Zastosowanie sieci neuronowych do sterowania procesem spawania elektronowego
Application of neural networks for electron beam welding
II Kongres Towarzystwa Próżniowego ; 13-17.05.2001 ; Warszawa, Polska
The measurements for the three largest groups of charged particles: backscattered electrons, really secondary electrons and ions were carried out. It was estimated that only the signals of direct component amplitude of these particles currents can serve as the input of neural networks. The computer simulations of various models of neural networks were done. The best result was obtained for network, which on the basis of amplitude signals measured with unfocused EB determines previously defined value of the focus current under which the weld will be made. It creates possibilities to weld outside of the forbidden zones and to obtain the compromise between depth of the weld and the lack of defects.
Bibliogr. 3 poz., wykr.