Method and system for selecting an AI-based method for integration into a CAE application

Abstract

The invention relates to a method for selecting an AI-based method for integration into a CAE (Computer Aided Engineering) application for a technical object or component design, in particular in vehicle development, comprising the following steps - capturing historical text data (350) from documents of a plurality of projects (P1, P2, ..., Pn), wherein each project (Pj) comprises at least one CAE application with at least one integrated AI method for a technical task such as component design; - generating training data (450) from the text data (350), wherein the text data (350) of each project (Pj) is transformed into tokens and vectors; - training an AI language model (430) with the training data (450), wherein the AI language model (430) learns through training to assign different AI methods to different CAE applications; - collecting input data (250) for a request for a future project (Pz), wherein the project (Pz) comprises a CAE application and the possibility of integrating an AI method into this CAE application is requested; - generating at least one recommendation (470) for selecting an AI-based method for the CAE application by the AI language model (430); - Output the recommendation (470) as output data (550).

Description

[0001] The invention relates to a method, a system and a computer program product for selecting an AI-based method for integration into a CAE (Computer Aided Engineering) application for a technical object or a component design, in particular in vehicle development.

[0002] In vehicle development and design, the CAE (Computer-Aided Engineering) process is a key step in which computer simulations and analyses are used to evaluate the behavior of components or the entire vehicle in different situations. The CAE process typically consists of several steps. First, a virtual model of the vehicle or individual components is created. This involves converting the geometry of the respective component or the vehicle geometry into a digital format. Then, the properties of the materials from which the components are made are defined in the model. This includes mechanical, thermal, and other material properties. The virtual model of the component or vehicle is decomposed into a network of small elements to enable numerical analysis. This step is crucial as it forms the basis for the simulation.CAE software is used to perform simulations to analyze the behavior of the vehicle or its components under various conditions. These can include structural analyses, crash simulations, thermodynamics, fluid dynamics, and other areas. The simulation results can then be analyzed and interpreted. Based on the results, design and material adjustments can then be made to optimize the design and improve the vehicle's performance, safety, or efficiency.

[0003] The CAE process offers several advantages, including the ability to reduce costly physical prototypes, shorten development time, and deepen the understanding of vehicle behavior in different scenarios. It is an integral component in modern vehicle development.

[0004] Alongside computer-aided engineering (CAE), artificial intelligence (AI) is playing an increasingly important role in vehicle development and engineering in general. However, CAE and AI are often used in different areas of vehicle development. Nevertheless, there are some connections and overlaps between CAE and AI. CAE is often used for simulation, optimization, and design. AI can support these processes by employing automated optimization methods, genetic algorithms, or machine learning to achieve better design and simulation results. AI, especially machine learning, can, for example, be integrated into CAE processes to create improved models of a vehicle or component and thus improve complex simulations. This can help develop more realistic models and increase the accuracy of CAE simulations.In particular, AI can be integrated into CAE processes to automate iterative tasks, optimize parameterization, and thus make the simulation process more efficient overall.

[0005] While CAE is used for virtual prototyping and testing to reduce the number of physical prototypes, AI can help create more realistic virtual prototypes and improve the accuracy of test simulations. Furthermore, AI techniques such as machine learning can be used in CAE applications to analyze large amounts of data and predict simulation behavior. This can help identify patterns in simulation results. This is particularly useful in fault detection and diagnosis and can help identify anomalies in simulation results.

[0006] In particular, simulations can be made adaptive through the use of AI in CAE models. This means that the CAE model learns during the simulation and adapts to changing conditions or unknown environments. This allows AI algorithms to automatically generate innovative designs that meet specific performance targets, significantly accelerating the design process.

