Novel Data Science Driven Chemical Agent Sensors: Towards Better Discrimination in Complex Environments

Chemical agents remain a persistent threat to both Warfighters and civilian personnel in both combat and non-combat environments. While many useful detection technologies exist to provide early warning, many of them suffer from similar drawbacks. Most common are unacceptable false positive response rates and inability to discriminate target analytes in complex or obscurant-laden environments. A typical approach to this problem is to develop material solutions for more specific and sensitive sensors. While useful, these solutions are often too expensive, not practical, or non-deployable. Our approach fuses both material and data science to extend the power of multiplexed sensor arrays towards more accurate and faster detection and discrimination. This presentation will discuss our approach towards vapor analyte detection using multiplex array systems combined with novel data science approaches. For vapor phase detection, sorptive and conductive graphene-based nanocomposite polymer films as well as select 2D conductive materials are used for semi-selective affinity towards chemical agent simulants in binary mixtures with common obscurants. Response data are pre-processed using a novel algorithm called the Variational Autoencoder Kalman Filter and the entire response dataset is used to train neural networks for analyte discrimination. To inform the neural networks to facilitate accurate and fast analyte discrimination, datasets far exceeding the capability of laboratory measurements are needed. Here too we describe our approach to generate simulated data to supplement and augment the classification algorithm machine learning process. We lastly demonstrate this approach to discriminate binary and tertiary mixtures of analytes with both similar and different chemical similarity both in and out of the precise of common environmental interferents.