Developing an integrated clustering algorithm based on fuzzy C-means for characterizing sound signals of vibration test rig

Abstract

Fuzzy-based integrated clustering application, or FICA, has been developed in this research. It effectively performs data clustering by considering seven clustering quality indexes, namely the partition coefficient, classification entropy, modified partition coefficient, Fuzzy silhouette width, Xie-Beni, partition index, and separation index, leading to more optimal cluster results. In addition, FICA is also equipped with correlation coefficient and coefficient of determination functionalities that describe the relationship in the clustered data. For validation purposes, a data set of 30 sound signals with sampling frequencies ranging from 11-20 kHz has been measured with a voltage of 6, 9, and 12 Volts for 30 seconds. Similarly, key parameters such as sampling frequency, sound pressure level, and power spectral density have been obtained from sound signals and have been clustered and compared with open-source software. Validation results have shown that only Fuzzy silhouette width reports different results, although the difference does not affect the selection of the optimal number of clusters. Both FICA and the software has recommended the same number of clusters, which was 2, for the two and three dimensions. In conclusion, an integrated, user-friendly, and accurate clustering application for engineering data has been successfully developed.