A Simulation-Data-Based Machine Learning Model for Predicting Basic Parameter Settings of the Plasticizing Process in...

Area of Science:

• Polymer processing
• Machine learning applications
• Injection molding optimization

Background:

• Traditional polymer processing relies on time-consuming and costly trial-and-error for optimal machine settings.
• Accurate prediction of machine parameters is crucial for efficient and high-quality plasticizing.

Purpose of the Study:

• To develop a simulation-driven machine learning workflow for determining optimal injection molding machine settings.
• To predict essential parameters like back pressure and screw rotational speed for desired plasticizing times and melt quality.

Main Methods:

• A machine learning model was trained using pre-processed data sets, incorporating material properties, screw geometry, and shot weight.
• Supervised machine learning algorithms were compared, with a neural network identified as the most effective approach.
• The model's predictions were validated through experiments on a real injection molding machine using diverse materials.

Main Results:

• The trained neural network model demonstrated excellent generalization on unseen data, achieving an overall absolute mean error of 0.27% and a standard deviation of 0.37%.
• Experimental validation confirmed that the predicted machine settings resulted in minimal deviations between real and desired plasticizing times.
• All tested operating points yielded good melt quality, validating the model's predictive accuracy.

Conclusions:

• The developed workflow effectively predicts suitable initial operating points for injection molding machines, significantly reducing planning costs.
• This approach enhances understanding of factors influencing melt quality in complex plasticizing processes.
• The simulation-driven machine learning model offers a powerful tool for optimizing polymer processing and improving operational efficiency.