Method and system for providing recommendations regarding a configuration process

Abstract

The invention relates to a computer-implemented method for providing a recommendation REC on a configuration process of configuring an industrial system SYS, the method comprising the steps of: computing (S1), by a trained graph neural network GNN, scores s of components c of a set C of configurable component types ct; generating (S2), based on the computed scores s, a recommendation REC for introducing at least one additional component c into the industrial system SYS; and outputting (S3), to a user, the generated recommendation REC, or executing the generated recommendation, by means of a user interface.

Description

Technical Field

[0001] This invention relates to a computer implementation method and recommendation engine for providing recommendations on the configuration process of configuring an industrial system. Background Technology

[0002] Industrial systems can include many different types of components, especially automation components. These automation components can include both hardware and software components.

[0003] Configuring an industrial system in an engineering project involves several key steps. Users must select appropriate components whose interactions satisfy all functional requirements arising from the intended use case. For this purpose, engineers responsible for configuring industrial systems typically utilize configuration software containing a catalog of available automation components. A fully integrated automation portal can provide access to a wide range of digital automation services, ranging from digital planning and integration engineering to transparent operation. Each automation component may include a set of technical characteristics or attributes that influence its capabilities and its compatibility with other automation components. Technical characteristics can include both static technical characteristics and configurable technical attributes. For engineering industrial systems, the selected components must be coupled such that the connectivity pattern allows for the anticipated real-world application. Therefore, it is not possible to represent an industrial system engineering scheme in the form of a flat list, but rather to consider the inherent topology. Finally, during the routine configuration process, the values ​​of the configurable technical attributes of the selected automation components are chosen by the user, ensuring component compatibility and that the entire subsystem can operate in the intended scenario.

[0004] Conventional systems can implement predefined recommendation rules, such as "if A, then B". Furthermore, implicit recommendations for the next component to be added to the industrial system can be provided by sorting the corresponding lists of possible options (i.e., all available components and / or all available components belonging to a certain category). This can be done either based on some manually defined human criteria or by employing one of the collaborative filtering techniques. Collaborative filtering techniques can be enhanced with additional contextual information, or information about the sequence in which items or components are added to an already partially configured industrial system. However, none of the conventional recommendation systems provide information about how the configured components are connected to each other. Summary of the Invention

[0005] Therefore, the object of the present invention is to provide a method and apparatus for increasing the efficiency of the configuration process for configuring industrial systems.

[0006] This objective is achieved through a computer implementation method according to the first aspect of the present invention.

[0007] According to a first aspect, the present invention provides a computer implementation method for providing a recommended configuration process for configuring an industrial system, wherein the method includes the following steps:

[0008] Scores for components within a set of configurable component types are calculated using a trained graph neural network.

[0009] Recommendations for introducing at least one additional component into an industrial system are generated based on the calculated scores.

[0010] The generated recommendations are output to the user through the user interface, or the generated recommendations are executed.

[0011] Furthermore, the graph neural network is trained to encode the component characteristics and topology of the industrial system configured by the configuration process. The component characteristics may include both static characteristics and configurable attributes of the respective components.

[0012] In a further possible embodiment of the computer implementation method according to the first aspect of the invention, the topology of the industrial system is represented by a topology map stored in memory.

[0013] The vertex set of the topology graph includes vertices representing configured components of the industrial system, and

[0014] The set of edges in the topology graph includes edges between two vertices, where each edge represents a connection between two corresponding components within the industrial system configured by the configuration process.

[0015] In a further possible embodiment of the computer implementation method according to the first aspect of the invention, each vertex in the vertex set of the topology graph of the industrial system representing corresponding components in the industrial system includes an associated feature vector specifying the technical attributes of the corresponding component.

[0016] In a further possible embodiment of the computer implementation method according to the first aspect of the invention, each feature vector of a vertex in the vertex set of the stored topology graph of the industrial system includes a one-hot encoding of the component type of the corresponding component.

[0017] In a further possible embodiment of the computer implementation method according to the first aspect of the invention, the feature vectors of all vertices in the vertex set of the stored topology graph of the industrial system form a feature matrix stored in memory.

[0018] In a further possible embodiment of the computer implementation method according to the first aspect of the invention, embedding for each configured component of the industrial system is performed by the trained graph neural network processing the feature vectors of corresponding vertices in the vertex set of the stored topology graph of the industrial system, together with the feature vectors of all adjacent vertices in the stored topology graph, to generate a context-aware embedding with the embedding size of the corresponding component.

