Особливості технічної реалізації штучного нейрона для розв’язання задач контролю і діагностики технічного стану авіаційного двигуна ТВ3-117

Abstract

Purpose. The purpose of this work is the mathematical modeling of a neuron and the possibility of its technical implementation, since an artificial neuron is the basic link of any neural network, in particular, a neural network, which is used to control and diagnose the technical state of TV3-117 aircraft engine in flight modes. A way is proposed based on bit-parallel processing of numerical data (synapses) in order to obtain the accuracy of the output signal of the neural network. Methodology. The work is based on neuroinformatics methods were applied to develop and training a neural network for implementing a linear on-board model of aircraft engine TV3-117. Results. According to the results of approbation of the developed mathematical model of the artificial neuron, it is determined that the considered approach will allow to increase the speed of the artificial neuron by two orders of magnitude in comparison with the existing models. Therefore, the proposed device for the implementation of the neuron can serve as a new elemental basis for building neurocomputers with higher speed of biological information processing and good survivability, in particular, developed on the basis of 64-bit mini-computer Raspberry Pi NanoPi M1 Plus with quad-core Allwinner H3 processor Allwinner Technology Co., Ltd. Originality. The scientific novelty of the results obtained lies in the fact that for the first time a way was proposed based on bit-parallel processing of numerical data (synapses) in order to obtain the accuracy of the output signal of the neural network. Practical value. The proposed device for implementing a neuron can serve as a new element base for building neurocomputers with a higher speed of processing biological information, developed on the basis of a 64-bit Raspberry Pi NanoPi M1 Plus mini-computer with a quad-core Allwinner H3 processor. It was determined that if it is necessary to increase the accuracy of algorithms and complicate their structure, it can lead to an increase in the number of artificial neurons, and, accordingly, to an increase in the number of computational operations performed. In this case, it is advisable to implement the proposed algorithms on the basis of faster computing devices compared to a standard microcontroller, which can be implemented on the basis of digital signal processors, programmable logic integrated circuits or specialized processors. Keywords: aircraft engine, neural network, artificial neuron, bit cut