Quality Prediction of Injection-Molded Parts Based on PLS-ANN

Abstract

A feed-forward three-layer artificial neural network (ANN) combined with Partial Least-Squares (PLS) was presented to predict the part weight of injection-molded products. Firstly, melt temperature, holding pressure and holding time which are the most important influenced factors of injection-molded parts quality were chosen as independent variables and part weight were chosen as dependent variable. Secondly, PLS was used to analysis the relationship among these variables and calculate the aggregate elements of independent variables and dependent variable. Here, dependent variable was single, so parts weight is the aggregate element of dependent variable. Thirdly, the principal elements of independent variables and dependent variable were used to construct an ANN. At last, the performance of PLS-ANN model was evaluated and tested by its application to verification tests. Results showed that the PLS-ANN predictions yield mean absolute percentage error (MAPE) in the range of 0.06% and the maximum relative error (MRE) in the range of 0.15% for the test data set, which can accurately reflect the influence of the injection process parameters on parts quality index under the circumstance of data deficiencies.