[0007] The integration of artificial intelligence (AI) into computer-aided engineering (CAE) therefore offers enormous potential for more efficient, accurate, and innovative solutions in vehicle development and other engineering applications. However, CAE users are often uncertain about when integration should occur and which AI methods are best suited for specific CAE process steps. Because CAE encompasses a wide variety of applications, including structural analysis, flow simulation, heat transfer, acoustics, and many others, the optimal use of AI can vary depending on the specific CAE application. Furthermore, CAE applications and AI require different types and amounts of data. Therefore, it is important to understand which data is relevant for AI and how it can be integrated into the CAE process. Another aspect is the rapid development of AI technologies.Here, it's sometimes difficult for developers to keep track of which methods are best suited for their specific application. The usability of AI tools and platforms can also vary, so a lack of familiarity leads to developer reluctance.

[0008] The object underlying the invention is to enable better integration of AI methods into various CAE processes for engineering applications such as vehicle design, thus creating more efficient, accurate, and innovative solutions in various engineering applications. In particular, the aim is to create the possibility of predicting, specifically for a specific design task, whether AI methods can be integrated into CAE processes and which AI methods are suitable for which CAE process step in a CAE application.

[0009] This object is achieved according to the invention with respect to a method by the features of patent claim 1, with respect to a system by the features of patent claim 7, and with respect to a computer program product by the features of patent claim 12. The further claims relate to preferred embodiments of the invention.

[0010] The method according to the invention enables the efficient selection of an AI-supported method approach for a technical object or component design within a computer-aided engineering (CAE) application. This allows the design process to be optimized and the performance of the CAE application to be improved through the use of advanced AI algorithms. Using the trained AI language model, a suitable AI method for the respective CAE application, such as a component design, can be selected from various AI-based methods.

[0011] By continuously providing new training data from new projects, a continuous further development of the system according to the invention is possible, so that the use of the latest AI technologies and algorithms is also guaranteed in future projects.

[0012] According to a first aspect, the invention provides a method for selecting an AI-based method for integration into a CAE (Computer Aided Engineering) application for a technical object or component design, particularly in vehicle development. The method comprises the following steps: - Collecting historical text data from documents of a large number of projects, where each project includes at least one CAE application with at least one integrated AI method for a technical task such as component design; - Generating training data from the text data, transforming the text data of each project into tokens and vectors; - Training an AI language model with the training data, whereby the AI language model learns through training to assign different AI methods to different CAE applications; - Collecting input data for a request for a future project, where the project includes a CAE application and the possibility of integrating an AI method into this CAE application is requested; - Transforming the input data into tokens for processing by the AI language model; - Generating at least one recommendation for selecting an AI-based method for the CAE application by the AI language model; - Output the recommendation as output data.

[0013] In a further development, the AI language model is intended to use neural networks and / or recurrent neural networks and / or convolutional neural networks and / or a transformer architecture.

[0014] In an advantageous embodiment, the transformer architecture uses an attention algorithm.

[0015] In a further embodiment, the AI method comprises the use of an AI model with neural networks, deep learning with deep neural networks, reinforcement learning, support vector machines (SVM), K-means clustering, Bayesian networks, genetic algorithms, recurrent neural networks (RNN), convolutional neural networks (CNN), transfer learning and / or natural language processing (NLP).

[0016] Advantageously, the vectors of the training data are embedded in an embedding model.

[0017] In particular, one possible CAE application is the simulation of the behavior of a dummy in a vehicle.

[0018] According to a second aspect, the invention provides a system for selecting an AI-based method for integration into a CAE (Computer Aided Engineering) application for a technical object or component design, particularly in vehicle development. The system comprises an input module, a database, an AI language model, and an output module, wherein the database is configured to capture historical text data from documents of a plurality of projects, each project comprising at least one CAE application with at least one integrated AI method for a technical task such as component design; wherein the database and / or the AI language model is / are configured to generate training data from the text data, wherein the text data of each project is transformed into tokens and vectors;wherein the AI language model is configured to be trained with the training data, wherein the AI language model learns through training to associate different AI methods with different CAE applications; wherein the input module is configured to capture input data relating to a query for a future project, wherein the project comprises a CAE application with the possibility of integrating an AI method into this CAE application; wherein the input module and / or the AI language model is / are configured to transform the input data into tokens for processing by the AI language model; wherein the AI language model is configured to generate at least one recommendation for selecting an AI-based method for the CAE application; and wherein the output module is configured to output the recommendation as output data.

