Premium
Design optimization of extrusion‐blow‐molded parts using prediction‐reliability‐guided search of evolving network modeling
Author(s) -
Yu JyhCheng,
Juang JyhYeong
Publication year - 2010
Publication title -
journal of applied polymer science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.575
H-Index - 166
eISSN - 1097-4628
pISSN - 0021-8995
DOI - 10.1002/app.31954
Subject(s) - taguchi methods , process (computing) , reliability (semiconductor) , computer science , convergence (economics) , artificial neural network , molding (decorative) , genetic algorithm , design of experiments , mathematical optimization , blow molding , algorithm , engineering , mathematics , mechanical engineering , artificial intelligence , materials science , machine learning , mold , power (physics) , physics , quantum mechanics , economics , composite material , economic growth , operating system , statistics
Abstract Industries often adopt a two‐stage design for blow‐molded parts. The part thickness distribution is first determined by structural analysis to satisfy loading requirements, and this is followed by programming of the die‐gap opening to realize the thickness distribution. This study proposes a soft‐computing‐based optimization scheme integrating part design and molding process control to search for the die‐gap programming of the molding process with minimum part weight while satisfying performance constraints. Finite element analysis tools are applied to simulate the extrusion‐blow‐molding process and structural analysis. To reduce the number of simulations, the proposed scheme first establishes a neural network (NN) model from a small experimental design to simulate the system response, and it searches for the model optimum with a genetic algorithm (GA). Because the prediction generality of an NN from small training samples will be limited, this work proposes fuzzy reasoning for the prediction reliability of the model to guide the GA search for a quasi‐optimum. The verification of the optimum is added to retrain the model, and the process iterates until convergence is reached. The iteration automatically distributes additional samples in the most probable space of the design optimum for the evolving model and improves the sampling efficiency. A high‐density polyethylene bottle design is presented to illustrate the application and for comparison with the Taguchi method and a simple iteration of NN and GA. The proposed scheme outperforms the other two and provides a feasible optimum from a robust convergence. © 2010 Wiley Periodicals, Inc. J Appl Polym Sci, 2010