Neural Network for Pulsed Ultrasonic Vibration Control of Electrical Equipment

Abstract

The technique based on contactless pulsed ultrasonic control of electrical equipment's low frequency vibrations is proposed. Experimental laboratory measurements were carried out under conditions when the frequency of ultrasonic probing pulses is comparable to the vibrations frequency of the controlled object's surface (fractions of Hz). In this case, it is proposed to use the simplest artificial neural network (ANN) with back error propagation to estimate the vibrations frequency in the ultrasonic sensing data processing. ANN training was carried out by numerical simulation of ultrasonic signals scattered on the vibrating surface, and then ANN was used to estimate the frequency of vibrations from experimental data. It is shown that at the frequency of ultrasonic sounding in 3-4 pulses for the vibrations period, the use of ANN allows to ensure the accuracy of determining the unsteady vibrations frequency not less than units of percent.