Robustness-aware multi-objective optimization of injection molding processing conditions and cooling channel geometry

The optimization of injection molding is strongly affected by variations in operating conditions, which influence part quality, cycle time, and energy consumption. Traditional approaches focus on nominal conditions and neglect unavoidable process fluctuations, whereas robust optimization seeks solutions that remain stable under variability. However, robustness is usually imposed through a single predefined formulation, even though different formulations may yield different optimization outcomes. This work investigates how the choice of robustness formulation affects multi-objective optimization in injection molding, using a two-ring case study modeled in Moldex3D with artificial neural network surrogates and a multi-objective evolutionary algorithm. In a first stage, the operating conditions (melt and mold temperatures) are optimized by comparing variance-based, average, worst-case, and expectation-based formulations through their Pareto fronts and solution rankings. In a second stage, the analysis is extended to conformal cooling channel geometries (two-ring, three-ring, spiral, and zig-zag), assessing how robustness against operating-condition variability influences geometry selection. The formulation reshapes the Pareto fronts and the solution rankings at both levels, but changes the selected solution unevenly: the recommended operating point stays essentially fixed at the lower boundary of the operating window, whereas the preferred cooling-channel geometry and the cost of imposing robustness depend on the formulation, because near-equivalent designs differ widely in geometry. Robustness is therefore not a neutral metric and must be defined according to the application and to the level, process parameters, or cooling-channel design, at which the engineering decision is made. The reported insensitivity of the recommended operating point to the robustness formulation is specific to the present material (PP, Borealis Borcom BG055AI) and operating window, whereas the general result is the asymmetry between the process-parameter and geometry levels.