[0019] In a further development, it is provided that the AI language model uses neural networks and / or recurrent neural networks and / or convolutional neural networks and / or a transformer architecture, and that the vectors of the training data are embedded in an embedding model.

[0020] In an advantageous embodiment, the transformer architecture uses an attention algorithm.

[0021] In a further embodiment, the AI method comprises the use of an AI model with neural networks, deep learning with deep neural networks, reinforcement learning, support vector machines (SVM), K-means clustering, Bayesian networks, genetic algorithms, recurrent neural networks (RNN), convolutional neural networks (CNN), transfer learning and / or natural language processing (NLP).

[0022] In particular, one possible CAE application is the simulation of the behavior of a dummy in a vehicle.

[0023] According to a third aspect, the invention provides a computer program product comprising executable program code configured to carry out the method according to the first aspect when executed.

[0024] The invention is explained in more detail below with reference to embodiments shown in the drawing.

[0025] It shows: Fig. 1 is a block diagram illustrating an embodiment of a system according to the invention; Fig. 2 a flow chart explaining the individual method steps of a method according to the invention; Fig. 3 is a block diagram of a computer program product according to an embodiment of the third aspect of the invention.

[0026] Additional features, aspects and advantages of the invention or embodiments thereof will become apparent from the detailed description taken in conjunction with the claims.

[0027] Fig. Figure 1 shows a system 100 according to the invention for selecting an AI-based method for integration into a CAE (Computer Aided Engineering) application for a technical object or component design, particularly in vehicle development. The system 100 according to the invention comprises an input module 200, a database 300, an evaluation module 400, and an output module 500.

[0028] The input module 200, the database 300, the evaluation module 400 and the output module 500 can each be provided with a processor and / or a memory unit.

[0029] In the context of the invention, a "module" can therefore be understood as, for example, a processor and / or a memory unit for storing program instructions. For example, the processor is specifically configured to execute the program instructions, so that the processor and / or the control unit performs functions to execute or implement the method according to the invention or a step of the method according to the invention.

[0030] In the context of the invention, a "processor" can be, for example, a machine or an electronic circuit. A processor can in particular be a main processor (Central Processing Unit, CPU), a microprocessor or a microcontroller, e.g., an application-specific integrated circuit or a digital signal processor, possibly in combination with a memory unit for storing program instructions. A processor can also be a virtualized processor, a virtual machine or a soft CPU. It can, for example, also be a programmable processor that is equipped with configuration steps for carrying out the aforementioned method according to the invention or is configured with configuration steps such that the programmable processor implements the inventive features of the method, the component, the modules or other aspects and / or sub-aspects of the invention.

[0031] In the context of the invention, a "storage unit" or "storage module" and the like can be understood as, for example, a volatile memory in the form of random access memory (RAM), a permanent memory such as a hard disk or a data storage device, or, for example, a replaceable storage module. The storage module can also be a cloud-based storage solution.

[0032] In the context of the invention, “data” refers to both raw data and already processed data, e.g. from measurement results from sensors or from simulation results.

[0033] In addition, communication connections are provided for the exchange and transmission of data between the individual modules, which are designed in particular as wireless communication connections, e.g., as a mobile radio connection (e.g., 4G LTE, 5G, 6G) and / or as near-field communication connections, e.g., Bluetooth®, Ethernet, NFC (Near Field Communication) or Wi-Fi®.

[0034] In particular, the database 300 and the analysis module 400 are integrated in a cloud computing infrastructure. A cloud computing infrastructure offers the ability to expand or reduce resources as needed, allowing computing power, storage space, or network resources to be easily adapted to changing requirements. This scalability enables cost optimization and efficient resource allocation without large investments in hardware. Furthermore, users can access applications and data from anywhere with internet access. This accessibility enables collaboration and seamless integration across different devices and locations. Cloud computing also offers flexibility in software deployment, allowing applications to be deployed and updated quickly and without interruption.Cryptographic encryption methods can also be used to protect the connection to the cloud computing infrastructure via a mobile connection.

