Fast approximation of fiber reinforced injection molding process using eikonal equations and machine learning
Topic(s) : Manufacturing
Co-authors :
Julian GREIF (GERMANY), Nils MEYER (GERMANY)Abstract :
Injection molding of thermoplastics is a widely employed method for manufacturing intricate geometries, allowing the incorporation of discontinuous glass or carbon fibers to enhance the mechanical properties of the resulting part. However, the mechanical characteristics of the fiber reinforced composite depend significantly on the local fiber orientation, making this a critical factor for the molded part.[1]
While commercial tools for simulating injection molding, often incorporating macroscopic fiber orientation models, are established, their computational demands lead to simulations lasting up to several hours. This complexity hinders seamless integration into the design process. Current research primarily concentrates on refining simulations for a more nuanced understanding of the fiber orientation process [2], fiber migration process [3], and flow-fiber coupling [4].
In this study, we propose a rapid approximation of the mold filling process through a combination of easily accessible nodal features and machine learning models. The approach utilizes the eikonal equation to quickly calculate distances from nodes to the inlet and cavity walls. Additionally, a structure tensor is applied to estimate local material distribution. Following machine learning model training, this enables the estimation of features typically derived from expensive mold filling simulations-such as fill time, freeze time, and fiber orientation-at a fraction of the computational cost of a full simulation. These advancements facilitate the direct integration of fiber orientation information into the design process. Moreover, these models can be extended to include warp analysis.
While commercial tools for simulating injection molding, often incorporating macroscopic fiber orientation models, are established, their computational demands lead to simulations lasting up to several hours. This complexity hinders seamless integration into the design process. Current research primarily concentrates on refining simulations for a more nuanced understanding of the fiber orientation process [2], fiber migration process [3], and flow-fiber coupling [4].
In this study, we propose a rapid approximation of the mold filling process through a combination of easily accessible nodal features and machine learning models. The approach utilizes the eikonal equation to quickly calculate distances from nodes to the inlet and cavity walls. Additionally, a structure tensor is applied to estimate local material distribution. Following machine learning model training, this enables the estimation of features typically derived from expensive mold filling simulations-such as fill time, freeze time, and fiber orientation-at a fraction of the computational cost of a full simulation. These advancements facilitate the direct integration of fiber orientation information into the design process. Moreover, these models can be extended to include warp analysis.