Multiple local damage detection method based on time-frequency representation and agglomerative hierarchical clustering of temporary spectral content

• Published in: Applied acoustics Vol. 147; pp. 44 - 55
• Main Authors: Sokołowski, J., Obuchowski, J., Wyłomańska, A., Kruczek, P., Zimroz, R.
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.04.2019
• Subjects: Clustering; Short-time Fourier transform; Source signals separation; Clustering; Source signals separation; Short-time Fourier transform
• ISSN: 0003-682X; 1872-910X
• DOI: 10.1016/j.apacoust.2018.04.025

Vibration signals acquired on machines operating in industrial conditions often consist of mixture of various components. From local damage detection perspective, signal could be considered as regular part with spectral content related to kinematics and normal operation of machine and periodic, wide-band disturbance of spectrum associated to local damage. Using time-frequency representation we could often see or automatically evaluate frequency bands in which these periodic excitations appear. A concept proposed in this paper assumes that the difference between regular and fault related signals spectra could be recognised by data mining techniques (simple cluster analysis is used here). When the resulting spectra are processed by the clustering algorithm due to the different characteristics of spectra from “healthy” and “unhealthy” segments of signals they shall be gathered in different clusters. The resultant groups are the base for creation of new time-frequency maps which in next step are inverted via Inverse Short-time Fourier transform (ISTFT). In result one might obtain the extracted signal. The efficiency of the proposed method will be proved with simulation and real data analysis and the results will be compared to the classical kurtogram technique.