[0035] The database 300 contains text data 350 for the process description of projects P1, P2, ..., Pn, which contain CAE applications combined with various artificial intelligence methods. A project Pj can, for example, relate to crash simulations involving a vehicle and a dummy inside. To simulate a collision situation, for example, a crash simulation is defined with predefined simulation parameters such as vehicle speed, obstacle properties, dummy seat position, seat design, vehicle interior fixtures, and safety devices such as airbags and seatbelts. Using this simulation configuration, a series of simulations is then performed. For example, the acceleration of a dummy's head in a vehicle can be simulated.

[0036] In an initial project (P1), artificial intelligence (AI) methods were also used to achieve more realistic and accurate simulation results. For this purpose, the data relevant to the behavior of a dummy in the vehicle, such as vehicle movements, impact forces, accelerations, and other physical variables, were prepared in such a way that they could be processed by an AI model. Furthermore, the first project (P1) defined which target variables or aspects of the dummy's behavior should be improved by the AI methods. These could be, for example, loads on specific body parts, head injuries, or other safety-relevant parameters. For this purpose, the method of machine learning was chosen to train an AI model to learn specific patterns and relationships from existing data. From the data that serves as input to the AI model, features relevant for training the AI model are identified and extracted.This can be, for example, the temporal progression of accelerations or the position of the dummy in the vehicle.

[0037] In a second project, P2, simulations were also conducted with the dummy, but with a different passenger cabin design. This project investigated how the forces acting on the dummy change depending on the choice of cabin material. Because the problem was more complex, deep learning techniques with deep neural networks were used as the AI model. Preparing the data for training the AI model was also more complex, as the AI model was required to learn more complex relationships.

[0038] In a third project, P3, reinforcement learning was used to answer the question of which interactive actions can influence the behavior of dummies. The challenge in this project was to conduct the AI model training process with well-prepared data to avoid overfitting. Overfitting is a common problem in the development of AI models. Overfitting occurs when a model is too well adapted to the training dataset, including random variations and noise. The AI model learns the data patterns and nuances so precisely that it has difficulty making generalized predictions for new data. However, the AI model developed in project P3 was successfully integrated into the CAE simulation of dummy behavior in the vehicle and demonstrated how changing scenarios influence the dummy's movements, deformations, or reactions.

[0039] What all projects P1, P2, and P3 have in common is that the combination of AI and CAE increased the accuracy of dummy simulations and thus improved vehicle safety.

[0040] In addition to the AI methods mentioned above, such as an AI model with neural networks, deep learning techniques with deep neural networks, and reinforcement learning, there are a number of other AI-based methods and techniques. These include - Support Vector Machines (SVM): A supervised learning algorithm used for classification and regression tasks. SVMs search for an optimal dividing line between different classes in a dataset. - K-means clustering: An unsupervised learning technique used to group data points into coherent clusters. It is particularly useful for pattern recognition and segmentation. - Bayesian networks: A probabilistic approach that uses probabilities to model relationships between different variables. Bayesian networks find application in inference and diagnosis. - Genetic algorithms: Genetic algorithms are inspired by evolutionary theory and are used for optimization and search problems. They use evolutionary steps such as mutation, crossing, and selection to find optimal solutions. - Recurrent Neural Networks (RNN): A special type of neural network specialized for processing sequential data. RNNs have applications in natural language processing, time series analysis, and other context-dependent problems. - Convolutional Neural Networks (CNN): Specialized neural networks used for processing raster data such as images. CNNs have proven particularly effective in image recognition and classification. - Transfer Learning: An approach in which a trained model is applied to another, similar task. This allows the model to benefit from previously learned features and reduce training effort. - Natural Language Processing (NLP): A subfield of AI that deals with the processing and understanding of human language.

[0041] Hybrid approaches can also be used, combining multiple AI techniques to achieve the best results.

