Data Mining Approach for Quality Prediction and Control of Injection Molding Process

Objectives: This paper proposes data mining approach to detect the causes of sink marks, short shot, burn marks and flash in 25 ml container cap and control the injection molding process to produce defect free product by optimal setting of process parameters. Methods and Statistical Analysis: Neural Net, Naive Bayes classifier and rule induction techniques are implemented to build the data mining models on training dataset acquired during the injection molding of 25 ml container cap. Findings: Neural Net and Rule Induction Models outperformed over Naive Bayes model (80%) with prediction accuracies of 95% on test dataset. Rule Induction model detected that sink marks are caused by high molding temperature, low injection speed, nozzle temperature and injection time. Low injection pressure, barrel temperature and mould temperature are responsible for short shot to occur in the product. Flash is caused by high mould temperature. Applications/Improvements: The custom prediction for quality of the product for a specific parameter setting can be made by applying Neural Net and Rule Induction models in order to control the process to ensure defect free caps. This approach can be applied for other injection molded products manufactured from other plastic materials.