Multiobjective optimization of injection molding using a calibrated predictor based on physical and simulated data

Abstract

This paper presents a method for improving injection molding processes having competing performance measures using multiobjective optimization. The procedure uses calibrated predictors that combine physical and simulated data to estimate the values of the performance measures. After the predictors are built, the values of the selected performance measures are estimated at a grid of process control variables, and a set of predicted Pareto solutions is identified using nondominance criteria. A refinement of the original Pareto solutions is obtained by predicting the performance measures at a finer grid of the process variables near the original Pareto set. Finally, as validation, a subset of these solutions is evaluated on the physical process. A case study with three performance measures is presented to show how the calibrated predictors allow the injection molding manufacturer to identify the processing conditions that optimize a process having competing objectives. POLYM. ENG. SCI., 57:248–257, 2017. © 2016 Society of Plastics Engineers

Simulation and multi-objective optimization to improve the final shape and process efficiency of a laser-based material accumulation process

“…Thus, it is not the best approach to obtain a single solution but rather the set of solutions corresponding to the best compromises. For this, we use the following definition for the concepts of non-dominated solutions and Pareto front as in [10] and [24]:…”

“…The corresponding input and output values are shown on Table 3 (runs 25-33). Afterwards, the incumbent Pareto front was updated and the new Pareto solutions are 1,4,7,8,12,14,15,16,18,19,21,22,23,24,[27][28][29][30][31][32][33]. Then, the stopping criteria were evaluated and since the R 2 of both models are larger than 0.98 the method stopped and the final Pareto solutions are reported.…”

“…The combination of optimization methods with computer simulations where simulations are used to transform input parameters (controllable processing variables, or CPVs) into the relevant performance measures (PMs) is an actual engineering need [19,24,27,29]. This requires evaluating optimization functionals on a large amount of candidate solutions, making it a demanding computational task when the simulations are numerically expensive [16].…”

“…Case Study 1: Evaluation of selected predicted Pareto solutions and derived quantities, iterations k = 1 (runs[16][17][18][19][20][21][22][23][24] and k = 2 (runs[25][26][27][28][29][30][31][32][33] …”

Challenges and Opportunities in Simulations and Computer Experiments in Industrial Statistics: An Industry 4.0 Perspective

“…A combination of physical and simulated experiments to finetune predictions for the manufacturing process injection moulding process outputs is used in ref. [78] The authors design both physical (full factorial with replications) and CEs (a maximin LHD). The predictors they suggest are based on Gaussian process models.…”