The purpose of this paper is to propose an automatic generation method integrating parametric design and artificial neural networks for women's jackets patterns to improve the efficiency of personalized garment pattern-making.
Firstly, the structure of the jackets is divided based on the modular design concept, and a library of modular component patterns is established. Subsequently, parametric modeling is employed to analyze the parameter constraints of the patterns. For the front and back panels of the jackets, three machine learning regression prediction methods – Back Propagation Neural Network (BPNN), Radial Basis Function Neural Network (RBFNN) and Support Vector Regression (SVR) – are utilized to establish predictive models for key point coordinates. A comparative analysis of the prediction results from these three models is conducted.
The comparative analysis reveals that BPNN achieves the highest accuracy in predicting key point coordinates, enabling more precise fitting of the coordinate values of critical points in the pattern. By integrating the parametric design approach with the predictive model, an automatic pattern generation system based on the AutoCAD platform is developed. To validate the feasibility of the proposed method, virtual try-on experiments are conducted using 3D virtual fitting technology, and the garment's performance is evaluated. The results confirm the rationality and effectiveness of the proposed automatic pattern generation approach. However, this study has several limitations: the small sample size (n = 25) leads to overfitting (R = 1.0000 on the test set is unrealistic); virtual fit was evaluated only qualitatively and no comparison with human pattern makers was performed. Therefore, the results should be considered preliminary. Future work with larger datasets and quantitative fit metrics is required.
This research provides a reference for the rapid generation of garment patterns and the development of personalized garment customization.
