Система безперервного вібромоніторингу стану технологічного обладнання з машинним навчанням класифікатору

Abstract

Vibration monitoring of technological equipment is characterized by the presence of unsteady complex vibration signals, which are characterized by the presence of time dependences of the amplitude, frequency, phase. In classical machine learning, data already obtained is usually randomly divided into training and test sets. Based on the training data set, a classifier is obtained, and with the help of a test one, the accuracy of this obtained data classifier model is checked. The developed software package solves the problem of identifying diagnostic vibration signals by selecting the time series of the test signal with minimal proximity based on wavelet coefficients. Selection in the test set is carried out in the control process at the minimum value of the coefficient of autocoherence, which is close to zero. Thus, the data classifier works continuously, enriching the mathematical model with recognition of emerging equipment defects, depending on a large number of random factors, installation conditions for the qualifications of staff, etc. A fundamentally new approach to organizing the identification of a vibration signal is that it is not the vibration monitoring signals recorded over a certain period of time that are identified, but the identification takes place in real time. This makes monitoring the state of technological equipment operational. The developed analysis algorithm makes it possible to implement a system of continuous vibration monitoring of technological equipment, to increase the accuracy of identification of diagnostic vibration signals through the use of an integrated approach to the analysis of the proximity of a test signal.