Predictive Methodology for Quality Assessment in Injection Molding Comparing Linear Regression and Neural Networks

Area of Science:

• Materials Science and Engineering
• Manufacturing Process Optimization
• Computational Modeling

Background:

• Increasing use of recycled polypropylene (PP) necessitates efficient design for manufacturing.
• Traditional trial-and-error methods for plastic part design are time-consuming and inefficient.
• Simulation and process modeling are crucial for developing plastic parts with recycled materials.

Purpose of the Study:

• To develop and compare prediction models for optimizing plastic part design using recycled PP.
• To evaluate the precision and correlation of linear regression and artificial neural network (ANN) models.
• To analyze the predictability of nonlinear behaviors and compensatory effects in injection molding processes.

Main Methods:

• Combined simulation with design of experiments (DOE) to create prediction models.
• Utilized linear regression and artificial neural network (ANN) fitting, specifically Back Propagation Neural Networks (BPNN).
• Input variables included eight injection parameters and geometry variations; output features covered seven process and part quality metrics.

Main Results:

• Back Propagation Neural Networks (BPNN) demonstrated suitability for correlating all quality features into a single predictive formula.
• The developed prediction models significantly accelerated the optimization of part design for multi-criteria decision-making.
• Application to an induction hob support design achieved a feasible 27% weight reduction.

Conclusions:

• The proposed simulation-driven methodology enhances the optimization of recycled plastic part design.
• BPNN models offer superior correlation capabilities for complex injection molding process parameters and quality features.
• Achieving significant weight reduction requires combining non-standard process parameters with non-uniform thickness distribution.