NEURAL NETWORK DIAGNOSTICS OF AIRCRAFT PARTS BASED ON THE RESULTS OF OPERATIONAL PROCESSES

Abstract

Context. The problem of synthesis of an optimal neural network model for diagnostics of aircraft parts after operational processes is considered. The object of the study is the process of synthesis of neural network diagnostic models for aircraft parts based on the results of operational processes Objective is to synthesize neural network diagnostic models of aircraft parts after operational processes with a high level of accuracy. Method. It is proposed to research the use of two approaches to the synthesis of neural network diagnostic models. So, using a system of indicators, the topology of the neural network is calculated, which will be trained using the method of Backpropagation method in the future. The second approach is based on the use of a neuroevolutionary approach, which allows for a complete synthesis of the neural network, dynamically modifying the topology of the solution in addition to the parameters. the final decisions are compared in the accuracy of work on the training and test data set. This approach will allow to determine the possibility and correctness of using neuroevolutionary methods for the synthesis of diagnostic models. Results. Neuromodels for diagnostics of aircraft parts based on the results of operational processes have been obtained. The obtained results of comparing the methods used for synthesis made it possible to form recommendations for the implementation of neuroevolutionary methods in the synthesis of diagnostic neuromodels. Conclusions. The results obtained during the experiments confirmed the operability of the mathematical software used and allowed us to form recommendations for further use of the considered methods in practice in order to synthesize diagnostic neuromodels. The prospects for further research may consist in expanding the input data sets in order to synthesize and study more complex topologies of neural network models.