Multivariate comparison of concentration profiles in materials analysis

Hierarchical cluster analysis (HCA) and principal components analysis (PCA) were applied to find groups between similar depth-profiles in thin-layers investigated by Rutherford backscattering spectrometry (RBS).HCA yields in one run an objective hierarchy of similarity for several profiles. Among the similarity coefficients examined the linear measure, the Euclidean distance and the exponential measure respond with different sensitivity to overall shifts in direction of the concentration axis, whereas the correlation measure relates to parallelism of the profiles.For agglomerative HCA with Euclidean distance, a lowest significant linkage level has been defined by use of Fisher'sF-test. For divisive HCA based also on Euclidean distance, the maximum of a separating function marks the most separating cluster step. The outcomes of both proposals agree for the data set investigated.PCA is useful for verifying the results of HCA and yields additional information about the data structure. In the actual example quite different positions of features (concentrations at definite depths) in the space of the two first principal components hint at peculiarities during the metallurgical coating process.