Past Researchers: Matt Aernecke
With the remarkable combination of high sensitivity and broad-band odor recognition, the human nose is one of the most advanced sensors known. The 'Artificial Nose' project involves the development of a microbead array system that is based directly on natural olfaction mechanisms. The microbead array is composed of silica and polymer microbeads coated with environmentally-sensitive fluorescent dyes. Brief vapor pulses give rise to a series of different temporal responses which are processed by various pattern recognition techniques . Classifiers are trained with databases of vapor responses and applied for subsequent vapor identification.
Figure 1 A series of steps are involved in the sensor system assembly and vapor recognition process: construction of a randomized microbead array (a), registration of sensor positions (b), collection of a database of vapor patterns (c), training of a classifier (d), and recognition of unknown vapors (e).
Microbead sensors are individually analyzed with respect to fluorescence intensity changes vs. time using imaging software and a fluorescence microscope with CCD-based detection. Arrays are densely packed with microbead sensors (up to 24,000) and each sensor type is represented in thousands of microbead replicates. The response signal-to-noise ratio improves by averaging the single bead responses by sensor type.
Sealed bubblers are filled with liquid analyte and purged with air to obtain a saturated headspace of vapors, which is subsequently diluted with air and delivered to the microarray in a pulsatile fashion. An automated vapor delivery system (Sensor Research and Development, Orono, ME) enables the generation of an unlimited number of vapors and vapor mixtures at preset concentrations and mixture composition. Because the vapor delivery and response acquisition software codes are integrated, hundreds to thousands of vapor responses may be collected without interruption. Our current system is pictured below.
Figure 2 Photo of the current imaging system used to acquire vapor responses. Hardware components include a high-speed CCD camera, a fluorescence microscope with objectives, a lamp, shutter and filter wheel controllers, a vapor delivery system and a vapor sparge system.
Figure 3 A cartoon schematic of the optical block.
Vapor-sensitive Microbead Sensors and Sensor Arrays
Individual sensors are fabricated using a fluorescent dye (solvatochromic, vapochromic, pH-sensitive, etc. ) in combination with micrometer-sized silica or polymer beads, or polymer-coated silica microbeads. The polymers and surface groups can act as concentrators and/or adsorbents of organic vapors which help maximize analyte/dye interaction. Sensor responses reflect changes in fluorescence intensity and wavelength shifts that occur as vapors are presented to the sensor array. Many parameters ( e.g. vapor diffusion through the polymer layer, polymer type, surface-vapor interactions, pulse time, and pulse regime) contribute to the optical response resulting in unique chemical signatures for a particular vapor-sensor combination. Each sensor stock contains millions of individual microbead sensors, allowing fabrication of thousands of arrays with highly reproducible sensor responses. Figure 3 shows remarkable sensor-to-sensor, as well as day-to-day reproducibility from one sensor type exposed to 4 vapors on 5 different days.
Figure 4 Reproducible responses to 50% saturated heptane (green), ethanol (magenta), acetone (red), and water (grey) vapors obtained by averaging individual responses of 50 sensors. Each group of 50 microbeads was randomly selected from one of the 5 arrays that were prepared and tested on 5 different days.
Pattern Recognition Techniques
Our current software, Nose Data Analyzer (NDA) (Department of Computer Science; Christopher Collier and Prof. Lenore Cowen), preprocesses raw sensor responses and organizes the data into user-defined databases. The preprocessing software enables the user to scale raw responses, register sensor locations on an array, visualize responses, etc. The databases are output into a format readable by pattern recognition software (Weka, University of Waikato). Figure 6 shows images of the graphical user interface (GUI) of the NDA data preprocessing software.
Figure 6 Images of the NDA software: averaged responses of one sensor type to three different vapors are visualized on the same plot (top) and responses of randomly positioned microsensors are color-coded after being registered by the software (bottom).
Current research is focused on detection of harmful vapors released in highly populated areas. Ideally, this application will use a sensor platform that collects information at preset time periods ( e.g. every minute) until a vapor is released. Such approach is necessary because it minimizes the exposure of sensors and therefore prolongs their usage time. We are developing an adaptive learning system that uses the shortest possible acquisition and data analysis times by applying optimized conditions, learned previously, to solve each discrimination query. The system adjusts its sensing process according to the changes in the environment ( e.g. presence of harmful vapors). Instead of using long processing times by analyzing the entire amount of information available from the sensor array, the system solves every problem in a hierarchical fashion ( e.g. the system answers a hierarchy of yes/no queries to obtain a final answer).
“Extending the Longevity of Fluorescence-Based Sensor Arrays Using Adaptive Exposure”, S. Bencic-Nagale, D.R. Walt. Analytical Chemistry , 2005 77, 6155-6162