Towards real-time detection and recognition of toxic industrial chemicals with temperature programmed microsensor arrays
Baranidharan Raman1,2, Douglas C. Meier1 and Steve Semancik1
1Chemical Science and Technology Laboratory,
National Institute of Standards and Technology (NIST), USA
2Laboratory of Cellular and Synaptic Neurophysiology,
National Institute of Health (NICHD, NIH),
35 Convent Dr Room 3A106, MSC 3715, Bethesda, MD 20892-3715, USA
Microsensor arrays developed in the Chemical Sciences and Technology Laboratory at NIST are MEMS-based conductometric chemical sensing devices that can offer considerable benefits over existing detection technology in terms of size, speed, and power consumption. Each array incorporates multiple temperature-controlled elements with differing conductive sensing oxide films applied via the chemical vapor deposition (CVD) process. These sensing elements can be individually addressed and programmed to cycle through many temperatures, capturing a greater range of temperature-dependent interactions between analytes and sensing films. In this study, we present advances made in tuning the NIST microsensor technology to achieve near real-time recognition of toxic industrial chemicals (TICs) in the presence of a variety of common interferences, such as bleach and paint fumes.
First, we take a statistical approach using Pearsonís correlation coefficients to assess orthogonality/similarity of conductometric responses from sensor materials. This correlation analysis greatly assists our selection of film thicknesses and compositions to incorporate within an array for a specific detection problem. Using dimensionality reduction techniques like principal component analysis and linear discriminant analysis to visualize the sensor array response, we then determine the sufficiency of available analytical information derived from the chosen materials and temperature programs for species recognition. Finally, we study the variation of signal baselines over extended operating times for our microsensors and develop self-calibration routines and training methods for drift reduction. We demonstrate the benefits of the proposed preprocessing stage for allowing previously-trained recognition models to remain viable for detection of certain low-level toxic industrial chemicals over time periods of months, even as a number of interfering mixtures are introduced into the sampled background. These advances are critical to the production of pre-programmed microsensors that can achieve near-real time, multi-species recognition relevant to homeland security and other applications.
CSTL Process Measurement Division (836.04),
Process Sensing Group, Physics (221), Room A307
100 Bureau Drive MS8362, Gaithersburg, MD 20899-8362, USA
Mentor: Steve Semancik
Sigma Xi member: No
Category: Mathematics and Statistics