[0042] The evaluation module 400 contains a language model 430 for artificial intelligence (AI), in particular for natural language processing (NLP). The AI language model 430 is based on neural networks, in particular deep neural networks and / or transformer architectures. These networks learn complex relationships and patterns in the text.

[0043] Neural networks are often cited as prototypes for AI algorithms. Neural networks are also interesting because they can be combined to increase learning capacity. Such coupled neural networks are also known as deep learning. A neural network consists of neurons arranged in multiple layers and connected to each other in various ways. A neuron is capable of receiving information from outside or from another neuron at its input, evaluating it in a specific way, and passing it on in a modified form at the neuron's output to another neuron or outputting it as the final result. Hidden neurons are located between the input neurons and the output neurons. Depending on the type of network, there can be several layers of hidden neurons. They ensure the forwarding and processing of information. Output neurons ultimately provide a result and pass it on to the outside world.The arrangement and connection of neurons creates different types of neural networks, such as feedforward neural networks (FNNs), recurrent neural networks (RNNs), or convolutional neural networks (CNNs). These neural networks can be trained using unsupervised or supervised learning.

[0044] The Transformer architecture is an advanced and highly scalable architecture for neural networks. The Transformer architecture consists of an encoder-decoder structure. The encoder processes the input sequence of data inputs, while the decoder generates the output sequence. In some applications, only the encoder can be used. The core of the Transformer architecture is the attention algorithm, which allows the AI model to focus on different parts of the input sequence when generating outputs. This mechanism helps handle complex dependencies in the data. Multiple attention computations can be performed in parallel. The results are then combined and projected linearly. The Transformer architecture provides an efficient way to capture complex relationships in sequences and is particularly suitable for large datasets.

[0045] The AI language model 430 is trained with the text data 340 of the projects P1, P2, ..., Pn. It is important to provide a sufficient number of projects P1, P2, ..., Pn with a wide range of data to avoid overfitting the language model 430. The performance of the AI language model 430 depends heavily on the quality of the training data and the network architecture of the AI language model 430.

[0046] The text data of projects P1, P2, ..., Pn can consist of various types of text documents created during the respective project. For example, it can be a project requirements document that defines the basic requirements for the CAE project. Objectives, specifications, and other project-related requirements are described here. Simulation reports summarize the simulation results. They often contain descriptions of the CAE models, assumptions made, results, and conclusions. They also contain text documents detailing the simulation models used, simulation techniques, parameters, and boundary conditions. In addition, the text documents of projects P1, P2, ..., Pn can contain validation and verification reports that compare the simulation results to ensure that the simulations correspond to real-world conditions.If custom code was used, comprehensive code documentation may also be available, including descriptions of functions, variables, and other relevant information. The text data may also include text documents used for presentations and meetings.

[0047] In addition, the text data 350 contains a report that provides a general overview of the AI methods used in the CAE project. It explains the selected algorithms, models, and techniques, including their applications and advantages. In addition to a general description of the AI methods, documentation of the data used for the AI model may also be included. This report describes the dataset, potential challenges, and the preprocessing process, including normalization, cleaning, and feature engineering. Furthermore, the report contains a detailed description of the AI model's architecture. This describes the layers, neurons, connections, and all other relevant aspects of the model. Furthermore, the AI model's training process may be documented in the report.This includes information about the hyperparameter settings, training duration, the selection of optimization algorithms, and performance metrics for evaluating the AI model's performance. A report on the interpretability and explainability of the AI model can also be provided to make the AI model's decisions understandable. Furthermore, a report on the integration of the AI model into the CAE simulation can be provided, describing the integration process, interfaces, and collaboration with existing CAE tools.

[0048] The exact structure and content of the reports may vary depending on the specific requirements of each CAE project. However, since documentation plays a crucial role in a component or vehicle development project, a large number of text documents are typically available for a project Pj.

