Automated creation of digital twin using graph-based industrial data

Abstract

A computer-implemented method for automatically creating a digital twinning of an industrial system having one or more devices includes accessing a triple repository including an aggregated ontology of graph-based industrial data synchronized with the one or more devices. A triple repository is queried for a specified device to extract structure information of the specified device defined by a tree including a hierarchy of nodes from the graph-based industrial data. Each node is assigned a neural network element based on a mapping between the node type and a predefined neural network element. Based on the tree topology, the assigned neural network elements are combined to create a digital twin neural network. The triple repository is then queried to extract real-time process data collected from a specified device at runtime from the graph-based industrial data and to adjust parameters of the digital twin neural network using the extracted real-time process data.

Description

Technical Field

[0001] The present disclosure generally relates to industrial automation and, in particular, to systems and methods for automatically creating digital twins of industrial systems using graph-based industrial data. Background Art

[0002] An increasing number of industrial automation engineering tools may incorporate digital twins of real-world industrial systems. Using digital twins to provide simulation data can be an important aspect of automatic control and diagnosis. However, creating digital twins that can accurately replicate industrial systems, such as factories or complex machines with various structural components, is typically both difficult and time-consuming. To create a digital twin, a functional representation that can represent the internal structure of the real-world industrial system must be created. This representation can be in the form of a machine learning (ML) model, such as a neural network, a Bayesian network, etc.

[0003] According to current practice, the architecture of the ML model that describes the digital twin functional representation is typically manually designed (hand-coded) by experienced personnel and then trained using real-world data. There is prior art literature that describes such manually designed architectures informed by expert design. However, such design may be highly dependent on experienced experts who have had the rare opportunity to participate in multiple such design projects and have witnessed firsthand what works and what does not. The process is both time-consuming and costly and may not be suitable for rapid development. Summary of the Invention

[0004] Briefly, aspects of the present disclosure provide a computer-implemented system and method that can utilize graph-based industrial data to automatically map the machine learning model architecture of a digital twin and adjust such a digital twin to replicate a real-world industrial system. The graph-based industrial data includes structural information of individual devices of an industrial system that can be extracted to map the digital twin neural network of a specified device. The graph-based industrial data also integrates real-time process data collected from individual devices at runtime, and these data can be extracted to adjust the digital twin neural network of the specified device.

[0005] According to a first aspect of the present disclosure, there is provided a computer-implemented method for automatically creating a digital twin of an industrial system including one or more devices. The method includes querying a triple store that includes an aggregated ontology of graph-based industrial data synchronized with the one or more devices to extract, for a specified device, structural information of the specified device defined by a tree of levels including nodes. The method further includes traversing the tree to identify the node types of individual nodes and, based on a mapping between the node types and predefined neural network elements, assigning a corresponding neural network element to each individual node. The method also includes combining the corresponding neural network elements based on the topology of the tree to create a digital twin neural network. The method also includes training the digital twin neural network by querying the triple store to extract real-time process data collected from the specified device at runtime from the graph-based industrial data and using the real-time process data to adjust the learnable parameters of the digital twin neural network.

[0006] Other aspects of the present disclosure relate to computing systems and computer program products including instructions executable by a processor to perform the above method and its alternative implementations.

[0007] Additional technical features and benefits can be achieved through the techniques of the present disclosure. Embodiments and aspects of the present disclosure are described in detail in the present invention and are considered to be part of the claimed subject matter. For a better understanding, reference is made to the detailed description and the drawings. Brief Description of the Drawings

[0008] The above and other aspects of the present disclosure can be best understood when the following detailed description is read in conjunction with the drawings. To easily identify the discussion of any element or action, the most significant digit in the reference numeral refers to the drawing number in which that element or action is first introduced.

[0009] Figure 1 A simplified system architecture is shown for implementing a method for automatically creating a digital twin of an industrial system using graph-based industrial data according to an exemplary embodiment.

[0010] Figure 2 A schematic diagram is shown of mapping a digital twin neural network from structural information extracted from graph-based industrial data.

