Towards an accurate pressure estimation in injection molding simulation using surrogate modeling

Abstract

The computational cost of high-fidelity injection molding simulations has been growing in the past years making it more and more challenging to use them for performing analyses such as optimizations or uncertainty quantification. Surrogate modeling offers a cheaper way to realize such studies and has been gaining attention in the field of injection molding simulation. In this work, we propose to compare three surrogate modeling techniques along with two design of experiment methods in their ability to predict the pressure signal at a surface node in a Moldflow simulation by varying process and modeling parameters. A Sobol sensitivity analysis is performed to study the contribution of the varied parameters on the pressure results. In addition, one of the generated models is used along with experimental pressure sensor data to improve the pressure estimation by calibrating the heat transfer coefficients during filling and packing as well as the pressure-dependency coefficient in the Cross-WLF viscosity model. This resulted in major improvements of the pressure predictions for all 27 considered cases in comparison to using the default heat transfer coefficients and viscosity model parameter.

Efficient identification of a flow-induced crystallization model for injection molding simulation

“…In fact, we propose that the generation of a surrogate model of a high-fidelity injection molding simulation model (where the surrogate input variables are the unknown material parameters in the high-fidelity simulation model) and the subsequent identification by calibrating the generated surrogate model is more efficient than the traditional approaches based exclusively on lab experiments. In a previous work dealing with default injection molding simulation [3], this methodology performed well for a low number of training data and input parameters.…”

“…Once the results of the high-fidelity simulation DoE are available, the generation of the surrogate models is done in MATLAB R2019b using proper orthogonal decomposition (POD) of the pressure signal and non-linear regression (NLR) of the POD basis coefficients. This methodology is partially analogous to the POD-NLR method presented in [3]. The following steps describe the process of generating the surrogate models using the POD-NLR technique used in this work:…”

“…132 simulations were chosen for training and 20 simulations for testing purposes. This choice is postulated to be appropriate as five parameters are being varied in this work in comparison to the six parameters changed in our previous work [3], where an acceptable accuracy for the estimation of the pressure signal was reached using 120 training simulations.…”

A methodology for determination the inlet velocity in injection molding simulations

“…Initially, it is relevant to note that most published works using Moldflow ® ignore the influence of mesh size. For example, Solanki et al [33] and Saad et al [34] recently presented studies addressing design and optimization of molds, respectively, using Moldflow ® . Despite the importance of mesh size, mainly in optimization problems, the authors of both works did not discuss the effects of mesh refinement.…”