Method and device for computer-aided design of technical systems

Abstract

According to the present invention, for corresponding components (K1, K2) of a technical system (TS), component names (KB1, KB2) and characteristic parameter names (KPB1, KPB2) of design-related characteristic parameters of the components (K1, K2) are read in and a search engine (SE) is controlled using these. Documents (D1, D2) found by the search engine (SE) are read in and component information (KA1, KA2), such as product information about the corresponding component, is extracted from them. The extracted component information (KA1, KA2) is fed to a machine learning routine (NN), which is trained based on a plurality of predefined training component information (TKA) and training characteristic parameter values ​​(TKPW) with respect to reproducing predefined training characteristic parameter values ​​(TKPW) based on the predefined training component information (TKA). Output data (OUT) of the machine learning routine (NN) are selected as characteristic parameter values ​​(KPW1, KPW2) and are embedded in a planning data record (PR1, PR2). Next, the planning data records (PR1, PR2) are output for the design of the technical system (TS).

Description

Background Art

[0001] Computer-aided design and planning tools are increasingly being used to design complex technical systems, such as manufacturing facilities, power grids, wind turbines, gas turbines, power plants, robots, industrial plants, or motor vehicles. For this purpose, the system components to be built are mapped to formalized planning components of a planning model. However, this can prove to be very costly, particularly in the case of complex systems, as the corresponding components are often provided by multiple manufacturers and specified in different, informal ways. Summary of the Invention

[0002] The object of the present invention is to specify a method, a device, a computer program product and a computer-readable storage medium, by means of which the complexity in designing a technical system can be reduced.

[0003] This object is achieved by a method according to the invention, by a device according to the invention, by a computer program product according to the invention, and by a computer-readable storage medium according to the invention.

[0004] According to the present invention, for computer-aided design of a technical system, the component name and the characteristic parameter names of the design-relevant characteristic parameters of the component are read in for the corresponding component of the technical system. A search engine is controlled using the read-in component name and / or characteristic parameter name as a search term, and the documents found by the search engine are read in. In this case, the corresponding characteristic parameter name can, in particular, indicate which characteristic parameter of the corresponding component is to be specified for the specific design of the technical system. Component information, such as product information about the corresponding component, is extracted from the found documents and fed to a machine learning routine that is trained based on a plurality of predefined training component information and training characteristic parameter values ​​to reproduce predefined training characteristic parameter values ​​based on the predefined training component information. A planning data record is also generated for the technical system, wherein output data of the machine learning routine are selected as characteristic parameter values ​​and embedded in the planning data record. The planning data record is then output for use in designing the technical system.

[0005] To carry out the method according to the invention, a device for computer-aided design of the technical system, in particular an assistance system, a computer program product and a computer-readable storage medium are provided.

[0006] The method according to the invention and the device according to the invention can be implemented or realized, for example, by means of one or more processors, application-specific integrated circuits (ASICs), digital signal processors (DSPs) and / or so-called "field programmable gate arrays" (FPGAs).

[0007] The advantage of the present invention is particularly evident in the fact that planning data records can be parameterized as automatically as possible. These parameterized planning data records can then often be incorporated directly into the data model of the automated planning method. By including a search engine, the most comprehensive information possible about available components and their characteristic parameters can also be used for the design. This generally increases the scope for optimization and, therefore, often leads to better planning results.

[0008] Advantageous embodiments and developments of the invention are explained in the following description.

[0009] Advantageously, the machine learning routine can be implemented with the aid of an artificial neural network, a recursive neural network, a convolutional neural network, an autoencoder, a deep learning architecture, a support vector machine, a data-driven trainable regression model, a k-nearest neighbor classifier, a physical model, and / or a decision tree. To train the machine learning routine implemented in this manner, a variety of efficient standard methods can be used.

[0010] According to an advantageous embodiment of the present invention, target values ​​for predefined design parameters for the technical system can be read in, and characteristic parameter values ​​can be selected based on the read-in target values. In this case, the design parameters can relate to and / or quantify predefined requirements, functions, boundary conditions, or other characteristics of the technical system or component. In particular, the design parameters can relate to operational, functional, or production-related characteristics, such as delivery times, maintenance intervals, and the like. A desired or required value or value range for the design parameter can be read in as the target value. In this selection, the characteristic parameter value that is closest to or approximately close to the target value can be selected. Alternatively or additionally, a characteristic parameter value that does not exceed or fall below the target value can be selected.

