Enhancing Formability of GH4169 Tubes in Cold Drawing Processes by Uncertain Optimization Using Neural Network-Based Surrogate Models

This study focuses on enhancing the formability of GH4169 tubes during the cold drawing process through uncertain optimization using neural network-based surrogate models. First, the mandrel and mandrel-free drawing processes are analyzed, and finite element models are developed for establishing the database. Second, deterministic and uncertain optimization models for the drawing processes are formulated: the objective for mandrel drawing is to minimize the difference between drawing stress and fracture strength, while the objective for mandrel-free drawing is to reduce tube wall thickness variation; the design variables are friction coefficient, die angle, and sizing-zone length. Then, neural network-based surrogate models for uncertain optimization are developed, trained, and validated. Finally, deterministic and uncertain optimization are performed for the mandrel and mandrel-free drawing of capillary tubes. The results indicate that the fracture probability during mandrel drawing decreased from 60.2% after deterministic optimization to 28.3% after uncertainty optimization. For mandrel-free drawing, the probability of wall thickness fluctuation exceeding 1% decreases from 73.7% to 61.5%. Cold drawing experiments confirmed that uncertainty optimization significantly enhances the forming stability of the mandrel drawing process for GH4169 capillary tubes. The method developed in this study contributes to advancing the high-precision manufacturing of high-performance capillary tubes.