[0011] Figure 3 An example of a computing system is shown according to the disclosed embodiment that is capable of supporting the automatic creation of a digital twin of an industrial system using graph-based industrial data. Detailed Description

[0012] Industrial automation system components are typically interconnected via dedicated networks using standard industrial protocols to enable access and data exchange. The development of current and future automation systems is increasingly focused on exchanging semantically rich information, aiming to enable flexible manufacturing scenarios. The Open Platform Communications Unified Architecture (OPC UA) is an industrial standard protocol of the OPC Foundation for manufacturer-independent communication, which aims to exchange industrial data, especially for automation purposes. In the field of factory automation, OPC UA is one of the most promising standards for device communication, which can elevate low-level signal exchange schemes to the semantic level and contribute to realizing flexible manufacturing scenarios. The information model of OPC UA has a semantically rich, graph-based data structure, which is specifically designed for automation purposes.

[0013] Embodiments of the present disclosure utilize graph-based industrial data (such as data obtained from the OPC UA information model) to model and tune a machine learning-based digital twin of an industrial system. The basic idea of the disclosed embodiments utilizes the fact that graph-based industrial data, especially industrial data obtained from the OPC UA information model, can integrate structural information and real-time process data ("real-time data"), thus providing a viable ontology that can be semantically mapped to a machine learning model architecture. Therefore, by leveraging the existing semantic data of factories and machines, the cost and time of developing digital twins can be significantly reduced.

[0014] The disclosed embodiments include a digital twin mapping module that can extract the structural information of a specified device from graph-based industrial data and automatically map it to a neural network representation of the device ("digital twin neural network"). The structural information can define the overall structure of an individual device or machine in a factory by expressing the hierarchy of various components of the device and how one component takes another component as input. Thus, the structural information can define the dependencies between the components of the device. The disclosed digital twin mapping module can reduce costs by significantly reducing expert involvement and further improve the reliability of the digital twin by using actual device structural information.

[0015] The disclosed embodiments further include a digital twin training module that can adjust the automatically generated digital twin neural network by extracting real-time data collected from a specified device at runtime and integrated into the graph-based industrial data. The disclosed digital twin training module can automate and simplify data acquisition and further reduce expert involvement.

[0016] Now turning to the drawings, Figure 1shows a system architecture for implementing a method of automatically creating a digital twin of an industrial system 100 using graph-based industrial data according to an exemplary embodiment. The various modules described in the present invention (including the query module 104, the digital twin mapping module 114, and the digital twin training module 116, including their components) can be implemented in a computing environment in various ways, for example, as hardware and programming. The programming for modules 104, 114, 116 can take the form of processor-executable instructions stored on a non-transitory machine-readable storage medium, and the hardware for the modules can include a processor for executing these instructions. The processing capabilities of the systems, devices, and modules described in the present invention (including the query module, the digital twin mapping module 114, and the digital twin training module 116) can be distributed among multiple system components, such as among multiple processors and memories, optionally including multiple distributed processing systems or cloud / network elements.

[0017] Referring Figure 1 , the industrial system 100 can include at least one (usually multiple) device 102. The device 102 can include, for example, machines in a factory floor, such as robots, CNC machines, etc. The device 102 can be connected to an industrial network. The query module 104 can be located within the aggregation layer of the industrial network or hierarchically assigned to the aggregation layer, for example, implemented by an edge or cloud application or integrated within an edge or cloud controller. As shown, the query module 104 can include an aggregation address space 106 communicatively connected to the device 102, a triple store 108 including an aggregation ontology of graph-based industrial data synchronized with the device 102 obtained through the aggregation address space 106, a query engine 110 for querying the triple store 108, and one or more endpoints 112 that can serve as a logical query interface assigned for interacting with a client system. The digital twin mapping module 114 and the digital twin training module 116 can be located in the client system 118, which can exchange query messages with the query module 104 in a query language supported by the assigned endpoints 112. The digital twin mapping module 114 and the digital twin training module 116 can extract structural information 120 and real-time process data 122 respectively by using the query engine 110 to query the triple store 108 to automatically create and adjust a digital twin neural network 124 as disclosed in the present invention.