[0011] The generated planning data record can also be used to configure a simulator for the technical system. The configured simulator can then be used to simulate the technical functions of the technical system and output a functional description of the simulated technical functions. The simulator can, for example, implement event-based and / or time-discrete simulations, in particular logistics simulations. In particular, operational functions of the technical system or parts thereof, such as power, speed, material transport, output, or other capabilities or characteristics, can be simulated. In particular, the characteristic parameter values ​​to be embedded in the planning data record can be selected so that deviations from the functional description from predefined functional requirements for the technical system are minimized. In this way, the functionality of the technical system can be adapted to the functional requirements and / or optimized.

[0012] According to another advantageous embodiment of the present invention, multiple planning data records can be generated, each of which has different characteristic parameter values ​​selected from the output data of the machine learning routine. For each of the generated planning data records, the technical functions of the technical system can be simulated using a simulator. A planning data record that optimizes the simulated technical functions can then be selected from the multiple generated planning data records for use in designing the technical system.

[0013] According to an advantageous embodiment of the present invention, component type information can be read in for the corresponding component. The extracted component information can then be fed together with the corresponding component type information to a machine learning routine trained specifically for the component type. For example, "production machine" or "transportation system" can be specified as component type information. Component type-specific training has proven to be very efficient in many cases, as the association of feature parameter names with feature parameter values ​​is often presented in a type-specific manner.

[0014] It can also be provided that the machine learning routine is trained based on a plurality of predefined training component data, training feature parameter names, and training feature parameter values ​​with respect to reproducing predefined training feature parameter values ​​based on the predefined training component data and training feature parameter names. Feature parameter names associated with the extracted component data can then be supplied to the machine learning routine. Explicitly taking into account the training feature parameter names when training the machine learning routine and explicitly taking into account the feature parameter names when using the trained machine learning routine can significantly improve the determination of feature parameter values ​​in many cases.

[0015] Advantageously, search engines can use web crawlers that specialize in product data to search data networks, in particular the Internet or an intranet. Specialized web crawlers can classify visited pages by topic and limit their index to pages relevant to the topic. In this way, the proportion of relevant documents in the documents found can be increased in many cases.

[0016] According to a further advantageous embodiment of the invention, the component information can be extracted from the found documents by means of a parser, for example an HTML or XML parser and / or a pattern recognizer.

[0017] The corresponding component names and / or characteristic parameter names can also be read in from a construction plan of the technical system. In particular, an electronic construction plan for the technical system, such as a CAD plan drawing, can be used as the construction plan.

[0018] It is also possible to extract a plurality of training characteristic parameter values ​​from a plurality of parameterized planning data records. A parameterized planning data record is to be understood in particular as a planning data record with characteristic parameter values ​​used, preferably in a uniform format. Suitable parameterized planning data records are often provided by previously completed designs of the technical system or can be derived therefrom. Preferably, parameterized planning data records from similar technical systems can be used. However, in many cases, parameterized planning data records from different technical systems can also be used advantageously, since cross-system correlations often exist between component names, characteristic parameter names, and characteristic parameter information, which can be learned by the machine learning routine. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0020] Figure 1 shows an assistance system according to the invention in its training phase for designing a technical system; and

[0021] Figure 2 The trained assistance system is shown in the application phase. DETAILED DESCRIPTION

[0022] Figure 1A schematic diagram shows an assistance system AS for designing technical systems in the training phase. The assistance system AS can be used to design a variety of technical systems, such as manufacturing facilities, power grids, wind turbines, gas turbines, power plants, robots, industrial plants, motor vehicles, or combinations thereof. The technical systems to be designed each include a plurality of different technically interacting components, such as electrical or mechanical components, manufacturing robots, machine tools, transport systems, conveyor belts, and / or other machines or machine parts.

[0023] For this design, in particular, it should be determined: which components with which characteristics are required and available for the construction of the technical system. Here, these components may come from multiple suppliers or manufacturers.

[0024] To specify this design, specific values ​​for characteristic parameters of the component should be specified or adapted for the respective component. Such component-specific characteristic parameters quantify properties of the respective component that are relevant for the design of the technical system, such as the size of the component, the resource consumption of the component, the performance of the component, and / or the functionality of the component.

[0025] Subsequently, for example, a non-parameterized planning model can be parameterized as a function of the determined characteristic parameter values, and the parameterized planning model can be used for the production or quantitative simulation of the technical system.