[0049] The database 300 and / or the language model 430 contain a software application for processing the text data 350 from the various projects P1, P2, ..., Pn. In this process, irrelevant information can first be removed from the collected text data 350. Subsequently, the text data 350 from the various projects P1, P2, ..., Pn is broken down into individual tokens (words, word components, or characters). These tokens can then be converted into multidimensional vectors to be represented in an embedding model. Each token is thus represented by its corresponding vector. These vectors can be generated using pre-trained word embeddings such as Word2Vec, GloVe, or FastText. Embedding models are often used as part of transformer architectures, particularly in the encoder section. Embedding models serve as representations of the input sequences of the text data and are optimized during model training.

[0050] To represent a text at the sentence level in the embedding model, the vectors of the sentence's words are combined. This can be done by averaging, summation, or other aggregation methods. An embedding model can thus operate at different levels, including word embedding, sentence embedding, or document embedding, depending on the specific requirements. The vectors of words, sentences, or documents can be represented in a multidimensional vector space, where querying is performed by comparing the position of the vectors in space. This transformation of the text data into vectors, which are integrated into an embedding model, allows for the detection of semantic similarities, since semantically similar words are represented by spatially close vectors in the embedding model.Text embedding thus enables a semantically meaningful text representation, in which semantically similar texts are represented by closely spaced vectors. The similarity between vectors can be assessed using metrics such as cosine similarity. A higher cosine similarity indicates greater semantic similarity between the texts.

[0051] With the help of the embedding model, the AI language model 430 can find relevant documents based on semantic similarity. Since text documents from projects P1, P2, ..., Pn with similar semantic meanings are positioned close to each other in the embedding model, they can be found when a user queries for a specific text type. Additionally, clustering algorithms can be used to find similar projects, e.g., all projects that use reinforcement learning.

[0052] By converting the text data 350 into tokens and vectors, training data 450 is created, which can now be processed by the neural networks, including the transformer architectures of the AI model 430. Furthermore, pre-trained embedding models can be used to better capture the semantic content of words or texts in the text data by leveraging previously learned semantic relationships.

[0053] The AI language model 430 is then trained with these data sets 440 using either supervised learning or unsupervised learning. In supervised learning, the model is trained with input-output pairs (e.g., sentence structures and their categories). In unsupervised learning, the AI model 430 learns patterns and structures in the text without explicit target categories.

[0054] This trained AI language model 430 can now be used to evaluate options for a future CAE project as to how and which AI methods can be used and integrated to achieve a specific result.

[0055] For this purpose, a user enters a request in the form of input data 250 into the input module 200. The input module 200 is typically a keyboard or a touchscreen. However, it can also be a voice input, which is converted into a text message by the input module 200 using appropriate language processing software. Such input of input data 250 for querying an AI language model 430 is typically referred to as a prompt. The prompt can be a question, an instruction, or a request.

[0056] A typical prompt might be: In a CAE project to improve the structural stability of the A-pillar in a vehicle cabin, the goal is to investigate the extent to which a different material choice increases safety while maintaining or improving the aesthetics of the passenger compartment. Does it make sense to use AI methods for this task, and if so, which AI methods could they be?

[0057] The AI language model 430 breaks this input data 250 into tokens to make it comprehensible for processing by the AI language model 430. The AI language model 430 processes the tokenized input and generates a recommendation 470 as a response. In particular, certain neurons and weights in the network are activated. The AI language model 430 generates the response based on the learned patterns and the information from the training process. This recommendation 470 is output as output data 550 by the output module 500 in natural language or in another agreed form.

[0058] For example, the response of the AI language model 430 in the form of the output data 550 could be: In a CAE project analyzing the stability of the A-pillar in a vehicle cabin, AI methods can be used in various aspects of the project to increase efficiency and gain more detailed insights. Here are some areas where AI approaches could be used: - Design optimization: AI can be integrated into optimization algorithms to automatically search for improved designs. By applying genetic algorithms or other evolutionary optimization methods, AI can help identify structural parameters that improve A-pillar performance under crash conditions. - Material selection: AI can assist in material selection for the A-pillar. By analyzing large data sets of materials, including their mechanical properties and costs, AI can provide recommendations for optimal materials while taking safety requirements into account. - Early fault detection: AI models with neural networks can detect potential problems in design or simulation data early on. These models can be based on anomaly detection and identify deviations from normal structural patterns or crash scenarios. - Simulation efficiency: AI can be used to increase simulation efficiency, for example, through machine learning models that can provide faster predictions instead of complex finite element calculations. This can lead to shorter simulation times. - Crash scenario analysis: AI can help identify complex patterns in crash scenario data. This could facilitate the identification of critical factors influencing A-pillar stability and lead to improved safety concepts.

