Shipboard equipment malfunctions detection based on new data mining methods

The article is devoted the Hilbert-Huang transform for the vibroacoustic signals from sensors installed on shipboard equipment to diagnose their condition (considered on the example of vibration-acoustic signal processing from rolling bearings) effectiveness study. The classical frequency-time analysis methods drawbacks critical analysis, in particular the Fourier transform, which is not applicable to non-stationary processes, and the Wigner-Ville distribution, which is subject to interference components, is carried out. As an adaptive alternative, the Hilbert-Huang transform method is considered. The main attention is paid to the first stage, which is part of the method — empirical mode decomposition. A detailed description of the standard mode decomposition algorithm is given, the essence of which is to sequentially decompose the original signal into intermodal functions, each of which is a mono-frequency component. To overcome the standard empirical mode decomposition algorithm known drawbacks, such as sensitivity to noise level and the mode mixing phenomenon, a modified version of it is proposed and tested in the work, built on the basis of the ensemble empirical mode decomposition method with an improved symmetric extension procedure to suppress boundary effects. Experimental research was conducted using an open database of real vibrationacoustic bearing signals with artificially created defects. It is shown that the modified algorithm effectively suppresses the mode mixing effect characteristic of the standard mode decomposition algorithm, which allows diagnostic features to be clearly localized in specific empirical modes. It is established that the application of the modified algorithm leads to the formation of a Hilbert spectrum with higher resolution, which is expressed in a decrease in the entropy of the spectrum. A quantitative comparison of the Hilbert spectrum clarity index for the considered algorithms is carried out, on the basis of which the effectiveness of applying the modified ensemble empirical mode decomposition together with the Hilbert-Huang transform is demonstrated due to the ability to isolate and clearly frequency-time localize informative defect features (such as impulse repetition frequency), which significantly increases the reliability of diagnostic conclusions and confirms the method's promise for implementation in technical condition monitoring systems. The results obtained demonstrate the high potential of applying the Hilbert-Huang transform together with the proposed modified ensemble empirical mode decomposition method for defect type accurate diagnosis and its temporal dynamics assessment, which is the basis for building technical signal diagnostics systems.