Open AccessApplication of artificial neural networks digital filtering in measurements
Author(s) -
O. A. Galkovskii
Publication year - 2020
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/734/1/012082
Subject(s) - component (thermodynamics) , artificial neural network , noise (video) , computer science , filter (signal processing) , signal (programming language) , sensitivity (control systems) , dimension (graph theory) , software , function (biology) , artificial intelligence , activation function , electronic engineering , engineering , mathematics , computer vision , physics , evolutionary biology , biology , pure mathematics , image (mathematics) , thermodynamics , programming language
The problem of this paper is suppression of the noise component of the useful signal in the researching of factors affecting the coefficient of friction of a new type of machine polymer guide rail. The parasitic component occurs during measurements, due to the vibration sensitivity of the sensor and the specifics of measurements. To suppress this component, a neural network program filter modelling the noise function has been developed. Filtering of the noise component of the signal occurs in real time, so the requirements of the software to the system resources are the determining factor, with the same accuracy characteristics of artificial neural networks. Thus, the topology of radial basis functions is optimal for the problem of modelling the physical noise function. The software implementation of the neural network filter is made in the “G” language of the LabView development environment. Approaching the number of neurons of the hidden layer to the dimension of the training sample increases the accuracy of the function simulation, but at the same time causes retraining. The accuracy of the simulation of the function of the proposed system was 0.142%.