[0059] A further prompt can now be used to query whether an AI model for material selection already exists. If such an AI model has already been used in one of the previous projects P1, P2, ..., Pn, the AI language model 430 can provide information about the project number under which it can be found. In this way, not only can the appropriate AI method be queried via the AI language model 430, but the database of the previous projects P1, P2, ..., Pn can also be searched specifically to be able to use project components, such as previously created AI models, for new projects.

[0060] The quality of the AI language model's responses depends on the quality of the training data, the size of the AI model, the architecture, and other factors. Furthermore, it is important to use clear and precise prompt wording to achieve the desired results.

[0061] The AI language model 430 can be further improved by fine-tuning it for specific tasks or through continuous learning. This can be achieved through feedback loops, evaluation, and continuous training with datasets from new projects.

[0062] In this way, the respective strengths, weaknesses, and areas of application of various AI-based methodological approaches can be even better captured. Recommendation 470 of the AI language model 430 regarding the appropriate selection of a specific AI-based method for use in different scenarios is thus placed on a sound basis for decision-making.

[0063] In particular, a comprehensive dataset from a large number of projects (P1, P2, ..., Pn), which ultimately represent case studies on the use of AI methods in CAE applications, allows for even better answers to questions about the performance and efficiency of various AI methods, e.g., with regard to accuracy, processing speed, and resource consumption. For example, the question of which application areas simple neural networks are most suitable for compared to deep neural networks and reinforcement learning can be better answered. Questions about the training effort required for the models for the individual methods and how complex their respective implementation is in practice can also be increasingly answered thanks to the amount of data provided by the AI language model 430.

[0064] In Fig. Figure 2 shows the procedural steps for selecting an AI-based method for integration into a CAE (Computer Aided Engineering) application for a technical object or component design, particularly in vehicle development.

[0065] In a step S10, historical text data 350 of documents of a plurality of projects P1, P2, ..., Pn are collected, wherein each project Pj comprises at least one CAE application with at least one integrated AI method for a technical task such as a component design.

[0066] In a step S20, training data 450 is generated from the text data 350, wherein the text data 350 of each project Pj is transformed into tokens and vectors.

[0067] In a step S30, an AI language model 430 is trained with the training data 450, wherein the AI language model 430 learns through training to assign different AI methods to different CAE applications.

[0068] In a step S40, the input data 250 for a request for a future project Pz are recorded, wherein the project Pz comprises a CAE application and the possibility of integrating an AI method into this CAE application is requested.

[0069] In a step S50, the input data 250 is transformed into tokens for processing by the AI language model 430.

[0070] In a step S60, at least one recommendation 470 for selecting an AI-based method for the CAE application is generated by the AI language model 430.

[0071] In a step S70, the recommendation 470 is output as output data 550.

[0072] Fig. 3 schematically illustrates a computer program product 900 comprising executable program code 950 configured to perform the method according to the first aspect of the present invention.

[0073] The method according to the invention enables the efficient selection of an AI-supported method approach for a technical object or component design within a computer-aided engineering (CAE) application. This allows the design process to be optimized and the performance of the CAE application to be improved through the use of advanced AI algorithms. Using the trained AI language model, a suitable AI method for the respective CAE application, such as a component design, can be selected from various AI-based methods.