[0026] Such a non-parameterized planning model can be, for example, a construction plan or another planning model in which the components and their design-related characteristic parameters are represented but not yet quantified, for example because the specific characteristic parameter values ​​of commercially available components are not yet known.

[0027] The assistance system AS has one or more processors PROC for carrying out the method steps of the assistance system AS and one or more memories MEM for storing data to be processed by the assistance system AS.

[0028] According to the present invention, the assistance system AS includes a trainable artificial neural network NN, by which a machine learning routine is implemented. The neural network NN is to be trained to extract characteristic parameter values ​​of a plurality of components of a technical system from textual documentation about these components in a uniform and formalized manner or to extract the characteristic parameter values ​​of these components into a uniform, machine-readable format.

[0029] To train the neural network NN, the auxiliary system AS reads in a plurality of predefined, parameterized planning data records PR of components whose characteristic parameter values ​​are known. The planning data records PR can be read in, for example, from one or more component catalogs, component libraries, or other databases. Preferably, in this case, planning data records PR are read in that can also be used for components of the technical system to be designed or similar technical systems.

[0030] These planning data records PR also include, in particular, a formalized parameterization of the relevant components; that is to say, the characteristic parameter values ​​of these components are uniformly and formally present in these planning data records PR.

[0031] A plurality of training component names TKB are extracted from these planning data records PR. The training component names TKB respectively represent components and may include, for example, the names "robot" or "conveyor belt". The corresponding training component names TKB also include training component type information TKT, for example in the form of names "production machine" or "transportation system". In addition, training feature parameter names TKPB are extracted for the corresponding components from these planning data records PR, which represent the feature parameters of the corresponding components to be specified for the specific design. For example, the names "length", "width", "weight", "positioning accuracy", "conveyance speed", "pipeline cross section", "pipeline friction value", "tank capacity" or "maintenance interval" can be used as training feature parameter names TKPB.

[0032] Furthermore, a plurality of training component data TKA is extracted from these planning data records PR. These training component data include textual data about the respective components, for example, product information from the manufacturer's website or from a component catalog or component library. These textual data contain, in particular, quantified values ​​of characteristic parameters of the respective components, but usually in a non-formatted text form.

[0033] As mentioned above, the neural network NN is to be trained with respect to extracting these characteristic parameter values ​​contained in informal text form in a standardized machine-readable format. For this training purpose, the planning data records PR also contain corresponding, uniformly formalized training characteristic parameter values ​​TKPW, as already indicated above. The training characteristic parameter values ​​TKPW are each assigned to a predetermined training component name TKB, a training characteristic parameter name TKPB, and training component information TKA. The corresponding training characteristic parameter value TKPW specifies the value or value range of the characteristic parameter represented by the assigned training characteristic parameter name TKPB in a standardized machine-readable format. For example, the formalized information "10 m" can be assigned as a characteristic parameter value to the component name "Conveyor Belt" along with the characteristic parameter name "Length" and the component information "Conveyor Belt Construction Plan: [...] Length: 10 m." Such component information can be extracted from the planning data records PR, for example, using so-called regular expressions.

[0034] In order to train the neural network NN, the training component names TKB, the training feature parameter names TKPB and the training component data TKA which correspond to one another are transmitted to the neural network as input data.

[0035] In this context, training is generally understood to mean the optimization of the mapping of input data to output data of a machine learning routine, implemented here with the aid of a neural network NN. This mapping is optimized during the training phase according to predefined, learned, and / or to-be-learned criteria. In this context, the deviation between desired and actual output data can be used as a criterion, in particular. Through this training, for example, the connections between neurons of the neural network NN or the weights of the network structure of these neurons can be optimized so that predefined criteria are met as closely as possible.

[0036] In the present exemplary embodiment, the goal is to achieve, through this training, that the output data OUT resulting from the input data by the neural network NN reproduce the corresponding formalized training characteristic parameter values ​​TKPW as accurately as possible.

[0037] For this purpose, the output data OUT are compared with the training characteristic parameter values ​​TKPW, which are transmitted to the auxiliary system AS and assigned to the input data TKB, TKPB, and TKA, respectively. Within the framework of this comparison, a deviation D between the output data OUT and the training characteristic parameter values ​​TKPW is determined, for example, the absolute value or the square of the difference. The deviation D is then fed back to the neural network NN. Based on the fed-back deviation D, the neural network NN—as indicated by the dashed arrow—is trained to minimize the deviation D, that is, to reproduce the training characteristic parameter values ​​TKPW as closely as possible using the output data OUT derived from the input data. In other words, the neural network NN is trained to extract uniformly formalized characteristic parameter values ​​from informal component documentation using the output data OUT. In this case, the deviation D represents the extraction error of the neural network NN.