[0018] Industrial entities, such as device 102, are typically equipped with sufficient storage, communication, and computing resources. According to the disclosed embodiments, each device 102 can include a corresponding OPC UA server running therein. Thus, when the device 102 is connected to an industrial network, the device can expose its structural information and real-time data through the corresponding UPC UA server. The OPC UA server of an individual device 102 can be communicatively connected to an aggregated address space 106. The aggregated address space 106 can be synchronized with the device 102, for example, through an aggregator server (not shown). The aggregated address space 106 can provide access to the OPC UA information model of each device 102, including the structural information and real-time data accumulated and delivered by the corresponding device 102.

[0019] In the OPC UA information model, each entity in the address space is a node. A node is the basic unit of data in the OPC UA address space and provides a standard way for an OPC UA server to represent an object to an OPC UA client. The OPC UA information model can provide the following hierarchical structure:

[0020] On the first or lowest layer (referred to as the meta layer), basic entities can be defined, such as node classes, attributes, references, etc.

[0021] The second layer (referred to as the OPC UA base layer) can be provided by the OPC Foundation itself. This layer can include specifications for base VariableTypes, server types, engineering units, etc.

[0022] In the third layer, at least one OPC UA companion specification can be used to define domain-specific models or schemas that extend the OPC UA model. Companion specifications are typically developed by domain experts, standardization bodies, or industrial machine vendors.

[0023] The fourth layer (referred to as the extension layer) can host original equipment manufacturer (OEM)-specific schema extensions written by OEMs, including, for example, a device vendor information (DVI) model that includes a device type description, a machine vendor information (MVI) model that includes a machine type description, and a machine user information (MUI) model that includes a process type or plant element type.

[0024] Finally, the fifth layer (referred to as the instance layer) located at the top of the hierarchical information model can include a device information model (DIM), i.e., an instance model for describing the structure and data items (including real-time data) of an individual device based on the schema defined in the layer.

[0025] The triple repository 108 includes a graph database that can store data as statements in a subject-predicate-object format (triples). An aggregated ontology of graph-based industrial data in the triples 108 can be derived based on the OPC UA information model provided by the aggregated address space 106. To facilitate querying in the semantically rich OPC UA information model, the OPC UA information model can be converted into a suitable target ontology. According to the disclosed embodiments, the aggregated ontology of industrial data in the triple repository 108 can include a Resource Description Framework (RDF) graph obtained by mapping the OPC UA information model provided by the aggregated address space 106 to a target ontology representation (within each layer) represented by a Web Ontology Language (such as OWL). Details of such mapping have been described in the published document WO2020104019 A1 based on an international patent application filed by the applicant, the content of which is incorporated herein by reference in its entirety.

[0026] The ontology included in the triple repository 108 can include a static part and a dynamic part. The static part can define hierarchical information of nodes (e.g., type-hierarchy), which can be generated by converting the OPC UA information model provided by the aggregated address space 106 into an RDF representation (i.e., into triples) as a result of an OWL mapping. Thus, the hierarchical information defined in the static part can include the structural information of the individual device 102. If the underlying OPC UA graph structure is updated, the static part can be modified, for example, when a new device is added to the industrial network, this update can be triggered. The dynamic part can be used to provide actual values (e.g., in OPC UA, the value attribute of a variable node such as temperature), which can be directly accessed on demand through the aggregated address space 106. In other words, the dynamic part can include the dynamic allocation of data values (i.e., real-time data), and the data values are collected from the individual device 102 in response to a query at runtime and integrated into the aggregated ontology in the triple repository 108 when such a query occurs.

[0027] To avoid the high complexity introduced by querying the semantic descriptions scattered in the aggregated ontology within the triple store 108, a suitable query language, such as SPARQL, can be used to query the triple store 108. SPARQL is the recursive acronym for SPARQL Protocol and RDF Query Language. According to the disclosed embodiments, the query engine 110 can be configured to execute SPARQL query requests transmitted by the endpoint 112 for the triple store 108. In one exemplary embodiment, the query engine 110 can be implemented using Apache Jena (an open-source semantic web framework for Java) together with Fuseki (a SPARQL query engine with an additional web interface supporting SPARQL queries). The endpoint 112 can be configured to exchange query messages in the query language SPARQL with the client system 118, or to convert query messages formulated by the client system 118 in different query languages into SPARQL. For a detailed description of using SPARQL to query the OPC UA information model, reference can be made to the published document WO 2020200404A1 based on an international patent application filed by the applicant, the content of which is incorporated herein by reference in its entirety.

