Abstract Determining optimal filling conditions in injection molding is essential for ensuring part quality and process stability. Despite the variety of existing approaches, this task remains difficult when the available process information is limited or difficult to obtain. Alternatively, this work presents a simple data-driven optimization strategy based on the combination of multivariable regression (MVR) and a genetic algorithm (GA). This scheme is implemented to predict processing conditions leading to complete cavity filling using only a reduced set of experimental data obtained under short-shot conditions. The case studied is a flat-plate geometry having a hot weld line in the mid plane. Specifically, second-order MVR models were trained to quantify the influence of the most relevant processing variables during the filling stage on two complementary filling indicators: part weight and projected area. Once fitted, the models were integrated into a GA-based inverse optimization scheme to search for parameter combinations that achieve complete filling targets. The parameter combinations predicted by the MVR-GA optimization scheme were experimentally validated, resulting in fully filled parts, with average deviations of approximately 2 % and 0.6 % for weight and projected area, respectively, relative to the nominal target values. The proposed framework allowed the data-efficient exploration of different practical scenarios, where modifying one or more processing parameters requires the corresponding adjustment of the remaining ones while still achieving complete filling.
Quintana et al. (2026) studied this question.