[0038] In the present embodiment, the training is performed component-type-specifically, that is, specifically for the type of the corresponding component, by taking into account the training component type information TKT. Alternatively or additionally, the training can also be performed group-specifically for frequently constructed component groups or component combinations.

[0039] To train the neural network NN, various standard methods of machine learning, in particular supervised learning, can be used.

[0040] Figure 2 The diagram shows the trained assistance system AS in the application phase, ie in the design of a concrete technical system TS. Figure 2 Use in Figure 1 The same reference numerals in the figures denote the same entities. Figure 1 As described, these entities can Figure 2 is implemented or realized in association with the functions of

[0041] In the present embodiment, the auxiliary system AS is intended to be used for designing a manufacturing facility as a technical system TS. Alternatively or additionally, the technical system to be constructed can also be a power grid, a wind turbine, a gas turbine, a power plant, a robot, an industrial facility or a motor vehicle or include such devices.

[0042] For the design of the manufacturing facility TS, a planning model PM that specifies the manufacturing facility TS in detail should be generated by a trained assistance system AS based on the construction plan BP of the manufacturing facility TS.

[0043] The manufacturing facility TS to be designed consists of a number of different technically interacting components. Figure 2For reasons of clarity, only two components K1 and K2 of a production facility TS are explicitly mentioned as examples. Components K1 and K2 can be, for example, production robots, machine tools, transport systems, conveyor belts, or other machines or machine parts. The properties of components K1 and K2 are each characterized by characteristic parameters of these components, as explained above.

[0044] The structure of the manufacturing facility TS, consisting of various components (here, K1 and K2) and their functional relationships, is described by a building plan BP. However, the building plan BP is not yet parameterized. That is, while it contains the component names KB1 and KB2 of components K1 and K2 and the characteristic parameter names KPB1 and KPB2 of their characteristic parameters, it does not yet or not fully include the specific values ​​of these parameters. As mentioned above, the component names KB1 and KB2 can include names such as "robot" or "conveyor belt." In the current embodiment, the component names KB1 and KB2 also include component type information, such as "production machine" or "transportation system."

[0045] As already indicated above, for the planning with the aid of the assistance system AS it is necessary to determine which components with which characteristic parameter values ​​are required and available for the construction of the production facility TS.

[0046] For this purpose, the component names KB1 and KB2 of the required components K1 and K2 and their characteristic parameter names KPB1 and KPB2 are first read in by the assistance system AS from the unparameterized building plan BP and stored in the memory MEM.

[0047] Next, the search engine SE is operated for the corresponding components K1 and K2 using the read-in names KB1 and KPB1 or KB2 and KPB2 as search terms. In this case, these search terms are passed to the search engine SE, which then returns references to documents in which these search terms preferably occur. For reasons of clarity, Figure 2 In the example, only two such references L1 and L2 to documents D1 and D2 are explicitly mentioned. An external or internal search engine for a data network WWW, such as the Internet or an intranet, can be used as the search engine SE. In the present embodiment, the World Wide Web (WW), as the data network www, is searched by the search engine SE. The documents D1 and D2 found are often hypertext or PDF documents from the web pages of the manufacturer or supplier of the corresponding component K1 or K2. These documents D1 and D2 typically contain informal, manufacturer-specific product information about the relevant component K1 or K2 in a manufacturer-specific format.

[0048] Preferably, the search engine SE for searching the data network WWW includes or uses a so-called web crawler, in particular a web crawler focused on product data. This web crawler classifies web pages and hyperlinks by topic, here regarding existing product data, and limits its index to web pages relevant to the topic.

[0049] The documents found by the search engine, here D1 and S2, are retrieved from the data network WWW via an auxiliary system AS and subjected to a feature extraction FE. The feature extraction FE includes an HTML and / or PDF parser that extracts text content from documents D1 and D2. The feature extraction FE also uses a downstream pattern recognizer specifically designed to recognize text patterns and extract the associated text content. Such a pattern recognizer can be implemented, for example, using so-called regular expressions. Such a feature extraction FE is often also referred to as a feature extraction system.

