A predictive modelling strategy for warpage and shrinkage defects in plastic injection molding using fuzzy logic and pattern search optimization

Steven O. Otieno, Job M. Wambua, Fredrick M. Mwema*, Edwell T. Mharakurwa, Tien Chien Jen, Esther T. Akinlabi

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

Quality control through defect minimization has been the central theme in plastic injection molding research. This study contributes to this course through the introduction of an alternative predictive modelling strategy for injection molding defects. Through multi-stage design of experiments, Computer Aided Engineering simulations, and intelligent algorithms, the study developed a warpage and shrinkage defects predictive model based on processing parameters. In the factorial design of experiment stage, the mains effect sizes, interaction effect sizes, and ANOVA were used for process parameter screening. Next, a Taguchi L25 design was used for the generation of predictive model training data. Fuzzy logic models were then developed to predict warpage and shrinkage defects based on given process parameters and the predictive capability of triangular and Gaussian membership functions was investigated. A pattern search algorithm was utilized to tune the developed predictive models. The resulting predictive model had root mean square error (RMSE) of 0.04, standard error of regression (S) of 9.6, and coefficient of determination (R2) of 98.7% for shrinkage prediction. The respective model metrics for warpage prediction were 0.005, 1.2, and 96.3%. The triangular membership function model had lower RMSE indicating a higher predictive accuracy whereas the Gaussian membership function model had lower S indicating a higher model reliability. Tuning of the predictive models using a pattern search algorithm reduced the RMSE and S and increased the models’ R2. The approach can be adopted by plastic processing industries to predict and control such (and related) defects for quality products and maximum productivity.

Original languageEnglish
Number of pages25
JournalJournal of Intelligent Manufacturing
Early online date9 Mar 2024
DOIs
Publication statusE-pub ahead of print - 9 Mar 2024

Cite this