[0028] To create a digital twin of the device 102 connected to the industrial network, the digital twin mapping module 114 can, for example, submit a query to the triple store 108 through the SPARQL interface as described above to extract the structural information 120 specifying the device 102. The structural information 120 can be defined by a tree including a hierarchy of nodes. For illustration, the structure of a robot can be defined by a hierarchy of nodes, where the robot itself defines the root node, and its components are represented as hierarchies (e.g., the robot's arm, the power train of the arm, the motors of the power train, the sensors associated with the motors, etc.), such that each component is identified as a node in the hierarchy.

[0029] Nodes can include ObjectTypes, ConfigurationTypes, and VariableTypes. The node types can be defined in the companion specification of the OPC UA information model and mapped to OWL classes in the RDF graph. The actual hierarchical structure and data values of the specified device can be defined in the instance layer. The query of the digital twin mapping module 114 can occur in the instance layer. For example, the root node in the instance layer can be located by using the device name or identifier specified in the query, and then the child nodes related to the hierarchy of the root node can be determined in sequence to retrieve the instance tree of the specified device.

[0030] The digital twin mapping module 114 can traverse the retrieved instance tree to identify the node types of individual nodes. The nodes of the instance tree can include one or more types of component nodes representing components of a specified device (e.g., a robotic arm, a powertrain, a motor, a software object, etc.), one or more types of sensor nodes including sensor data associated with the specified device (e.g., a motor temperature sensor), and one or more types of configuration nodes including configuration data of the specified device (e.g., the speed of a robotic arm). The type information of each node can be determined by traversing the instance tree in the direction from the leaf node to the root node. Generally, sensor nodes can appear as leaf nodes, and component nodes can appear as intermediate nodes of the instance tree. A configuration node refers to a node that configures the behavior of a tree-shaped topological element. Configuration nodes can generally appear as leaf nodes, but in principle can also appear as intermediate nodes. For example, in the case of an OPC UA RDF graph, when traversing the instance tree, sensor nodes can be identified as nodes represented by variable types that can be filled by a value attribute (e.g., AnalogTypes), component nodes can be identified as nodes represented by ObjectTypes, and configuration nodes can be identified as nodes represented by ConfigurationTypes.

[0031] After identifying the node types, the digital twin mapping module 114 can assign corresponding neural network elements to each individual node based on the mapping between the node types and predefined neural network elements. For each node type described in the companion specification, the mapping can include stored corresponding neural network structure elements. The neural network elements for a given node type can include, for example, a layer of neuron nodes, or can even include a small neural network. The architecture of the neural network elements for each node type in the companion specification (e.g., the number of layers of neuron nodes, the number of neuron nodes in each layer, the connections between nodes, etc.) can be determined by heuristic methods, such as based on the experiments of domain experts, and stored in the mapping. Once the mapping is created, the digital twin mapping module 114 can use this mapping as a lookup table to assign neural network elements to the nodes of any queried OPC UA device.

[0032] Next, the digital twin mapping module 114 can use the topology of the instance tree to combine the corresponding neural network elements assigned to the individual nodes of the instance tree to create a digital twin neural network 124. As described above, the instance tree can define the overall structure of a specified device by expressing the hierarchy of the various components of the device and how one component takes another component as input. Since the instance tree can define the dependencies between the components of a specified device, the topology of the instance tree can be utilized to combine the neural network elements corresponding to each node of the instance tree to create the digital twin neural network 124. For example, according to the disclosed embodiments: the neural network element corresponding to the sensor node can form the output layer of the digital twin neural network 124; the neural network element corresponding to the configuration node can form the input layer of the digital twin neural network 124; and the neural network element corresponding to the component node can form one or more hidden layers of the digital twin neural network 124. To appropriately process dynamically changing real-time data, the digital twin neural network 124 can include a recurrent neural network (RNN) architecture.

