<br/>The large amount of radiant energy released during a nuclear detonation fuses elements form the environment into a glassy solid referred to as Trinitite or melt glass. These sorts of samples are of interest to scientists and are often collected by first responders after an event. By studying the formation of melt glass samples, the chemistry of the nucleation and particle formation can be predicted so that general features of the melt glass could be known before collection. Appropriate characterization of melt glass features before collection could assist in limiting the dose to first responders by reducing the time required to collect samples from high radiation environments.<br/>Optical emission spectroscopy of a high temperature environment surrounding a continuous dusty plasma was used to study the molecular formation process of melt glass. Due to the nature of the system, the optical spectra were dominated by lines from elemental Ar and H used to sustain the plasma. In order to study the more interesting features obscured by “background” artificial neural networks and support vector machines were used to remove unwanted background features. A classification support vector machine (SVM) was used to classify collected spectra into two categories, “Urban” and “Desert” based on spectral features. SVM’s were able to classify spectra with around 98% accuracy. Linear discriminant methods had a lower accuracy and decision trees were avoided due to their tendency to over train. A supervised trained neural network was used to map the collected spectra of a complex matrix combined with a spike of interest to a spectra representative of only the spike. Once the network was appropriately trained, it was tested using the same matrix combined with a different spike.

Neural networks for removal of background matrix peaks from optical emission spectra

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.