The identification of low fatigue damage using fuzzy double clustering framework

Abstract

Identifying damaging or non-damaging events in long records of fatigue data is a crux of recapitulating pristine data. In this article, a fuzzy double clustering framework (DCf) is utilized to classify the fatigue segment by exploiting two typical statistical features; kurtosis and the standard deviation. In the first stage, segments are assigned to a number of similar groups to generate multi-dimensional prototypes. Then, the resulting multi-dimensional prototypes are projected onto each featuring space of the input variables. On each dimension, a hierarchical clustering is applied to extract the information granules. For ease of interpretability, the granules are translated into a set of antecedent-consequent rules by means of a fuzzy set theory where for the model output, two distinct classes namely low and high with different degrees of evidence are assigned. The results reveal that the fatigue segments could be classified according to the value of kurtosis and standard deviation in a specific range where further, it can be a part of a fatigue data editing process