Virtual Fabric Drape Parameter Estimation With Neural Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for estimating physical property parameters in virtual clothing simulation lack accuracy and efficiency, particularly in reproducing the drape of fabrics, as they either rely solely on mechanical devices or learning models without combining speed and flexibility.
Innovation Solution
A method utilizing a neural network trained on the correlation between physical property parameters and fabric draping, updating these parameters based on error minimization through an optimizer, and generating drape simulation results using a spring mass model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a mechanical device is used to measure fabric properties, then measurement accuracy is improved, but measurement speed and efficiency deteriorate
Solution Approach 1:
The patent replaces mechanical measurement devices with a neural network-based estimation system. The neural network learns the correlation between physical property parameters and fabric drape characteristics from training data, enabling rapid prediction without mechanical measurement equipment. This substitution maintains accuracy while dramatically improving speed.
Solution Approach 2:
The patent creates a virtual copy of the fabric drape simulation process through the neural network. Instead of physically measuring fabrics, the system copies the relationship between physical properties and drape outcomes from training data, allowing instant prediction of fabric behavior without actual measurement or physical simulation.
2Productivity
If a learning model is used to estimate physical property parameters, then estimation speed is improved, but accuracy and flexibility deteriorate
Solution Approach 1:
The patent implements a feedback mechanism where the neural network predicts physical property parameters, the drape simulator uses these parameters to generate virtual fabric drape results, and the system compares these results against actual target data. The error between predicted and actual results feeds back to refine the parameter estimation, improving accuracy while maintaining speed.
Solution Approach 2:
The patent dynamically adjusts physical property parameters based on feedback from drape simulation results. The system modifies parameters such as stretch stiffness and bending stiffness according to the error between simulated and target drape characteristics, enabling continuous optimization of accuracy while preserving the speed advantage of neural network estimation.
3Manufacturing precision
If comprehensive optimization is performed to improve simulation accuracy, then manufacturing precision is improved, but computational time and complexity increase
Solution Approach 1:
The patent performs preliminary action by pre-training the neural network on a comprehensive dataset of fabric drape characteristics and physical property correlations. This pre-training captures essential drape patterns and material relationships in advance, allowing rapid prediction during actual design iterations without requiring time-consuming real-time optimization.
Solution Approach 2:
The patent segments the complex drape simulation process into two distinct stages: a fast neural network-based parameter estimation stage and a more computationally intensive drape simulation verification stage. This segmentation allows the system to perform rapid initial predictions using the neural network, then only perform detailed simulation optimization when necessary, reducing overall computational time while maintaining precision.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
Disclosed are a method and an apparatus for estimating a physical property parameter. According to an embodiment, the method for estimating a physical property parameter for a drape simulation of a virtual fabric comprises the steps of: generating a mesh by applying a physical property parameter corresponding to a virtual fabric to a neural network; acquiring drape data corresponding to the type of target data related to drape of the virtual fabric, on the basis of the mesh; and updating the physical property parameter on the basis of an error between the acquired drape data and the target data.