engleski

Optimization of aluminum extrusion and die design using neural networks and genetic algorithms

The approach to the optimization of aluminum extrusion process and die design, based on artificial neural network and genetic algorithms, is presented. The artificial neural network is trained on extrusion experiments data to be able to describe complex dependencies between extrusion influence parameters and section properties. Therefore, it can be applied in analysis of different problems in extrusion practice. Two examples in extrusion of hollow sections by means of hollow die were analyzed by using this procedure. The length of the charge welds as a function of billet temperature, extrusion ration and the height of welding chamber were analyzed in the first example. Experiments were performed durin extrusion of tubes of 1000 aluminum alloy by means if bridge die in laboratory conditions. The second example was connected to extrusion of thick walled hollow section of AlZnMg4.5 aluminum alloy by means of porthole die in real conditions. The longitudinal welds strength was analyzed as the function of billet temperature, extrusion rate, and Zr content. In both examples composite plans of experiment were used. Trained artificial neural network forward pass was implanted in genetic algorithms were researched out and presented in the paper. The results of the performed optimization were compared with methods that are more conservative. Proposed approach is proved as efficient, robust and very reliable.

extrusion; aluminum; genetic algorithms; artificial neural network; optimization

nije evidentirano