[0019] In a further possible embodiment of the computer implementation method according to the first aspect of the invention, embedding is performed for all configured components of the industrial system to generate a first embedding matrix. ,

[0020] Where n is the number of vertices v in the vertex set V of the stored topological graph G, and d is the embedding size.

[0021] In a further possible embodiment of the computer implementation method according to the first aspect of the invention, embedding is performed for all component types of the component to generate a second embedding matrix Z, wherein ,

[0022] Where m is the number of configurable component types, and d is the embedding size.

[0023] In a further possible embodiment of the computer implementation method according to the first aspect of the invention, the second embedding matrix Z and the transposed first embedding matrix H are executed. T Matrix multiplication is used to compute the score matrix S, where ,

[0024] Each entry s of the calculated score matrix S ij Includes a score s, which indicates the selected component type ct. i Component c and connect it to component type ct j The plausibility of the already configured component c.

[0025] In a further possible embodiment of the computer implementation method according to the first aspect of the invention, the recommendation for introducing at least one additional component c into the industrial system is generated for the component type that has the highest score s in the calculated score matrix S.

[0026] In a further possible embodiment of the computer implementation method according to the first aspect of the invention, in response to a user introducing an additional component c into an industrial system via a user interface, the stored topology of the industrial system is automatically expanded using additional vertices corresponding to the introduced component c, and expanded using edges between the additional vertex v and the vertices of at least one component c to which the additional component c is already connected.

[0027] In a further possible embodiment of the computer implementation method according to the first aspect of the invention, the feature vector x corresponding to the vertex v of the additional component c v The one-hot encoding of component type ct of the additional component c is initialized, and the feature vector x v The remaining entries are set to zero.

[0028] In a further possible embodiment of the computer implementation method according to the first aspect of the invention, the initialized feature vector x corresponding to the vertex v of the additional component c v The vector is passed through the trained graph neural network (GNN) to generate an embedding vector h for the corresponding added component c. n+1 The embedding vector h n+1 It is fed into the prediction model g to generate prediction vectors. The prediction vector It contains entries that are output to the user via the user interface, serving as recommendations for the technical attributes of the corresponding added component c.

[0029] According to another aspect, the present invention further provides a recommendation engine.

[0030] According to a second aspect, the present invention provides a recommendation engine for providing recommendations regarding the configuration process of an industrial system.

[0031] The recommendation engine is adapted to compute a score s of component c in a set C of configurable component types ct using a trained graph neural network (GNN), and to generate recommendations for introducing at least one additional component c into the industrial system based on the computed score s.

[0032] The generated recommendations are output to users or executed automatically via a user interface, and the graph neural network (GNN) is trained to encode the component features and topology of the industrial system configured by the configuration process, wherein the component features include the static features and configurable attributes a of the corresponding component c. Attached Figure Description

[0033] In the following description, different aspects of possible embodiments of the invention are described in more detail with reference to the accompanying drawings.

[0034] Figure 1 A flowchart illustrating a possible exemplary embodiment of a computer-implemented method according to a first aspect of the present invention is shown;

[0035] Figure 2 A diagram illustrating a possible exemplary embodiment of a recommendation engine according to another aspect of the present invention is shown;

[0036] Figure 3 A further flowchart illustrating a computer implementation method according to a first aspect of the invention is shown. Detailed Implementation

[0037] As from Figure 1 As can be seen from the flowchart, a recommended computer implementation method for providing a configuration process for configuring an industrial system SYS may include several main steps in possible embodiments.

[0038] The computer implementation method according to the present invention can be used to provide assistance in the process of configuring engineering projects concerning an industrial system SYS. This computer implementation method can use a trained graph neural network (GNN) to provide recommendations regarding the configuration process of the industrial system. The graph neural network (GNN) is a connectionist model that captures graph dependencies via message passing between nodes in the graph. Unlike standard neural networks, graph neural networks can preserve state, which can represent information from its neighborhood at arbitrary depths.

[0039] In such Figure 1 In the computer implementation method according to the first aspect of the present invention illustrated in the flowchart, in the first step S1, the score s of component c of the set of configurable component types C is calculated by a trained graph neural network (GNN).

[0040] In a further step S2, recommendations for introducing at least one additional component c into the industrial system SYS are generated based on the calculated score s. Finally, in a third step S3, the generated recommendations are either output to the user via a user interface or executed automatically.

