MLP-VAE 3D Field Prediction Without Repeated Simulations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating 3D engineering components, such as turbine blades, involve costly and time-consuming simulations and data management, particularly in computational fluid dynamics (CFD) and finite element analysis (FEA), with high storage requirements for multiple geometry variations.
Innovation Solution
A method using a Hybrid Multilayer Perceptron-Variational Autoencoder (MLP-VAE) model to compress and predict 3D field data, eliminating the need for costly geometry updates and simulations by directly mapping geometry parameters to result data, achieving over 99% compression rate while preserving data details.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional simulations and measurements are conducted for each geometry variation, then accurate performance prediction is achieved, but time consumption and computational resources increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-training the VAE model on a comprehensive dataset of geometry-parameter to performance-result mappings before actual design optimization. This pre-computed knowledge is stored in the latent space, allowing rapid prediction without repeated full simulations. The encoder-decoder architecture learns the underlying relationships in advance, so when a new geometry variant is evaluated, the system can quickly map its parameters through the trained model to obtain performance predictions, avoiding time-consuming CFD/FEA simulations for each individual case.
Solution Approach 2:
The patent uses copying by creating a virtual model (the VAE neural network) that replicates the behavior of expensive physical simulations. Instead of repeatedly executing actual CFD or FEA simulations for each design variant, the system uses the trained VAE model as a surrogate that copies the essential input-output relationships. This virtual copy can be evaluated instantaneously multiple times without the computational burden of the original simulation processes, while maintaining acceptable prediction accuracy for design space exploration.
2Adaptability or versatility
If multiple geometry variations are simulated and stored, then comprehensive design data is available, but storage requirements become enormous
Solution Approach 1:
The patent applies dimensionality change by transforming the high-dimensional geometry parameter space and performance result space into a compressed latent space through the VAE encoder. Instead of storing complete 3D field data for each geometry variation (which would require hundreds of gigabytes to terabytes), the system projects this data into a lower-dimensional latent representation that captures the essential variability. This dimensional reduction allows comprehensive coverage of design space while storing only the compressed latent vectors, dramatically reducing storage requirements while preserving adaptability for generating predictions of new geometry variants.
Solution Approach 2:
The patent uses parameter changes by transforming the physical geometry parameters and performance results into latent space parameters through the trained VAE model. The system changes the parameter representation from detailed 3D field data to compressed latent vectors that encode the essential information. This parameter transformation allows the system to maintain comprehensive design exploration capability while working with a reduced parameter set that requires minimal storage, as the latent parameters can be combined and manipulated to generate predictions for unseen geometry variations.
3Loss of information
If detailed 3D field data is generated for each variant, then complete performance information is obtained, but data processing and management become complex
Solution Approach 1:
The patent applies extraction by isolating and removing the redundant information from detailed 3D field data through the VAE encoding process. The encoder extracts only the essential features and patterns that define performance characteristics, discarding redundant details that do not contribute to predictive accuracy. This extraction process produces compact latent representations that retain the critical performance information needed for design optimization while eliminating the management burden of complete detailed field data for each geometry variant.
Data Source
AI summary
A system and method of creating 3D field data of at least one specimen of an engineering component includes obtaining a first set of 3D field data, defining at least one geometry parameter, training a variational autoencoder model (VAE), splitting the VAE into an encoder model and a decoder model, 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), training the MLP-VAE to map values, defining to at least partially define geometry data of at least one additional specimen of the engineering component, using the trained MLP-VAE to predict 3D field data related to the at least one additional specimen by 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.