Neural Classifiers of Vibroacoustic Signals in Implementation on Programmable Devices (FPGA) - Comparison
D. Dąbrowski, Witold Cioch · Acta Physica Polonica A · 2011
The research includes comparative analysis of the effect of the recording of weight vectors and input data of selected neural classifiers with fixed-point numbers.Research has been conducted due to insufficient literature on the influence of such a recording on the correct classification of vibroacoustic signals by neural networks.This current issue was brought up in authors' earlier researches, concerning realization of neural classifiers on programmable logical devices field programmable gate array, with the application of fixed-point processor.During the analysis, three types of neural classifiers were compared in the tests: classifier based on neural networklength vector quantization, classifier using radial neural networks -radial basis functions, and third -counter propagation neural network.The problem stated was to recognize technical state of gear transmission DMA-1 in variable operating conditions.Vectors, based on estimates derived from processed vibroacoustic signals were used as teaching material.