

16. The use of the method according to any one of claims 1 to 12 for providing a list of affected device elements.

17. The use of the influence graph according to claim 12 for training a machine learning program (ML).

Images