The injection molding process is versatile for fabricating thermoplastic products for various application purposes. There has been an increased demand for most of these manufactured products recently, and the expectation tends to increase with time. S...
The injection molding process is versatile for fabricating thermoplastic products for various application purposes. There has been an increased demand for most of these manufactured products recently, and the expectation tends to increase with time. Significant attention has been drawn to the demolding stage because it’s the final step in the injection molding process where most of these products are manufactured. The reason is that the demolding stage produces the demolding force, which is quantified as a factor contributing to the quality of the final molded parts. However, there has been an effort in establishing prediction models and optimizing the process parameters for the demolding stage to achieve good quality products through the demolding force. Still, most of the established approaches are based on traditional techniques, which lacks the robustness for prediction and optimizing the process parameters in terms of their advantage. These concerns have motivated the investigation to experimentally measure the demolding force in the injection molding process and develop some robust models through artificial intelligence to predict the demolding force efficiently. To address this issue, a design setup was proposed that creates a modification to the supporting ejector rod behind the demolding plate to accommodate the load cell in the injection molding machine for measuring the demolding force. In addition to the design setup is the data logger system, which receives the signal from the load cell. The collected dataset was evaluated, analyzed (root mean square error and coefficient of determination), and used for the model development. The model development and optimal parameter selection were established in three forms using the process parameters: melt temperature, mold temperature, filling time, and packing pressure. First, the artificial neural network was proposed for predicting the demolding force for a thermoplastic polymer material using the process parameters for three different algorithms, namely, Levenberg-Marquardt, Broyden-Fletcher-Goldfarb-Shanno quasi-Newton, and scale conjugate gradient. Secondly, the long short-term memory was also proposed to forecast the demolding force in real time in the injection molding process. Three optimization algorithms were applied: stochastic gradient descent with momentum, root mean square propagation, and adaptive momentum estimation. The developed models, through their respective optimized algorithms, have demonstrated their significant robustness in forecasting the demolding force, which can be verified by the agreement with the experimental dataset. Thirdly, a novel intelligent system for optimal parameter selection, called VRML-TOPSIS, has been proposed. The system platform is a combination of variogram model, machine learning model and TOPSIS, which has demonstrated a unified decision criteria for predicting the demolding force by selecting the optimal process parameters. In summary, this dissertation developed for the first time artificial intelligence models for the demolding force prediction and optimization in the injection molding process. The developed models have demonstrated their robustness in predicting the demolding force and will significantly help improve the properties of polymer material for comparable designed products.