Data-Driven Property Estimation for Protective Clothing
Final rept. Jan 2012-Dec 2013
ARMY NATICK SOLDIER RESEARCH DEVELOPMENT AND ENGINEERING CENTER MA
Pagination or Media Count:
This report details an exploratory effort for making data-driven prediction of barrier properties, completed in 2013 at the U.S. Army Natick Research, Development and Engineering Center NSRDEC, as part of the integrated protective fabric system IPFS project sponsored by the Defense Threat Reduction Agency DTRA. Desorption data for 23 organic solvents data chemicals from butyl rubber were measured and comprehensively analyzed to estimate diffusion coefficients. Using commercial computational chemistry software, machine-readable structures were built and numerous molecular descriptors calculated for these solvents and several threat agents and simulants query chemicals. MatlabR codes were developed and probed to implement a machine learning technique --Artificial Neural Networks. Trained using the descriptors and diffusion coefficients of the data chemicals, the network is able to make good predictions for query chemicals for which only the descriptors are known. Cheminformatics --demonstrated in this work for characterizing threat agents --has a broader scope, in computational toxicology.
- Chemical, Biological and Radiological Warfare
- Protective Equipment