[0074] By continuously providing new training data from new projects, a continuous further development of the system according to the invention is possible, so that the use of the latest AI technologies and algorithms is also guaranteed in future projects. Reference symbol 100 systems 200 input module 250 input data 300 database 350 text data 400 evaluation module 430 AI language model 450 training data 470 Recommendation 500 output module 550 output data 900 Computer program product 950 program code

Claims

[1] Procedure for selecting an AI-based method for integration into a CAE (Computer Aided Engineering) application for a technical object or component design, particularly in vehicle development, with the following steps - capturing (S10) historical text data (350) from documents of a plurality of projects (P1, P2, ..., Pn), wherein each project (Pj) comprises at least one CAE application with at least one integrated AI method for a technical task such as component design; - generating (S20) training data (450) from the text data (350), wherein the text data (350) of each project (Pj) are transformed into tokens and vectors; - training (S30) an AI language model (430) with the training data (450), wherein the AI language model (430) learns through the training to assign different AI methods to different CAE applications; - capturing (S40) input data (250) for a request for a future project (Pz), wherein the project (Pz) comprises a CAE application and the possibility of integrating an AI method into this CAE application is requested; - Transforming (S50) the input data (250) into tokens for processing by the AI language model (430); - generating (S60) at least one recommendation (470) for selecting an AI-based method for the CAE application by the AI language model (430); - Output (S70) the recommendation (470) as output data (550). [2] The method of claim 1, wherein the AI language model (430) uses neural networks and / or recurrent neural networks and / or convolutional neural networks and / or a transformer architecture. [3] The method of claim 2, wherein the transformer architecture uses an attention algorithm. [4] Method according to one of claims 1 to 3, wherein the AI method comprises the use of an AI model with neural networks, deep learning with deep neural networks, reinforcement learning, support vector machines (SVM), K-means clustering, Bayesian networks, genetic algorithms, recurrent neural networks (RNN), convolutional neural networks (CNN), transfer learning and / or natural language processing (NLP). [5] Method according to one of claims 1 to 4, wherein the vectors of the training data (450) are embedded in an embedding model. [6] Method according to one of claims 1 to 5, wherein a possible CAE application is the simulation of the behavior of a dummy in a vehicle. [7] A system (100) for selecting an AI-based method for integration into a CAE (Computer Aided Engineering) application for a technical object or a component design, in particular in vehicle development, comprising an input module (200), a database (300), an AI language model (400), and an output module (500), wherein the database (300) is designed to capture historical text data (350) from documents of a plurality of projects (P1, P2, ..., Pn), wherein each project (Pj) comprises at least one CAE application with at least one integrated AI method for a technical task such as a component design; wherein the database (300) and / or the AI language model (400) is / are designed to generate training data (450) from the text data (350), wherein the text data (350) of each project (Pj) is / are transformed into tokens and vectors;wherein the AI language model (430) is designed to be trained with the training data (450), wherein the AI language model (430) learns through training to assign different AI methods to different CAE applications; wherein the input module (300) is designed to capture input data (250) relating to a query for a future project (Pz), wherein the project (Pz) comprises a CAE application with the possibility of integrating an AI method into this CAE application; wherein the input module (300) and / or the AI language model (430) is / are designed to transform the input data (250) into tokens for processing by the AI language model (430); wherein the AI language model (430) is designed to generate at least one recommendation (470) for selecting an AI-based method for the CAE application; and wherein the output module (500) is configured to output the recommendation (470) as output data (550); [8] The system (100) of claim 7, wherein the AI language model (430) uses neural networks and / or recurrent neural networks and / or convolutional neural networks and / or a transformer architecture, and wherein the vectors of the training data (450) are embedded in an embedding model. [9] The system (100) of claim 8, wherein the transformer architecture uses an attention algorithm. [10] The system (100) of any one of claims 7 to 9, wherein the AI method comprises using an AI model with neural networks, deep learning with deep neural networks, reinforcement learning, support vector machines (SVM), K-means clustering, Bayesian networks, genetic algorithms, recurrent neural networks (RNN), convolutional neural networks (CNN), transfer learning, and / or natural language processing (NLP). [11] System (100) according to one of claims 7 to 10, wherein a possible CAE application is the simulation of the behavior of a dummy in a vehicle. [12] A computer program product (900) comprising an executable program code (950) configured to carry out the method according to any one of claims 1 to 7 when executed.

Images