APPLICATION OF ARTIFICIAL NEURON NETWORKS FOR PREDICTION OF LASER TREATMENT OPERATION QUALITATIVE RESULTS

Abstract

Purpose. Attempt of the use of artificial neuron network for a prognosis and management precision of laser hardening treatment operation. Methodology approach. For the decision of task the three-layered network with signal direct distribution, the syntax of which looks like in the environment of Matlab is used: Net=netff(minmax (X) [P, S, Q]{ logsig, logsig, logsig }. Findings. There are results of artificial neuron networks application for prediction of laser surface treatment operation qualitative dates, in particular, to precision of size its results. For the decision of task the three-layered network with signal direct distribution, the syntax of which looks like in the environment of Matlab is used: Net=netff(minmax (X) [P, S, Q]{ logsig, logsig, logsig }. The select network paradigm features are: a variety of synapses is after a dimension and action orientation; possibility of synapses groups forming after their influence objects and dependence of synapses level and them weights from the regime operation realization parameters precision of which is forecast. For three-layered persеptron with direct distribution of signals the system of mutually independent entrances and proper synapse connections is developed, that allowed to linearize model of operation size indexes’ precision under different charts of its organization. Findings. The possible level of prognoses’ error was although got, it was succeeded to use training of network with the use of reverse distribution of signal for realization of it adjusting by functional influence on the synapse coefficients. Originality. However, the found out the structure of operation error allows purposefully to use not regime methods and facilities for increasing of treatment precision or add to the optimization model of process the additional functional of misclosures for the search of the optimum modes of treatment.