3D Fabric Draping Simulation Using Neural Network

The patent uses machine learning models to copy and learn from the complex nonlinear relationships between physical property parameters and fabric draping shapes observed in training data. Instead of requiring users to manually tune parameters based on theoretical physics, the system copies the patterns from trained models to generate accurate simulations quickly, resolving the contradiction between simulation accuracy and parameter tuning difficulty

The patent transforms the approach by changing from traditional physics-based parameter tuning to machine learning-based parameter prediction. The system uses pre-trained neural networks that have learned optimal parameter relationships from extensive simulation or experimental data, allowing users to obtain accurate draping results without manually adjusting complex physical parameters, thus significantly reducing tuning time and improving ease of operation

The patent applies preliminary action by pre-training machine learning models offline using extensive simulation data or experimental measurements before actual use. The complex training process is completed in advance, and during actual draping simulation, the system only needs to infer results from the pre-trained model using user input parameters. This separates the complex computational task (training) from the usage phase, enabling fast simulation while managing system complexity through pre-computation

The patent introduces machine learning models as intermediary components between simple user inputs (physical property parameters) and complex fabric draping outputs. The ML model acts as a mediator that encapsulates the complex nonlinear relationships, translating straightforward parameter inputs into accurate 3D draping shapes without requiring users to understand or compute the underlying complexity, thus achieving fast simulation while managing system complexity through abstraction