Injection Molding Product Weight: Online Prediction and Control Based on a Nonlinear Principal Component Regression Model Yi Yang, Furong Gao Department of Chemical Engineering, The Hong Kong University of Science and Technology, Clearwater Bay, Kowloon, Hong Kong, People’s Republic of China Weight is an important quality characteristic of injection- molding products. The current work focuses on the on- line prediction and closed-loop control of the product weight. Previous researchers used the process set- points as the inputs to establish weight prediction model. These models cannot reflect the weight varia- tions at a given setting. In this study, an online weight prediction model has been developed, with the process variable trajectories as the inputs, using a principal com- ponent regression (PCR) model. A nonlinear enhance- ment has been made to improve the prediction accuracy of the PCR weight model. Based on such an online prediction, a closed-loop weight control system has been developed and tested experimentally. POLYM. ENG. SCI., 46:540 –548, 2006. © 2006 Society of Plastics Engineers INTRODUCTION Quality is the ultimate goal of injection-molding control, like any manufacturing process. The quality attributes of injection-molded products can be roughly divided into three categories: (i) the dimensional properties such as weight, length, and thickness, (ii) the surface properties represented by the appearance of surface defects, such as sink marks, record grooves, and jetting, and (iii) mechanical or optical properties, such as tensile strength and impact strength. Among these three categories, the product weight is an important quality attribute for the following reasons. First, it has a close relationship to other quality properties, particu- larly those other dimensional properties. For instance, Harry [1, 2] had shown a strong linear correlation between the product length and weight. Second, product weight is a good indication of process stability. A definition of the quality says, “Quality is inversely proportional to variabil- ity” [3]. The performance of the process and quality control can be monitored by the product weight. Furthermore, the weight control is of great commercial interest. Online pre- diction of part weight has been a difficult but promising task, which has been an active research topic in the past decades. The necessity of controlling the product weight in injec- tion molding has been experimentally demonstrated in the early 1990s. Davis and Hudson [4] evaluated the weight variation before and after process interruption. They used a Box-Jenkins model to fit the weight measurements, and found a nonrandom variation during the process. By exam- ining the influence of key process variables, namely injec- tion pressure, packing pressure, cavity pressures, injection stroke, and average cavity pressures, Harry [1, 2] confirmed a close relationship between these process variables and the part weight. The investigation of Schenker [5] suggested that packing pressure, mold temperature, and hydraulic oil temperature were the most significant variables that affect the product weight. Srinivasan et al. [6] proposed a method to control the part weight in thermoplastic injection molding. A linear empir- ical model has been developed to correlate the weight with the set-points of mold temperature, melt temperature, pack- ing time, and packing pressure, and a PI controller is de- signed to control the part weight by adjusting the set-point of packing pressure in a cycle-based fashion. Reasonable control performance had been achieved. However, an online weight measurement mechanism is necessary, which is not feasible in the current molding industry. Kamal et al. [7] developed a weight prediction method based on the PVT relationship of amorphous thermoplas- tics. On the basis of this prediction model, they later [8] developed a closed-loop control system to control the prod- uct weight online with reasonable accuracy. Their approach, however, is limited to amorphous materials. For semicrys- talline materials, because of the complicated crystallization phenomenon experienced in the molding process, it is dif- ficult to control the part weight, using this method. Kim et al. [9] developed a modified factorial design Correspondence to: Furong Gao; e-mail: kefgao@ust.hk Contract grant sponsor: Hong Kong Research Grant Council; Contract grant number: 601104. DOI 10.1002/pen.20522 Published online in Wiley InterScience (www.interscience.wiley. com). © 2006 Society of Plastics Engineers POLYMER ENGINEERING AND SCIENCE—2006