[0033] Thus, the disclosed digital twin mapping module 114 can use the vast amount of information obtainable from semantically rich graph-based industrial data (such as OPC UA RDF graphs) to intelligently create a digital twin neural network for a specified device that is both easy to train and provides accurate results. In the past, for each new device, a digital twin neural network had to be manually created (hand-coded) from scratch. According to the disclosed embodiments, the mapping can be coded once to establish a one-to-one correspondence between the node types and the neural network elements. Once the mapping is established, the digital twin mapping module 114 can use the mapping to automatically create the digital twin neural network of any specified device, as described above, without further manual operation.

[0034] Figure 2Illustrates how an instance of a digital twin neural network 124 is created from an instance tree 120 extracted from an RDF graph using a digital twin mapping module 114. As described above, the instance tree 120 can be retrieved by locating the root node NR in the instance layer of the RDF graph using the device name identifier specified in the query (e.g., ABC robot), and then sequentially determining the child nodes associated with the hierarchy of the root node NR. Then, the tree 120 is traversed in the direction from the leaf nodes to the root node to identify the node types of the nodes. In the illustrated example, the node Nc represents a configuration node, the node Ns represents a sensor node, and the node Ni represents a component node. For each node, based on its node type, a predefined neural network element is assigned using the stored mapping described above. Then, the assigned neural network elements are combined using the topology of the instance tree 120 to create the digital twin neural network 124, such that the configuration node Nc forms the input layer of the digital twin neural network 124, the sensor node Ns forms the output layer of the digital twin neural network 124, and the component node Ni forms the intermediate or hidden layer of the digital twin neural network 124. The description shown is simplified. For example, in some embodiments, the number of hidden layers of the digital twin neural network 124 can correspond to the number of levels of the component nodes Ni in the instance tree 120.

[0035] Continuing to refer to Figure 1 , the digital twin training module 116 can use real-time data from the specified device 102 to automatically adjust the automatically generated digital twin neural network 124. During the training process, the digital twin training module 116 can, for example, query the triple store 108 through the SPARQL interface described above to extract real-time process data 122 (real-time data) collected from the specified device 102 at runtime from graph-based industrial data. According to the disclosed embodiments, the OPC UA server of each device 102 can communicate the data values of such real-time process data during the operation of the device 102. In response to the query submitted by the digital twin training module 116, the data values of the specified device 102 can be dynamically integrated into the aggregated ontology of the graph-based industrial data at runtime through the aggregated address space 106, whereby the training process can be fully or substantially automated.

[0036] The extracted real-time process data 122 can include configuration data and sensor data. The real-time process data can be stored as time-series data in graph-based industrial data. The training process can involve using the extracted real-time process data 122 to adjust the learnable parameters (e.g., weights, biases) of the digital twin neural network 124, which can include an RNN. According to the disclosed embodiments, the digital twin training module 116 can use the configuration data to define the inputs of the digital twin neural network 124 and use the sensor data as the ground truth. The training process can include iteratively using the learnable parameters of the digital twin neural network 124 to generate an output based on the input configuration data and adjusting the learnable parameters to reduce the error between the output and the ground truth defined by the sensor data. These steps can be continuously executed over a number of epochs until a convergence criterion is met. For example, the convergence criterion can be met after a predefined number of epochs or when the error function is minimized.

[0037] Figure 3 FIG. shows an example of a computing system 300 according to the disclosed embodiments, which can support the automatic creation of a digital twin of an industrial system using graph-based industrial data. The computing system 300 includes at least one processor 310, which can take the form of a single or multiple processors. The processor 310 can include a central processing unit (CPU), a graphics processing unit (GPU), a neural processing unit (NPU), a microprocessor, or any hardware device suitable for executing instructions stored on a memory including a machine-readable medium. The computing system 300 also includes a machine-readable medium 320. The machine-readable medium 320 can take the form of any non-transitory electronic, magnetic, optical, or other physical storage device for storing executable instructions, such as Figure 3 the digital twin mapping instructions 322 and the digital twin training instructions 324 shown. Thus, the machine-readable medium 320 can be, for example, a random access memory (RAM), such as a dynamic RAM (DRAM), flash memory, spin transfer torque memory, an electrically erasable programmable read-only memory (EEPROM), a storage drive, an optical disc, etc.

