Data Mining Approach for Quality Prediction and Fault Diagnosis of Injection Molding Process

Abstract

Objectives: To implement data mining approach to diagnose the causes of faults occurring in the injection molding product and to predict the quality of product for a particular setting of process parameters. Methods and Statistical Analysis: Decision Tree, k-Nearest Neighbor (k-NN) and Polynomial by Binomial classification techniques are used to build the data mining models by training them on dataset collected during the injection molding of a cap for 25 ml container. Findings: These models are evaluated on test dataset and their prediction accuracy is found to be 95%. Sink marks are caused by low injection speed, nozzle temperature and injection pressure. Low nozzle and mould temperatures and injection pressure resulted in short shot. High barrel temperature at Zone 2 and injection speed are responsible for burn marks in the product. Applications/Improvements: The higher prediction accuracy of these models is helpful in predicting the quality of product before its manufacture and thereby avoiding the production of defective parts. This approach can be further extended for injection molded parts made out of various plastic materials and process conditions.