Optimal Deformation of a Small Plastic Light-guide Using Machine Learning Algorithms

Abstract

Lensknob is a component that transmits light to users. It is essential to minimize the deformation to transmit the light uniformly. As a method of finding injection molding parameters capable of minimizing the deformation, the amount of deformation of the Lensknob was predicted in advance by numerical analysis of the injection molding. However, because it takes a considerable amount of time to analyze, we used the Decision tree as a Machine Learning model. As the injection molding parameters, we set the melting temperature, cooling time, holding time, holding pressure, and ram speed. We set the injection molding parameters based on the range recommended by Moldflow. A full factor method of factor 5 level 3 was applied in the experiment. We predicted the parameters for minimizing the deformation through the Decision tree learned with 243 experimental data. We set the criteria to evaluate the performance of the Decision tree. The parameters predicted by the Decision tree improved the deformation by about 10.37%.