Quality Prediction for Injection Molding by Using a Multilayer Perceptron Neural Network.

Area of Science:

• Manufacturing Engineering
• Materials Science
• Artificial Intelligence

Background:

• Injection molding is crucial for high-precision product mass production, demanding stringent quality control.
• Current quality assessment relies on machine parameters or costly mass inspections, which are often inaccurate or inefficient.
• Accurate, real-time quality judgment for each part is essential but challenging to achieve with existing methods.

Purpose of the Study:

• To develop a fast and automatic quality inspection system for injection-molded parts.
• To enable accurate prediction of finished product geometry using in-mold sensor data.
• To overcome the limitations of machine parameters and costly traditional inspection methods.

Main Methods:

• Utilized a multilayer perceptron (MLP) neural network model.
• Integrated quality indices derived from in-mold pressure sensor data.
• Extracted key indices such as first-stage holding pressure, pressure integral, residual pressure drop, and peak pressure.
• Trained and tested the MLP model using these quality indices to predict geometric widths.

Main Results:

• The MLP model achieved high accuracy (exceeding 92%) in predicting geometric widths.
• Key quality indices extracted from pressure curves demonstrated a strong correlation with part quality.
• The proposed method effectively reflects melt flow state and molding quality.

Conclusions:

• The developed MLP model combined with quality indices provides a feasible solution for fast and automatic quality inspection in injection molding.
• This approach enhances the accuracy and efficiency of quality judgment for injection-molded parts.
• The method offers a cost-effective alternative to traditional mass inspection techniques.