Characterization of Wavelet Decomposition Strain Signal Using the K-Mean Clustering Method

Abstract

This paper aims to study the characterisation of time-frequency domain to analyse the fatigue strain signal due to weaknesses in time domain and frequency domain approaches. The objectives were to determine the behaviour of strain signal, characterise the fatigue life of strain signal and validate the fatigue life in time-frequency domain. The strain signal was obtained using data acquisition devices and strain gauges on two types of road condition including highway and industrial area. The acquired signals were analysed with time domain, frequency domain and time-frequency domain approaches. In time-frequency domain, the signals were decomposed using 4th Daubechies discrete wavelet transform. To validate the effectiveness of time-frequency approach in characterising vibration fatigue signal, fatigue data was clustered by mapping of the data based on the spectrum energy, root-mean-square and fatigue life obtained. The clustering was performed by comparing the centroid values which both data had five clusters as the optimum data clustering with 0.836 average distance to centroid. From this, the relationship between fatigue life, root-mean-square and spectrum energy can be determined and thus a new fatigue life criterion was developed.