Component Simulation Surrogates for Repeating Geometry Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current simulation methods for components are computationally inefficient, especially when dealing with complex geometries and varying boundary conditions, which hinders real-time simulation capabilities in design and operational phases.
Innovation Solution
The method involves generating a surrogate model using physically informed machine learning techniques, such as autoencoders and physically informed neural networks, to simplify simulations by parameterizing geometry and boundary conditions, reducing the complexity of the simulation problem and enabling faster computation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If traditional simulation methods are used for components with repeating geometric shapes and varying boundary conditions, then simulation accuracy is maintained, but simulation time and computational resources are excessively high
Solution Approach 1:
The component is divided into repeating geometric shapes (RGS) that can be independently identified and processed. Each RGS is treated as a separate unit with its own surrogate model, allowing parallel computation and reducing overall simulation time while maintaining accuracy through modular analysis
Solution Approach 2:
Surrogate models are created as simplified copies of the full physics models for each repeating geometric shape. These surrogate models capture the essential physics behavior but require significantly less computational resources, enabling rapid evaluation while preserving simulation accuracy through physics-informed machine learning
2Measurement precision
If the number of parameters for boundary conditions and geometry is increased to maintain simulation accuracy, then simulation fidelity is improved, but computational efficiency decreases
Solution Approach 1:
The approach transforms the simulation from using many detailed parameters to using a few key parameters that define the repeating geometric shapes and their boundary conditions. The physics-informed surrogate models learn the complex parameter relationships during training, enabling accurate predictions with reduced parameter sets and faster computation
3Reliability
If traditional numerical methods are applied to solve the complete simulation problem, then comprehensive physical accuracy is achieved, but real-time simulation capability is not attained
Solution Approach 1:
Physics-informed surrogate models are pre-trained offline using traditional numerical methods to learn the underlying physics of the repeating geometric shapes. Once trained, these surrogate models can be rapidly evaluated in real-time applications without requiring the full computational power of the original numerical methods, thus achieving both accuracy and speed
Solution Approach 2:
The traditional mechanical numerical simulation system is replaced with a machine learning-based surrogate model system. The physics knowledge is embedded into the neural network architecture and training process, allowing the surrogate models to capture physical behavior while enabling real-time computation that traditional mechanical methods cannot achieve
Data Source
Figure 1~2
AI summary
The invention relates to a method for simulating a component (12) using an electronic computing device (10), comprising the steps of: - providing a digital model (16) of the component (12) to be simulated; - determining at least one repeating geometric shape (18) of the component (12) within the digital model (16); - generating a substitute model (14) for the at least one geometric shape (18); and - simulating the component (12) by evaluating at least one simulation parameter (20), wherein the simulation parameter (20) is applied to the substitute model (14). The invention further relates to a computer program product, a computer-readable storage medium, and an electronic computing device (10).