Optimized injection-molding process for thin-walled polypropylene part using genetic programming and interior point solver

In this paper, the injection molding technique is selected, and different internal and external defects have been evaluated, including warpage, short shot, and shrinkage. Different geometric and injection machine inputs such as gate type, filling time, part cooling time, holding pressure time, and melt temperature have been chosen, respectively. The Taguchi method is applied to find the best level of each parameter. In recent years, researchers have been focused on using machine learning techniques with a combination of meta-heuristic methods to calculate the optimal process parameters in injection molding to increase the quality of the final product. However, the computational load of the machine learning methods and meta-heuristic algorithms is far behind in handling real-time applications. The main motivation of this study is to reach an accurate model with a low computational load to handle the real-time computational load even in the presence of lower CPU power controller mechanisms. Then, the genetic programming method is employed to extract the optimal mathematical model of the injection-molding process, which relates the processing parameters, including part cooling time, filling time, melt temperature, and holding pressure time, to output which is the combination of shrinkage rate, short shot, and warpage. The extracted optimal mathematical formulation of the genetic programming method is employed inside the interior point nonlinear programming solver via the fmincon function of MATLAB software to calculate the optimal parameters of the process as fast as possible. The genetic programming results are compared with previous methods such as decision tree, support vector regression, and multilayer perceptron to prove the acceptable accuracy of the first part of the paper with the lower computational load. Then, the means square error between the finite element method and the extracted result using the hybrid genetic programming and interior point nonlinear programming solver is 47.06%, 93.75%, and 83.63% lower than previous methods, including decision tree, support vector regression, and multilayer perceptron, respectively.