Parameter Optimization in Cluster Identification Algorithms for Characterizing Nanoclusters in Al–Mg–Si–Cu Alloys

Abstract

Abstract Optimization of user-defined parameters (Dmax, Nmin, order (K)) in the Density-based Spatial Clustering of Applications with Noise (DBSCAN) algorithm, used to characterize nanoclusters in Al–0.9% Mg–1.0% Si–0.3% Cu (mass %), was conducted. Ten combinations of parameters with a given K were considered for samples naturally aged (NA) and preaged (PA) at 100°C. We confirmed four types of unphysical clusters, artificially formed, by analyzing composition with size, atomic density, and atomic arrangement inside clusters. The optimum combinations minimizing those unphysical clusters were obtained for both NA and PA samples. Meanwhile, to evaluate the reliability of the optimum combination, volume rendering and isosurfacing were performed. As a result, regions of high solute concentration were confirmed, and those regions are in good agreement with the position of the clusters obtained by applying the optimum combination in DBSCAN. Furthermore, by comparing the optimum combinations with the fixed parameters widely used until now, we showed that for each dataset, considering independent parameters obtained in the same method is desirable rather than using fixed parameters. Consequently, an idea of determining the algorithm parameters for characterizing the nanoclusters in Al–Mg–Si(–Cu) alloys was introduced.