The Density Control Of The MIM Injection Molded Billet Based On The Particle Swarm Optimization And Neural Networks

Abstract:The metal injection molding process as an efficient powder metallurgy technology for parts near net shape forming has been widely applied; process control methods need to be studied in order to eliminate product defects and improve product quality. The density distribution of the green body produced by the stage of injection molding is an important indicator, which affects the quality of the finished product.This paper, firstly, summarizes the development and application of neural networks and the particle swarm optimization. Through the analysis of The influence of the metal injection molding process parameters on the quality of the injection molding green body, six major process parameters which are controllable and the density distribution of the green body produced by the stage of injection molding are chosen for the study. The orthogonal table is used to select research data. The fluid simulation module CFX of the finite element software ANSYS is used to simulate the injection molding process of the Pentagon mould to provide data for the training samples of the neural networks model.Secondly, through training the neural networks by the sample data, this paper establishes the density distribution prediction model based on the BP neural networks which can reflect the complex nonlinear relationship between the six inputs-injection molding process parameters and four outputs-the density distribution of the green body. The generalization ability of the model is strong by the verification of the testing data. The relative error of the prediction density and the volume fraction of the density distribution are within0.45%and5%.Eventually, the control problem of the density distribution in the stage of injection molding is transformed to a constrained nonlinear optimization problem through combining the particle swarm optimization with the density distribution prediction model based on the BP neural networks.Improved particle swarm optimization is chosen to solve the problem. The feasibility of the algorithm is tested by the basic multimodal function which is used to test the performance of the optimization algorithm. Through the verification of the control results by the ANSYS numerical simulation, the relative error of the density and the volume fraction of the density distribution control in the injection molding don’t exceed0.331%and3.5%. A global optimization method based on the interior point method is used to compare with the particle swarm optimization. This paper explores an effective way for the control of the green body’s density distribution in the metal injection molding.19figures,17tables,76reference documentations...