[0041] The graph neural network (GNN) used to calculate the score s in step S1 has been trained to encode the component characteristics and topology of the industrial system SYS configured during the configuration process. Component characteristics may include static characteristics and configurable technical attributes of the corresponding component c. The static characteristics or attributes of the automation component c are immutable and do not change over time. These static characteristics may include, for example, the size, height, or volume of the corresponding automation component c. Another example of static characteristics may include, for example, the number of ports provided by the corresponding automation component. A further example of the static characteristics of the automation component c is the applied line voltage or supply voltage for the corresponding automation component c. In addition to static characteristics, component c may also include configurable technical attributes, such as the temperature range in which the automation component c can be used in the industrial system SYS, or whether the corresponding automation component c must be fail-safe.

[0042] The topology of the corresponding industrial system SYS to be configured during the configuration process can be represented by a topology graph G = (V, E) and can be stored in the recommendation engine's memory. The vertex set V of the topology graph G contains vertices v, where each vertex v represents a configured component c of the industrial system SYS. Furthermore, the edge set E of the topology graph G contains two edges v. i v j The edge between them, whereby the edge represents two corresponding components c within the industrial system SYS to be expanded during the configuration process. i c j The connection between them.

[0043] Each vertex v within the vertex set V of the topology graph G of the industrial system SYS—representing the corresponding component c in the industrial system SYS—includes an associated eigenvector x. v The eigenvector x of vertex v v Specify the technical attributes of the corresponding automation component c. These technical attributes can include both static characteristics and configurable technical attributes of the corresponding automation component c. The eigenvector x of each vertex v within the vertex set V of the stored topology graph G of the industrial system SYS. v This can include one-hot encoding of the component type ct of the corresponding automation component c. In a possible embodiment, the feature vector x of all vertices v in the vertex set V of the stored topology graph G of the industrial system SYS is... v The feature matrix X is formed and stored in the memory of the recommendation engine.

[0044] The embedding of each configured component c for the industrial system SYS is performed as follows: the feature vector x of the corresponding vertex v in the vertex set V of the stored topological graph G of the industrial system SYS is processed by a trained graph neural network GNN. vTogether with the eigenvectors x of all adjacent vertices v in the stored topological graph G v To generate a context-aware embedding h with an embedding size d corresponding to component c. v Adjacent vertices v can include directly adjacent vertices v in the graph, and can also include vertices v indirectly connected via several hops in the topological graph G. The graph neural network GNN is trained to encode both the component features and the topology of the corresponding project. Partial, i.e., incomplete projects are represented by the topological graph G = (V, E). Each component c has an associated feature vector x. v The feature vector x v The configured technical attributes are specified.

[0045] Embedding is performed on all configured components c in the industrial system SYS to generate a first embedding matrix. , where n is the number of vertices v in the vertex set V of the stored topological graph G, and d is the embedding size.

[0046] Embedding is performed for all component types ct of component c to generate a second embedding matrix Z, where , where m is the number of configurable component types ct, and d is the embedding size. To allow for efficient computation at the node level, an embedding is generated for each configured component c represented by the corresponding vertex v in the vertex set V. For this purpose, a trained graph neural network (GNN) is employed. The trained GNN takes the feature vector x of a given center node or vertex v in the graph as its input. v Together with all eigenvectors x of its neighboring vertex v v Take as input to generate context-aware embedding h v Heuristically, the forward pass of a graph neural network (GNN) first aggregates the feature vectors x of all automated components c connected to the vertex v of the central node. v Then, in the second step, the graph neural network (GNN) combines this neighborhood information with the feature vector x. v Combining to produce an embedded h v It is possible to stack multiple layers of a graph neural network (GNN) to obtain a more expressive encoder. The computation can be repeated or iterated for each node or vertex v to form a first embedding matrix H.

[0047] Since the goal is to generate scores s for all component types c in a set of component types C, d-dimensional embeddings for all component types ct are generated. This can be achieved by reusing components of a graph neural network (GNN) or via embedding lookup. In most cases, the resulting embedding matrix is... Let m be the number of configurable component types ct, and d be the embedding size.

[0048] In a further step, the second embedding matrix Z is subjected to the transpose of the first embedding matrix H. T Matrix multiplication is used to compute the score matrix S, where Each entry s of the calculated score matrix S ij Includes a score s, which indicates the selected component type ct. i Component c and connect it to component type ct j The suitability or appropriateness of the already configured component c. Therefore, when performing matrix multiplication corresponding to the linear decoding step. S = Z • H T At that time, the score s is generated for each item. Therefore, if it is calibrated correctly, then Entry S ij Includes a score s, which indicates the selected component type ct. i And connect it to the component type ct j The rationality of the already configured component c. Based on this, it is possible to generate the following recommendation for the user: add component c with the highest score s to the partially configured industrial system SYS.

