Deconvolution of Post-Detonation Mixtures of Soot
MADELINE STRICKLIN, Ryan Farley, James E. Lee, Rachel Huber, Allison Aiken,
Los Alamos National Laboratory Abstract Number: 98
Working Group: Source Apportionment
AbstractMost atmospheric soot is formed through incomplete combustion (e.g., from vehicle emissions, in-home heating/cooking, power plants, wildfires), and consists of submicron particles. While combustion soot is relatively well-studied, we present on the properties of detonation soot, which is produced by shock-driven decomposition at much higher temperature and pressure regimes. Bulk detonation soot exhibits a range of chemical properties depending on the fuel, detonation conditions, and environment, which can be useful for defining source-dependent signatures.
We use Soot Particle – Aerosol Mass Spectrometry (SP-AMS) to study these chemical signatures to distinguish different types of soot produced from different high explosive (HE) composites and different experimental conditions. The mass spectra identify chemical signatures that might be associated with different HE composites that are composed of “organic” materials (CHON) similar to natural fuels that produce combustion soot.
While distinguishing between pure samples of detonation soot is useful and we have done using a kernel-based algorithm, it is particularly advantageous to identify when a single sample of soot originates from multiple HE composites. Here, we consider mixtures of post-detonation soot according to the relative proportions of individual ions that exist within different moiety families (e.g., organics, nitrogen-containing organics, black carbon). This framework allows us to describe a sample of soot as a mixture of multinomial distributions over a fixed number of family types, where each multinomial distribution corresponds to a particular source of soot. We then use Latent Dirichlet Allocation to “deconvolve” this sample into its different parts. This allows us to infer not only the contributions of different types of HE to a mixture of soot, but also the profiles associated with these different types of HE. A comparison between this method and other more traditional AMS deconvolution techniques, such as positive matrix factorization (PMF), will also be presented.
This abstract has supplementary material available. Click PDF icon to download.