[0050] In the present exemplary embodiment, a feature extraction FE identifies text sections in documents D1 and D2 that contain feature parameter names KPB1 and KPB2. Component data KA1 or KA2, respectively, that are textually assigned to the feature parameter names KPB1 and KPB2, are then extracted from the identified text sections. These component-specific, textual component data KA1 and KA2 from the manufacturer typically contain quantized values ​​for the feature parameters of the corresponding components, but usually in informal textual form.

[0051] For the respective component K1 or K2, the names KB1 and KPB1 and the component documentation KA1 or the names KB2 and KPB2 and the component documentation KA2 are transmitted in association with one another as input data to the trained neural network NN. Output data OUT are derived from the input data by the trained neural network NN. The output data OUT are transmitted from the neural network NN to a selection module SEL, which selects component-specific and characteristic parameter-specific characteristic parameter values ​​from the output data OUT. For reasons of clarity, Figure 2 In the example, only two characteristic parameter values ​​KPW1 and KPW2 are explicitly mentioned. In this case, the characteristic parameter value KPW1 or KPW2 specifies the value or value range of the characteristic parameter designated by the characteristic parameter name KPB1 or KPB2.

[0052] In this way, with the aid of the trained neural network NN, uniformly formalized characteristic parameter values ​​KPW1 and KPW2 can be extracted in a planning-related manner from the informal, manufacturer-specific documents D1 and D2 .

[0053] Characteristic parameters KPW1 and KPW2 are each associated with the associated characteristic parameter name KPB1 or KPB2 and embedded in the planning data record PR1 or PR2 for the associated component K1 or K2. In this way, the corresponding planning data record PR1 or PR2 is parameterized, i.e., specific characteristic parameter values ​​in the corresponding planning data record PR1 or PR2, in this case KPW1 or KPW2, are used for characteristic parameters that have not yet been determined. Thus, the planning data record PR1 or PR2 essentially constitutes a parameterized entity of the planning component.

[0054] In the manner described above, a plurality of parameterized planning data records are generated for a plurality of documents found by the search engine SE and a plurality of characteristic parameter values ​​specified therein.

[0055] The assistance system AS includes a simulator SIM for deriving an optimized planning model PM that specifies the manufacturing facility TS in detail from the generated planning data records. The simulator SIM simulates the resulting dynamic behavior of the manufacturing facility TS, the behavior of the corresponding components, and / or the technical interaction of the components, based on one or more parameterized planning data records. The simulation is preferably performed for different, particularly typical and / or predetermined operating situations. For this purpose, the simulator SIM includes a configurable system model of the components and their technical interaction.

[0056] Using the simulator SIM, one or more technical functions of a manufacturing facility TS, such as material transport and / or workpiece processing, are simulated for a plurality of parameterized planning data records. Subsequently, a component-specific planning data record OPR that optimizes the technical function is selected from the plurality of parameterized planning data records and embedded in the planning model PM. This optimization selection is preferably performed for each of the components and / or specifically for each component type.

[0057] The planning data records and / or planning models optimized in this way can preferably be stored in a database or planning library across planning projects. Such a database or planning library can advantageously be used for other planning projects.

[0058] As an alternative to or in addition to the selection of the optimized planning data record OPR described above, the selection of characteristic parameter values ​​from the output data OUT of the neural network NN can also be controlled by means of the simulator SIM. In this case, the selection can be carried out, in particular, based on the technical functions of the production facility TS that are simulated specifically with respect to characteristic parameter values ​​and their comparison with predefined functional requirements. In this case, for example, the transport speed of the conveyor assembly required for the production facility TS, the positioning accuracy of the robot assembly, the resource consumption, and / or other target values ​​for the design parameters can be predefined as functional requirements.

[0059] The optimized planning model PM generated in this way includes a plurality of optimized parameterized planning data records OPR, which can specify the manufacturing facility TS in detail and be used for the production or quantitative simulation of the optimized manufacturing facility TS as follows. Finally, the optimized planning model PM is output for the design of the manufacturing facility TS by the assistance system AS.

[0060] The present invention allows for the largely automatic generation of entities for planning components, which can often be incorporated directly into the model of a predefined planning method. By automatically incorporating an Internet search using a search engine SE, the most comprehensive possible information about available components and their characteristic parameters can be considered during planning. In particular, by automatically considering multiple alternative solutions in the planning method, a larger solution space can often be evaluated, for example within the framework of heuristic planning and / or optimization schemes. This generally increases the optimization scope and, therefore, often leads to better planning results.

