THESIS
2020
viii, 72 pages : illustrations (some color) ; 30 cm
Abstract
Injection molding is one of the essential polymer processing technologies, whose operation can generally be divided into five phases: filling, holding, cooling, metering, and demolding. Traditionally, the initial values of the parameter settings are highly empirical during the mold clamping phase. The lack of standard parameter tuning rules and excessive dependence on skilled engineers could significantly reduce the operating efficiency of the injection molding machines. In response, this thesis proposes to optimize parameter settings in the mold opening stage to minimize manufacturing costs.
Two optimization methods, including model-based and model-free ones, are designed. In the model-free approach, a simultaneous perturbation stochastic approximation algorithm is utilized to seek th...[
Read more ]
Injection molding is one of the essential polymer processing technologies, whose operation can generally be divided into five phases: filling, holding, cooling, metering, and demolding. Traditionally, the initial values of the parameter settings are highly empirical during the mold clamping phase. The lack of standard parameter tuning rules and excessive dependence on skilled engineers could significantly reduce the operating efficiency of the injection molding machines. In response, this thesis proposes to optimize parameter settings in the mold opening stage to minimize manufacturing costs.
Two optimization methods, including model-based and model-free ones, are designed. In the model-free approach, a simultaneous perturbation stochastic approximation algorithm is utilized to seek the optimal parameter settings. The operating parameters are moved along the direction to decrease the gradient of the cost function. On the other hand, the model-based method first uses artificial neural networks to map the operating parameters to the manufacturing cost, and then optimize the parameter sets by using a genetic algorithm. The above two methods can lead to close-to-optimum results. The model-free approach is more efficient but easily trapped by the local optimum. The model-based approach, on the other hand, turns out to be more accurate when sufficient training data is provided. Finally, sensitivity analysis is carried out to investigate the influence of the parameter variations on the manufacturing cost.
Post a Comment