The effect of linkages in the Hierarchical Clustering of Auto-Regressive algorithm for Defect Identification in Heat Exchanger Tubes

Abstract

Pattern recognition approach based on Auto-Regressive (AR) algorithm is an alternative way to provide a more accurate defect identification from stress wave propagated along ASTM A179 heat exchanger tubes. The AR algorithm characterizes the shape of the stress wave signals by AR coefficients and clustered using ‘centroid’ linkages. However, the increase of number of stress waves limiting the function of clustering into meaningful groups. This paper proposes the ‘ward’ linkages as an improved hierarchical clustering method to define the defect features from the reference tube signals and those from the artificially induced defective tubes. The clustering results from the ‘ward’ linkages were represented via a dendrogram showing the hidden pattern between clusters. The defect in the heat exchanger tubes are easily interpreted from the dendrogram and can be successfully identified from Maximum Group Distance Criteria (MGDC). The pattern recognition approach using ‘ward’ linkages in AR algorithm has been shown to effectively identify the defects in the heat exchanger tubes.