Multi-modal Deep Learning Surrogate Model for High-Fidelity Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning-based surrogate models for high-fidelity simulations are limited to single design scopes, making them inadequate for practical design tasks that require consideration of multiple design parameters, such as geometry, layout, and boundary conditions, leading to inefficiencies in simulation processes like magnetic field, fluid dynamics, and heat diffusion.
Innovation Solution
The development of multi-modal deep learning-based surrogate models that utilize multiple AI models to generate high-fidelity simulations by fusing information from various design scopes through concatenation, gating, pooling, averaging, or tensor-based approximation, while incorporating physics constraint models to ensure feasibility and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional Finite Element Method (FEM) simulation is used, then high-fidelity simulation accuracy is achieved, but computational cost and time consumption increase significantly
Solution Approach 1:
The patent creates a surrogate model that copies the essential input-output behavior of the complex FEM simulation system. Instead of running the full FEM simulation every time, the trained neural network model serves as a lightweight copy that reproduces simulation results much faster, achieving high-fidelity predictions without the computational burden of the original FEM process
Solution Approach 2:
The patent performs preliminary training of the neural network model using FEM simulation data before actual use. This preliminary action involves collecting training data through FEM simulations, training the model offline, and then deploying the trained model for rapid predictions. The computationally expensive FEM simulations are performed once during training rather than repeatedly during actual simulation tasks
2Loss of time
If single-scope deep learning surrogate models are used, then computational cost is reduced, but the ability to handle multiple design scopes (geometry, layout, boundary conditions) is limited
Solution Approach 1:
The patent designs a universal multi-modal neural network architecture that can handle multiple design scopes simultaneously. The model accepts multiple input modalities (geometry, layout, boundary conditions) and processes them through shared and modality-specific neural network components, enabling a single model to perform multiple simulation tasks across different design scopes without requiring separate models for each scope
Solution Approach 2:
The patent segments the input data into multiple modalities corresponding to different design scopes (geometry, layout, boundary conditions). Each modality is processed through dedicated neural network branches that extract modality-specific features, which are then fused to produce the final prediction. This segmentation allows the model to handle diverse design scopes while maintaining specialized processing for each type of input
Data Source
AI summary
A method of using multiple artificial intelligence models for generating a high fidelity simulation includes generating, by a computing device, multiple artificial intelligence models. Each artificial intelligence model simulating an industry design process. The computing device further fusing the multiple artificial intelligence models to generate a best-fit proposed industry design process. The computing device utilizes a physics constraint model to determine whether the best-fit proposed industry design process is feasible. The best-fit proposed industry design process is displayed in response to determining that the best-fit proposed industry design process is feasible.