Engineering Component 3D Field Prediction via MLP-VAE Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating 3D engineering components, such as turbine blades, are costly and resource-intensive due to the need for repeated simulations and data storage of high-volume CFD and FEA data, which cannot be efficiently reused for future optimizations.
Innovation Solution
A method using a Hybrid Multilayer Perceptron-Variational Autoencoder (MLP-VAE) model to compress and predict 3D field data of engineering components, directly mapping geometry parameters to result data, eliminating the need for costly simulations and data storage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional iterative geometry optimization using CAE simulations is used, then design accuracy is improved, but computational cost and time consumption increase significantly
Solution Approach 1:
The method performs preliminary actions by training the MLP-VAE model on a dataset of geometry-parameter to field-data mappings before actual optimization. This pre-trained model can then rapidly predict field data for new geometry variations without requiring full CAE simulations, significantly reducing optimization time while maintaining accuracy
Solution Approach 2:
The invention creates a computational copy (the MLP-VAE surrogate model) that replicates the behavior of expensive CAE simulations. This copy can be queried rapidly for design evaluations, avoiding the need to run time-consuming simulations for each design iteration while preserving the essential physics-based predictions
2Adaptability or versatility
If conventional CAE simulations are conducted for multiple geometry variations, then design exploration is improved, but data storage requirements increase dramatically
Solution Approach 1:
The method extracts only the essential mapping relationship between geometry parameters and field data by training the MLP-VAE model on a limited dataset. Once trained, the model stores only the learned parameters and weights rather than the full simulation datasets, enabling extensive design exploration with minimal storage requirements
Solution Approach 2:
The invention transforms the problem from storing large volumes of simulation data to storing and manipulating a compact set of model parameters. The MLP-VAE learns to represent complex field data through a small number of parameters that can be efficiently stored and used to generate predictions for any geometry variation
3Reliability
If repeated CAE simulations are performed for robust design variations, then design robustness is improved, but computational resources are excessively consumed
Solution Approach 1:
The method performs preliminary training of the MLP-VAE model on a diverse set of geometry variations that cover the robustness design space. This pre-trained model can then rapidly evaluate numerous design scenarios to assess robustness without requiring additional expensive CAE simulations for each evaluation
Solution Approach 2:
The computational copy (MLP-VAE model) enables rapid evaluation of design robustness across multiple scenarios. Instead of running energy-intensive CAE simulations for each robustness assessment, the pre-trained model provides accurate predictions at minimal computational cost, allowing comprehensive robustness analysis
Data Source
Figure 1~2(d)
Figure 3~5
Figure 6(a)~6(b)
AI summary
The invention relates to a method of creating 3D field data of at least one specimen of an engineering component, comprising the following steps: S1 - Obtaining a first set of 3D field data related to a first group of specimens of the engineering component, the 3D field data comprising geometry data and result data; S2 - Defining at least one geometry parameter of the first group of specimens; S3 - Training a variational autoencoder model (VAE) to compress the 3D field data to a latent vector and restore it from the latent vector; S4 - Splitting the VAE into an encoder model and a decoder model, wherein nodal weights of the encoder model and decoder model learned in step S3 are set permanent for the steps S5 to S8 of the method; S5 - Connecting a multilayer perceptron network model (MLP) to an input layer of the decoder model of the VAE to form a Hybrid Multilayer Perceptron - Variational Autoencoder model (MLP-VAE) ; S6 - Training the MLP-VAE to map values of the at least one geometry parameter of the first group of specimens to the first set of 3D field data; S7 - Defining at least one value of the at least one geometry parameter to at least partially define geometry data of at least one additional specimen of the engineering component not included in the first group of specimens; S8 - Using the trained MLP-VAE to predict 3D field data related to the at least one additional specimen, comprising directly mapping the respective at least one value of the at least one geometry parameter to respective predicted result data of the at least one specimen.