[0049] Users can now proceed in different ways. They can add the recommended automation component c to the existing partially configured industrial system SYS, or they can add another unrecommended component c to the industrial system SYS. As a further option, users can decide not to add any further components c and finalize the project.

[0050] If the user adds the recommended component c or another unrecommended component, this corresponds to adding a new vertex v. n+a Add it to the topology graph G and connect it to an existing node or vertex v. j In this case, graph G is expanded with new vertices and edges, and the process is iterated. Otherwise, the configuration process can terminate if the user does not add any further components.

[0051] In response to a user introducing an additional component c into the industrial system via a user interface, the stored topology graph G of the industrial system SYS can be automatically expanded using the additional vertex v corresponding to the introduced component c, and the edges between this additional vertex v and the vertices of component c that are already connected to the additional component c are also expanded. The eigenvector x corresponding to vertex v of the additional component c...v The one-hot encoding of component type ct of additional component c can be initialized, and the feature vector x v The remaining entries are set to zero. In a possible embodiment, the initialized eigenvector x corresponds to the vertex v of the addendum c. v It can be passed through a trained graph neural network (GNN) to generate embeddings h for the corresponding added component c. n+1 The embedded h n+1 It can be fed into the prediction model g to generate prediction vectors. The prediction vector It contains entries that are output to the user via the user interface, serving as recommendations for the technical attributes of the corresponding added component c.

[0052] Figure 2 A diagram illustrating a possible embodiment of a recommendation engine 1 according to another aspect of the invention is shown. Figure 2 The recommendation engine 1 illustrated in the diagram comprises three main modules in the illustrated embodiment. Recommendation engine 1 receives partially configured project items via an input interface, represented by a topology graph G having vertices v connected by edges. Each vertex v or node in graph G represents a configured component c of the corresponding industrial system SYS. These automation components c can include both hardware and software components. These automation components c can, for example, include programmable logic controllers (PLCs), human-machine interfaces (HMIs), motion controllers, server amplifiers, variable speed drives, or robot components. Depending on the use case, a wide variety of different hardware or software components can be configured and used during the configuration process of the industrial system SYS. Figure 2 In the illustrated example, the graph G of the partially configured engineering process includes nine vertices v1 to v9, each vertex representing the corresponding component c of a specific component type ct. For example... Figure 2 As illustrated, vertices v1, v2, v3, and v7 represent component c of the first component type ct. Furthermore, vertices v4 and v9 belong to the same component type ct. Additionally, components c represented by vertices v5 and v8 belong to the same component type ct. Vertex v6 ​​represents component c of another component type ct. Vertices v are connected via edges in the edge set E, which represent connections between two corresponding components within the industrial system SYS.

[0053] Recommendation Engine 1 can be used to provide recommendations on the configuration process for configuring and extending the corresponding industrial system (SYS). For example... Figure 2The recommendation engine 1 shown is adapted to: compute scores s for components c of a set C of configurable component types ct using a trained graph neural network (GNN), and automatically generate recommendations based on the computed scores s to introduce at least one additional component c into the industrial system SYS. The generated recommendations can be output to a user via a user interface or automatically executed by the system's execution engine.

[0054] In the illustrated example, recommendation engine 1 recommends vertex v of graph G'. 10 The additional component c is added to the partially configured industrial system SYS, as shown in Figure G, which is supplied to the recommendation engine 1. 10 Component c is represented as follows: This component c is connected to component c represented by vertices v6, v8, and v9. The recommendation engine 1 according to the present invention not only provides recommendations on which type of component c must be added to the partially configured industrial system SYS, but also provides recommendations on which other components c it should be connected to. Furthermore, the recommendation engine 1 also provides information on the configurable properties of the added automation components c. Figure 2 In the example illustrated, the configurable properties of the added component c include a temperature value of 60 as the maximum temperature of the added component c. Furthermore, it indicates that the added component c should be fail-safe.

[0055] exist Figure 2 In the illustrated embodiment, the recommendation engine 1 includes three main modules: an encoder module 2, a link prediction module 3, and an attribute prediction module 4.

