DIAGNOSIS METHODS FOR MECHANISMS AND MACHINES BASED ON EMPIRICAL MODE DECOMPOSITION OF A VIBROSIGNAL AND THE WILCOXON TEST

Abstract

Methods for diagnosing mechanisms and machines based on the analysis of vibration signals are considered. In particular, the comparison of various algorithms for analyzing vibration signals in the time and frequency domains was made, methods for selecting diagnostic features and methods for secondary processing were analyzed. The purpose of the study is to develop algorithms for selecting the vibration signal envelope based on empirical mode decomposition and decomposition of the signal into intrinsic mode functions, algorithms for the spectral estimation of envelopes and to choose a criterion for making a decision on object classification. It is proposed to choose the non-parametric Wilcoxon signed-rank test to determine the statistical significance of the difference between the parameters of normal and faulty objects. The multichannel microcontroller system for collecting data from an accelerometer and transmitting it to a computer via a local Wi-Fi network, including a number of independent data gathering nodes connected to a common distributed computing system, has been developed and experimentally studied. The computer processing of the recorded vibration signals for serviceable and faulty mechanisms was performed, including data decoding, Hilbert-Huang transform, spectral analysis using the Welch and Yule-Walker methods, and the choice of a diagnostic feature that provides maximum reliability of recognition. Based on the results of the work, it was determined that the empirical mode decomposition makes it possible to obtain vibration signal envelopes suitable for further diagnostics. Recommendations are developed for choosing the intrinsic mode function and the spectral analysis algorithm, it is determined that the first intrinsic mode function is the most informative for the mechanism under study. In accordance with the Wilcoxon criterion, the degree of diagnostic reliability was numerically determined in the analysis of the spectral power density of the vibration signal and the amplitude of peaks, and the comparison of probabilities of error-free recognition for various modifications of the algorithm was made.