Quality control using a multivariate injection molding sensor

Abstract

Injection molding part quality is modeled using a multivariate sensor. Melt pressure and temperature are respectively obtained through the incorporation of a piezo-ceramic element and infrared thermopile within the sensor head. Melt velocity is derived from the transient response of the melt temperature as the polymer melt flows across the sensor’s lens. The apparent melt viscosity is then derived based on the melt velocity and the time derivative of the increasing melt pressure given the cavity thickness. Quality metrics taken into account are finished part thickness, width, length, weight, and tensile strength. A 12-run, blocked half-fractional design of experiments was performed to derive predictive models for part mass, dimensions, and structural properties. Several predictive part quality models were created using data from the machine, a suite of commercial sensors, the multivariate sensor, and combinations thereof. The results indicate that multiple orthogonal streams of process data yield higher-fidelity models with coefficients of determination approaching one. Furthermore, best subset analysis indicates that the most important process data are gathered from in-mold sensors, where the acquired information is closest to the states of the polymer forming the final product.