[0038] The computing system 300 can execute the instructions stored on the machine-readable medium 320 via the processor 310. Executing these instructions (e.g., the digital twin mapping instructions 322 and the digital twin training instructions 324) can cause the computer system 300 to implement any of the technical features described in the present invention, including any features of the digital twin mapping module 114 and the digital twin training module 116 described above.

[0039] The systems, methods, apparatuses, and logics described above (including the digital twin mapping module 114 and the digital twin training module 116) can be implemented in a variety of different ways, in a combination of a variety of different hardware, logics, circuits, and executable instructions stored on a machine-readable medium. For example, these modules can include circuits in a controller, a microprocessor, or an application specific integrated circuit (ASIC), or can be implemented with discrete logic or components or a combination of other types of analog or digital circuits, combined on a single integrated circuit or distributed among multiple integrated circuits. A product, such as a computer program product, can include a storage medium and machine-readable instructions stored on the medium, which, when executed in an endpoint, a computer system, or other device, cause the device to perform any of the operations according to the above description, including operations according to any features of the digital twin mapping module 114 and the digital twin training module 116. The computer-readable program instructions described in the present invention can be downloaded from a computer-readable storage medium to a corresponding computing / processing device, or downloaded to an external computer or an external storage device via a network (such as the Internet, a local area network, a wide area network, and / or a wireless network).

[0040] The processing capabilities of the systems, apparatuses, and modules (the digital twin mapping module 114 and the digital twin training module 116) described in the present invention can be distributed among multiple system components, such as among multiple processors and memories, optionally including multiple distributed processing systems or cloud / network elements. Parameters, databases, and other data structures can be stored and managed separately, can be combined into a single memory or database, can be logically and physically organized in a variety of ways, and can be implemented in a variety of ways, including data structures such as linked lists, hash tables, or implicit storage mechanisms. Programs can be part of a single program (e.g., a subroutine), independent programs distributed across multiple memories and processors, or implemented in a variety of different ways, such as in the form of a library (e.g., a shared library).

[0041] Although the present disclosure has been described with reference to specific embodiments, it should be understood that the embodiments and variations shown and described are for illustrative purposes only. Those skilled in the art can implement modifications to the current design without departing from the scope of the patent claims.

Claims

1. A computer-implemented method for automatically creating a digital twin of an industrial system including one or more devices, the method comprising: Querying a triple store that includes an aggregated ontology of graph-based industrial data synchronized with the one or more devices to extract, for a specified device, structural information of the specified device defined by a tree including a hierarchy of nodes, Traversing the tree to identify the node types of individual nodes and, based on a mapping between the node types and predefined neural network elements, assigning a corresponding neural network element to each individual node, Combining the corresponding neural network elements based on the topology of the tree to create a digital twin neural network, and Training the digital twin neural network by querying the triple store to extract real-time process data collected from the specified device at runtime from the graph-based industrial data and using the real-time process data to adjust learnable parameters of the digital twin neural network.

2. The method according to claim 1, wherein The nodes of the tree defining the structural information include component nodes containing one or more types of components representing components of the specified device, sensor nodes including one or more types of sensor data associated with the specified device, and configuration nodes including one or more types of configuration data of the specified device.

3. The method according to claim 2, wherein, Based on the topology of the tree, the corresponding neural network elements assigned to the individual nodes are combined such that: The neural network elements corresponding to the one or more types of sensor nodes form an output layer of the digital twin neural network, The neural network elements corresponding to the one or more types of configuration nodes form an input layer of the digital twin neural network, and The neural network elements corresponding to the one or more types of component nodes form one or more hidden layers of the digital twin neural network.

4. The method according to claim 3, wherein, The real-time process data extracted from the graph-based industrial data includes configuration data and sensor data.

