Neural Network Simulation Encoding Surface Volumetric Points
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing numerical simulation methods for complex physical phenomena, such as fluid flow and thermal transfer, are resource-intensive and time-consuming, particularly due to difficulties in updating boundary conditions across subdomains in parallel calculations, limiting their efficiency in optimization contexts.
Innovation Solution
A neural network configuration that encodes surface and volumetric points of interest, along with simulation conditions, to generate correlated simulation values, reducing the need for iterative calculations by learning from multiple types of data during a training phase, thereby improving prediction efficiency and reducing energy consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional numerical simulation methods (CFD, FEA) are used to simulate complex physical phenomena, then simulation accuracy is maintained, but calculation time and computer resource consumption become excessively high
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing simulation results in a database for various geometric configurations and boundary conditions. During actual simulation, the system queries this pre-computed database to obtain results directly, avoiding the need to perform time-consuming numerical calculations from scratch. This transforms the simulation process from active computation to passive retrieval, dramatically reducing calculation time while maintaining accuracy.
2Productivity
If domain decomposition method with parallel calculations is used, then calculation efficiency is improved, but boundary condition updates across subdomains become complex and time-consuming
Solution Approach 1:
The patent applies segmentation by dividing the simulation problem into two independent parts: geometric parameter extraction and boundary condition determination. The system extracts geometric parameters from the input geometry and queries the database using these parameters alone, without needing to complexly coordinate boundary conditions across subdomains. This segmentation simplifies the parallel calculation architecture by eliminating the complex boundary condition update mechanism required in traditional domain decomposition methods.
3Loss of time
If machine learning techniques are applied for prediction, then calculation time is reduced, but prediction accuracy may compromise simulation precision
Solution Approach 1:
The patent applies copying by creating a comprehensive database that copies and stores exact simulation results from traditional numerical methods for various geometric configurations. The machine learning system then queries this database to retrieve pre-computed accurate results rather than relying solely on approximate predictions. This copying approach ensures that the prediction system maintains the precision of traditional methods while achieving the speed of database retrieval.
Data Source
AI summary
The disclosure relates to a neural network configured for a numerical simulation of a physical phenomenon, such as a fluid flow, a thermal transfer or a calculation of a mechanical structure, by joint learning from physical data of several types correlated with each other from a plurality of numerical training simulations.