[0056] Encoder module 2 includes a trained graph neural network (GNN) that can compute a score s for component c of a set C of configurable component types ct. Based on the computed score s, recommendations for introducing at least one additional component into the industrial system are generated. Link prediction module 3 is used to predict links or edges indicating which components c should be connected to the added component c. Attribute prediction module 4 can use a prediction model g to generate prediction vectors. The prediction vector It has entries that are output to the user via the user interface, as recommendations for the technical attributes of the corresponding added component c, such as fail-safe = true and maximum operating temperature = 60.

[0057] Figure 3 Further diagrams illustrating a computer implementation method according to the present invention are shown. (e.g.) Figure 3 As illustrated in the diagram, the partially configured engineering project, represented by graph G, is supplied to a trained graph neural network GNN that performs the encoding ENC. The linear decoder DEC performs the second embedding matrix Z with the transposed first embedding matrix H. TMatrix multiplication is used to compute the score matrix S, where each entry S of the computed score matrix S is... ij Includes a score s, which indicates the selected component type ct. i Component c and connect it to component type ct j The rationality of the already configured component c. Then, in the illustrated example, the user can perform component selection SFL to extend an existing partial project, where the extended project or industrial system SYS can be illustrated by Figure G', as well as... Figure 2 As shown in the example, the extended engineering project or industrial system SYS, illustrated by the extended graph G', can then be processed again by a trained graph neural network GNN to provide the encoded ENCI of the system, including the new added component c. This can be fed to a predictive model g to predict the configurable technical properties a or features of the corresponding component c.

[0058] Graph Neural Networks (GNNs) can be trained from historical engineering projects. The training data T can consist of historical engineering projects configured in the past. This means... Each G i The network representation corresponding to the engineering project, and X i Corresponding to the feature matrix. Heuristically, first delete (G i X i The structure of the project is partially determined, and then these structures are recovered by the recommendation system according to the invention. Therefore, it is possible to obtain an approximation of a partially configured engineering project. The system SYS is then trained and validated by its ability to complete the project, as it is pre-configured before the deletion step. More specifically, it is possible to train the parameters of a graph neural network (GNN) to recommend vertices in the set V that result in a loss L. R The previously deleted items. Furthermore, it is possible that the recovery resulted in the predicted loss L. P The characteristic matrix X i The ability to train a graph neural network (GNN) and a prediction model g. In a possible implementation, a joint loss can be formed. L = L R + L P This allows the entire model to be trained end-to-end.

[0059] Using the computer implementation method according to the invention, it is possible to leverage historical examples of industrial engineering projects to provide not only details about where selected components c are connected, but also to predict the values ​​of the technical attributes of the corresponding components c. This computer implementation method and engine are more scalable than conventional systems that rely on manually defined rules. Furthermore, depending on the complexity of the underlying technical system, the computer implementation method according to the invention allows the discovery of patterns that are more complex or ambiguous than those easily specified by domain experts responsible for maintaining the rule set, serving as a basis for recommendations. Compared to existing data-driven methods, the computer implementation method according to the invention is able to provide details about how automation components c should be connected to each other and the values ​​of the configurable technical attributes of the corresponding automation components. This is possible using a trained graph neural network (GNN) and a prediction model g. The computer implementation method according to the invention can be integrated into configuration software tools to increase the efficiency of the configuration process for complex industrial systems SYS. This computer implementation method can be executed by a recommendation assistance system. In a possible embodiment, the recommendation engine 1 can operate in real time while the user is configuring an engineering project or industrial system SYS. Depending on the specificity of the use case, the quality and quantity of training data, the proposed computer implementation method can be used to either assist the user in configuring the industrial engineering system SYS or to automate the configuration process. The computer implementation method according to the present invention not only provides component selection and topology construction, i.e. how the automation component c is connected, but also provides selection of values ​​for configurable technical attributes of the added component c, such as whether the added component c should be fail-safe, or values ​​related to the maximum permissible operating temperature of the corresponding added component c.

Claims

1. A computer implementation method for providing a recommended REC (Registered Execution Code) for configuring an industrial system SYS. The method includes the following steps: - A trained graph neural network (GNN) is used to compute a score s for component c of a set of configurable component types ct, wherein the score s is obtained by generating a first embedding matrix H for the configured component c of the industrial system SYS and a second embedding matrix Z for the configurable component type ct, and performing an operation on the second embedding matrix Z and the transpose of the first embedding matrix H. T The matrix multiplication is used to calculate the score s, and the score s indicates the rationality of selecting component c of component type ct and connecting it to the configured component c. - Generate a recommended REC based on the calculated score s for introducing at least one additional component c into the industrial system SYS; - Utilize the user interface to output the generated recommendation REC to the user, or execute the generated recommendation, and The graph neural network (GNN) is trained to encode the component features and topology of the industrial system SYS configured by the configuration process, wherein the component features include the static features and configurable attributes a of the corresponding component c.