5. The method according to claim 4, wherein, Iteratively, the learnable parameters of the digital twin neural network are adjusted by performing the following steps: Defining an input to the digital twin neural network using the configuration data, Generating an output using the learnable parameters of the digital twin neural network and adjusting the learnable parameters to reduce an error between the output and a ground truth defined by the sensor data.

6. The method according to any one of claims 1 to 5, wherein The real-time process data is stored as time series data in the graph-based industrial data.

7. The method according to claim 6, wherein The digital twin neural network includes a recurrent neural network.

8. The method according to any one of claims 1 to 7, wherein Each of the one or more devices includes a respective Open Platform Communications Unified Architecture (OPC UA) server communicatively connected to an aggregated address space, wherein the aggregated ontology of the graph-based industrial data is derived based on an OPC UA information model provided by the aggregated address space.

9. The method according to claim 8, wherein, The aggregated ontology of the graph-based industrial data includes a Resource Description Framework (RDF) graph obtained by converting the OPC UA information model provided by the aggregated address space into a target ontology.

10. The method according to claim 9, wherein, Query the Resource Description Format (RDF) graph via the SPARQL interface.

11. A non - transitory computer - readable storage medium comprising instructions that, when processed by a computing system, configure the computing system to perform the method according to any one of claims 1 to 10.

12. A computing system for automatically creating a digital twin of an industrial system including one or more devices, the computing system comprising: One or more processors, A non - transitory memory in communication with the one or more processors, the non - transitory memory including algorithmic modules executable by the one or more processors, the algorithmic modules including: A digital twin mapping module configured to: Query a triple store that includes an aggregated ontology of graph - based industrial data synchronized with the one or more devices to extract, for a specified device, the structural information of the specified device defined by a tree including levels of nodes, Traverse the tree to identify the node types of individual nodes and, based on a mapping between the node types and predefined neural network elements, assign a corresponding neural network element to each individual node, and Based on the topology of the tree, combine the corresponding neural network elements to create a digital twin neural network, And A digital twin training module configured to: Query the triple store to extract real - time process data collected from the specified device at runtime from the graph - based industrial data and use the real - time process data to adjust the learnable parameters of the digital twin neural network.

13. The computing system according to claim 12, wherein, The nodes of the tree defining the structural information include component nodes containing one or more types representing components of the specified device, sensor nodes including one or more types of sensor data associated with the specified device, and configuration nodes including one or more types of configuration data of the specified device.

14. The computing system according to claim 13, wherein, The digital twin mapping module is configured to, based on the topology of the tree, combine the corresponding neural network elements assigned to the individual nodes such that: The neural network elements corresponding to the one or more types of sensor nodes form the output layer of the digital twin neural network, The neural network elements corresponding to the one or more types of configuration nodes form the input layer of the digital twin neural network, and The neural network elements corresponding to the one or more types of component nodes form one or more hidden layers of the digital twin neural network.

15. The computing system according to claim 14, wherein, The digital twin training module is configured to cause the real - time process data extracted from the graph - based industrial data to include configuration data and sensor data.

16. The computing system according to claim 14, wherein, The digital twin training module is configured to, over multiple iterations, adjust the learnable parameters of the digital twin neural network by performing the following steps: Define the input of the digital twin neural network using the configuration data, Generate an output using the learnable parameters of the digital twin neural network, and adjust the learnable parameters to reduce the error between the output and the ground truth defined by the sensor data.

17. The computing system according to any one of claims 12 to 16, wherein The real-time process data is stored as time series data in the graph-based industrial data.

18. The computing system according to claim 17, wherein, The digital twin neural network includes a recurrent neural network.

19. The computing system according to any one of claims 12 to 18, wherein, Each of the one or more devices includes a respective Open Platform Communications Unified Architecture (OPC UA) server communicatively connected to an aggregated address space, wherein an aggregated ontology of the graph-based industrial data is derived based on an OPC UA information model provided by the aggregated address space.

20. The computing system according to claim 19, wherein The aggregated ontology of the graph-based industrial data includes a Resource Description Framework (RDF) graph obtained by converting the OPC UA information model provided by the aggregated address space into a target ontology.

Images