As part of our AI x CAE activities, we have been actively investigating and researching surrogate models. Here are some of the topics we have covered.
Also, the results of comparing and verifying the surrogate model with the FEM analysis results are discussed in the validation result page.
What is CAE simulation?
CAE stands for Computer Aided Engineering, and is a method of reducing the cost and increasing the efficiency of product design and evaluation by using computers to simulate the performance of products affected by external forces, heat, fluids, and electromagnetic waves, mainly in the product design stage.
CAE simulation has the following advantages:
・ Not only is it more cost effective and efficient than experiments using actual objects, but it also enables design and evaluation even when actual objects are not available. ・ With the recent development of PC processing speed, CAE can be applied to a wider range of applications and time can be reduced.
Problems with CAE simulation
As the application of CAE expands, more and more people are looking for higher accuracy and shorter simulation times. This has led to the following problems with CAE simulation.
As the content of the study became more and more complex, the simulation time became more and more time consuming to perform the calculations from scratch.
As CAE simulation methods have become more complex, the cost of software and hardware has increased, there has been a shortage of specialized engineers, and training costs have increased. Software maintenance costs have also risen.
Solving problems in CAE simulation
In order to meet the challenges of CAE simulation, the idea of a “surrogate model” that replaces CAE analysis (simulation) with an AI model has emerged. This model replaces traditional CAE analysis with an AI-driven approach.
A surrogate model offers several advantages in overcoming CAE simulation issues:
Reduced Analysis Time: Unlike traditional methods, AI does not need to perform calculations from scratch, significantly cutting down the time required for analysis. While training an AI model can be time-consuming, the execution time of the trained model is much faster. Additionally, AI model training can be automated, requiring minimal human intervention.
Ease of Use: Although training AI models requires specialized knowledge, evaluating and using these models does not. This makes the technology accessible and user-friendly, even for those without extensive AI expertise.
Comparison of surrogate models using conventional CAE and 2D AI
Problems with surrogate model using 2D AI
Although surrogate models can address some of the problems of CAE simulation, there are also problems with 2D AI-based surrogate models, as follows:
Incompatibility with 3D Data: In the design field, handling 3D data instead of 2D data has become the mainstream, and existing 3D data cannot be used.
Accuracy Issues: Even if AI is trained on a 3D object using 2D data, the accuracy of AI evaluation is poor, as the model struggles to capture the full complexity of 3D structures.
Data Creation and Volume: In order to train the AI, it is necessary to create new 2D data. In addition, a large amount of training data is required to create 2D data by photographing 3D objects from multiple angles. This process demands a substantial amount of training data, making it resource-intensive.
Responding to problems with surrogate model using 2D AI
In order to deal with the limitation of the 2D AI surrogate model, a shift to 3D AI surrogate model that can handle 3D data effectively is required instead of the 2D AI surrogate model.
High Accuracy: By learning from three-dimensional data, AI can achieve higher evaluation accuracy.
Utilization of Existing Data: Existing 3D data can be directly used, eliminating the need to create new 2D data.
However, handling 3D data in AI (deep learning) presents its own challenges. Techniques such as Convolutional Neural Networks (CNNs) for 3D meshes, Graph Convolutional Networks (GCNs), Transformers, and Physics-Informed Neural Networks (PINNs) are crucial for developing effective 3D AI surrogate models.
Currently, we are the only company providing this advanced technology for 3D meshes, we are the pioneer in the world in applying 3D. With these techniques, we are pioneering several groundbreaking applications:
3D Shape Classification: Classifying 3D data based on shape.
3D Shape Synthesis: Designing new 3D shapes by combining two different 3D datasets.
3D Shape Similarity: Matching 3D data that can match by shape and size.
Moreover, we are continuously researching and developing world-leading 3D AI surrogate models, pushing the boundaries of what is possible with AI in the CAE field.
Comparison of 2D AI surrogate model and 3D AI surrogate model
Preparing for Surrogate Model with 3D AI
Summary
Essential 3D Technologies: To develop 3D surrogate models, advanced technologies like 3D mesh convolution (CNN for 3D mesh), Graph Convolutional Networks (GCNs), and Physics-Informed Neural Networks (PINNs) are crucial. We are currently the only company offering this cutting-edge technology for 3D meshes.
Solving CAE Simulation Issues: A surrogate model can be used to solve the problems of CAE simulation. Furthermore, the limits of CAE simulation can be exceeded.
Limitations of 2D : Traditional AI models, which rely on 2D data, struggle with 3D data Implementing 3D surrogate models would allow for more accurate and precise simulations.