2. The computer implementation method according to claim 1, wherein the topology of the industrial system SYS is represented by a topology graph G=(V, E) stored in memory. The vertex set V of the topology graph G includes vertices v representing configured components c of the industrial system SYS, and The edge set E of the topological graph G contains two vertices v i v j The edge between them, whereby the edge represents two corresponding components c within the industrial system SYS configured by the configuration process. i c j The connection between them.

3. The computer implementation method according to claim 2, wherein each vertex v in the vertex set V of the topology graph G of the industrial system SYS, representing the corresponding component c in the industrial system SYS, includes an associated feature vector x specifying the technical attributes of the corresponding component c. v .

4. The computer implementation method according to claim 3, wherein each feature vector x of vertex v in the vertex set V of the stored topology graph G of the industrial system SYS v This includes the one-hot encoding of the component type ct of the corresponding component c.

5. The computer implementation method according to claim 3 or 4, wherein the feature vector x of all vertices v in the vertex set V of the stored topology graph G of the industrial system SYS is... v The feature matrix X is formed and stored in memory.

6. The computer implementation method according to claim 3 or 4, wherein the embedding of each configured component c of the industrial system SYS is performed by processing the feature vector x of the corresponding vertex v in the vertex set V of the stored topological graph G of the industrial system SYS by the trained graph neural network GNN. v Together with the eigenvectors x of all adjacent vertices v in the stored topological graph G v To generate a context-aware embedding h with an embedding size d corresponding to component c. v .

7. The computer implementation method of claim 6, wherein the embedding of h is performed for all configured components c of the industrial system SYS. v To generate the first embedding matrix , Where n is the number of vertices v in the vertex set V of the stored topological graph G, and d is the embedding size.

8. The computer implementation method of claim 7, wherein embedding is performed for all component types ct of component c to generate a second embedding matrix Z, wherein , Where m is the number of configurable component types ct, and d is the embedding size.

9. The computer implementation method according to claim 8, wherein the second embedding matrix Z and the transposed first embedding matrix H are executed. T Matrix multiplication is used to compute the score matrix S, where , Each entry s of the calculated score matrix S ij Includes a score s, which indicates the selected component type ct. i Component c and connect it to component type ct j The rationality of the already configured component c.

10. The computer implementation method of claim 9, wherein the recommendation for introducing at least one additional component c into the industrial system SYS is generated for the component type ct that has the highest score s in the calculated score matrix S.

11. The computer implementation method of claim 2, wherein in response to a user introducing an additional component c into an industrial system SYS via a user interface, the stored topology graph G of the industrial system SYS is automatically expanded using additional vertices v corresponding to the introduced component c, and expanded using edges between the additional vertices v and vertices of at least one component c to which the additional component c is already connected.

12. The computer implementation method according to claim 11, wherein the feature vector x corresponding to the vertex v of the additional component c v The one-hot encoding of component type ct of the additional component c is initialized, and the feature vector x v The remaining entries are set to zero.

13. The computer implementation method of claim 12, wherein the initialized feature vector x corresponding to the vertex v of the additional component c. v The embedded h is passed through the trained graph neural network (GNN) to generate an embedding h for the corresponding added component c. n+1 The embedded h n+1 It is fed into the prediction model g to generate prediction vectors. The prediction vector It contains entries that are output to the user via the user interface, serving as recommendations for the technical attributes of the corresponding added component c.

14. A recommendation engine device for providing a recommended REC (Registration Reference Code) for configuring an industrial system SYS. The recommendation engine device is adapted to: A trained graph neural network (GNN) computes a score s for a set of configurable component types ct for a component c, wherein the score s is obtained by generating a first embedding matrix H for the configured component c of the industrial system SYS and a second embedding matrix Z for the configurable component type ct, and performing an operation on the second embedding matrix Z and the transpose of the first embedding matrix H. T The matrix multiplication is used for calculation, and the score s indicates the rationality of selecting component c of component type ct and connecting it to the configured component c, and Based on the calculated score s, a recommended REC for introducing at least one additional component c into the industrial system SYS is automatically generated, wherein the generated recommended REC is output to a user via a user interface or is automatically executed by the recommendation engine device, and wherein the graph neural network GNN is trained to encode the component features and topology of the industrial system SYS configured by the configuration process, wherein the component features include the static features and configurable attribute a of the corresponding component c.

Images