Claims

1. A method for computer-aided design of a technical system (TS), wherein For the corresponding components (K1, K2) of the technical system (TS), - reading in component names (KB1, KB2) of the components (K1, K2) and characteristic parameter names (KPB1, KPB2) of design-related characteristic parameters of the components (K1, K2), - operating a search engine (SE) using the read component names (KB1, KB2) and / or characteristic parameter names (KPB1, KPB2) as search terms, - reading in documents (D1, D2) found by the search engine (SE) and extracting component information (KA1, KA2) therefrom; supplying the extracted component data (KA1, KA2) to a machine learning routine, which is trained on the basis of a plurality of predetermined training component data (TKA) and training characteristic parameter values ​​(TKPW) with respect to reproducing predetermined training characteristic parameter values ​​(TKPW) on the basis of the predetermined training component data (TKA); Execute a trained machine learning routine, where The executing uses input data and generates output data, wherein the input data includes the extracted component data (KA1, KA2); generating a planning data record (PR1, PR2) for the technical system (TS), wherein output data (OUT) of the machine learning routine are selected as characteristic parameter values ​​(KPW1, KPW2) specific to the components (K1, K2) and specific to the design-related characteristic parameters and are embedded in the planning data record (PR1, PR2); outputting the planning data records (PR1, PR2) for designing the technical system (TS); wherein, in the generated plurality of planning data records (PR1, PR2), the plurality of planning data records each have characteristic parameter values ​​(KPW1, KPW2) selected differently from the output data (OUT) of the machine learning routine; For the respectively generated planning data records (PR1, PR2), the technical functions of the technical system (TS) are simulated by means of a simulator (SIM); and A planning data record (OPR) which optimizes the simulated technical functions is selected from a plurality of generated planning data records (PR1, PR2) for the design of the technical system (TS).

2. The method according to claim 1, characterized in that The machine learning routine is implemented with the aid of an artificial neural network (NN) and / or a recursive neural network and / or a convolutional neural network and / or an autoencoder and / or a deep learning architecture and / or a support vector machine and / or a data-driven trainable regression model and / or a k-nearest neighbor classifier and / or a physical model and / or a decision tree.

3. The method according to claim 1 or 2, characterized in that Reading in target values ​​of design parameters predetermined for the technical system; The characteristic parameter values ​​(KPW1, KPW2) are selected according to the read-in target values.

4. The method according to claim 1 or 2, characterized in that configuring a simulator (SIM) of the technical system (TS) with the aid of the generated planning data records (PR1, PR2); simulating the technical functions of the technical system (TS) by means of a configured simulator (SIM); and Outputs the functional description of the simulated technological functions.

5. The method according to claim 4, characterized in that Characteristic parameter values ​​(KPW1, KPW2) to be inserted into the planning data records (PR1, PR2) are selected such that deviations of the functional description from functional requirements specified for the technical system are reduced.

6. The method according to claim 1 or 2, characterized in that For the corresponding components (K1, K2), read in the component type data; The machine learning routine is trained specifically for the component type; and The extracted component information (KA1, KA2) is fed to the machine learning routine together with the corresponding component type information.

7. The method according to claim 1 or 2, characterized in that The machine learning routine is trained based on a plurality of predetermined training component data (TKA), training feature parameter names (TKPB), and training feature parameter values ​​(TKPW) with respect to reproducing predetermined training feature parameter values ​​(TKPW) based on the predetermined training component data (TKA) and training feature parameter names (TKPB); and The characteristic parameter names (KPB1, KPB2) associated with the extracted component information (KA1, KA2) are supplied to the machine learning routine.

8. The method according to claim 1 or 2, characterized in that The search engine (SE) searches the data network using a web crawler focused on product data.

9. The method according to claim 1 or 2, characterized in that The component information (KA1, KA2) is extracted from the found documents (D1, D2) by means of a parser and / or a pattern recognizer.

10. The method according to claim 1 or 2, characterized in that The corresponding component names (KB1, KB2) and / or characteristic parameter names (KPB1, KPB2) are read in from a construction plan (BP) of the technical system (TS).

11. The method according to claim 1 or 2, characterized in that A plurality of training feature parameter values ​​(TKPW) are extracted from a plurality of parameterized planning data records (PR). 12 . An apparatus (AS) for computer-aided design of a technical system (TS), the apparatus being configured to carry out the method according to claim 1 . 13 . A computer program product configured to carry out the method according to claim 1 .

14. A computer-readable storage medium having the computer program product according to claim 13.

Images