Characterization and differentiation of aluminum powders used in improvised explosive devices—Part 3: Comparison of statistical analysis methods

Abstract Determining the extent to which sources of aluminum (Al) powders, often used as fuel in improvised explosive devices (IEDs), can be differentiated is important for forensic investigations and gathering intelligence. Previous work developed effective methods of characterizing Al powders using micromorphometric features of the Al particles and a multistage sampling approach. Since then, ~100 additional samples from Al powder sources representing five powder types used in IEDs and 33 product distributors have been added to the dataset. Using this large dataset, a study confirmed that 200 randomly selected fields of view (FOV) are needed to accurately characterize the powder. Three different statistical methods, each using a different method of summarizing the large volumes of data, are used: a modified Wasserstein distance score nearest neighbor classifier for the FOV means, an ASTM‐style match interval for means of the FOV means, and a linear discriminant analysis for the means of means of means. Two of the methods classify each questioned subsample to an Al powder sample, whereas the ASTM‐style method classifies questioned/known‐source subsample pairs as matching or non‐matching. All three classifiers show that Al powder sources can be discriminated, although samples of the same powder type or made of Al products from the same distributor are often confused. Analysis of Al powder samples from three casework IEDs shows they were likely made using Al powder from Al‐containing paint products. These results are integral to closed‐set classification of Al powders where the source of a questioned subsample is contained in the database.