engleski

An Evolution and Biology Based Approach to the optimization of the Extrusion Process

A new approach to the optimization of extrusion process has been performed with Neural Network (ANN) and Genetic Algorithm (GA) based procedure. ANN trained on the technology process track record should be able to describe complex dependencies between influence parameters and product outgoing characteristics and is also supposed to be the fitness function. The optimization of the technological process based on genetic algorithm is proposed. The influence parameters are treated as heritage and combined with ANN forward pass being the fitness function. The ANN is trained presenting extrusion experiment data-patterns. The objective (fitness) function value is calculated with ANN forward pass. The optimization results are compared with a more conservative approach using experiment planing and regression analysis. The mean error of the difference between regression and ANN approach is used as a comparison measure. The influence of different training parameters and choice of training patterns are investigated. Optimization is performed on the extrusion of thick-walled AlZnMg (7000) hollow sections. Optimization variables are: billet temperature; extrusion rate; and Zr content. The output is tensile strength of the longitudinal welds. The advantages of the method presented are its robustness and flexibility.

optimization;extrusion process; neural network; genetic algorithm

nije evidentirano