Inverse Modeling and Optimization of the Processing Parameters of Gas Assist Injection Molding Process by Neural Networks
| Autor: | Zi-San Sam, 沈子珅 |
|---|---|
| Rok vydání: | 2000 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 88 The effort of this thesis is to derive the optimal processing parameters of the Gas-Assisted Injection Molding process (GAIM) by the use of an artificial neural network (ANN). This approach is proposed because the effects of processing conditions, such as the applied pressure, plug material, its geometry and extruding velocity, and the local temperature, are highly nonlinear and fully coupled. The complexity of the process makes the operation design a difficult task, even with the existence of a reliable physical model. Usually, prolonged tests must be conducted. The proposed method can benefit many of the similar processes during the setup stage. Data from tests carried out on GAIM are used to train an artificial neural network serves as an inverse model of the process. The inverse model has the desired product parameters as inputs and the corresponding processing parameters as outputs. It can provide the processing parameters needed to yield products with the desired dimensions. The structure, together with the training methods of the ANN, is investigated. As the size of the hidden layer of an ANN dramatically affects its performance, this makes the architecture design an important issue. In this thesis, we use the genetic algorithms (GAs) to train the connection weights, as well as node number of the hidden layer of a neural network. A modification of fitness function in GAs is modified to take into account the simplicity of the structure of ANN. The approach based on GAs is presumed to perform a global search of the weight space and should be less likely to become stuck in a local minimum than back propagation. A disadvantage of probable slow training rate compared to back propagation is acceptable since the optimization is not conducted on-line. The feasibility of the proposed approach is also investigated for similar processes. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|