Feature Selection and Sensor Array Optimization in Machine Olfaction

Feature Selection and Sensor Array Optimization in Machine Olfaction

Alexander Vergara (University of California, USA) and Eduard Llobet (University Rovira i Virgili, Spain)
Copyright: © 2011 |Pages: 61
DOI: 10.4018/978-1-61520-915-6.ch001

Abstract

In this context, the main objective of this chapter is to provide the reader with a thorough review of feature or sensor selection for machine olfaction. The organization of the chapter is as follows. First the ‘curse of dimensionality’ and the need for variable selection in gas sensor and direct mass spectrometry based artificial olfaction is discussed. A critical review of the different techniques employed for reducing dimensionality follows. Then, examples taken from the literature showing how these techniques have actually been employed in machine olfaction applications are reviewed and discussed. This is followed by a section devoted to sensor selection and array optimization. The chapter ends with some conclusions drawn from the results presented and a visionary look toward the future in terms of how the field may evolve.
Chapter Preview
Top

Introduction

Machine olfaction applications would greatly benefit from gas sensors with high sensitivity and specificity or selectivity to target analytes, low cross-sensitivity to interfering species, fast response and full reversibility of the detection mechanism and, finally, no long term drift (Göpel, Hesse, & Zemel 1991, Gardner & Bartlett 1999, Nanto & Stetter 2003). However, such ideal gas sensors are largely unrealizable today. Real sensors show a trade-off between sensitivity, selectivity, response time and reversibility. While highly sensitive and selective sensors showing fast responses are typically associated with strong analyte to sensing material interactions, full reversibility is only achieved if the interaction between analytes and sensing material is weak. A weak enough interaction allows the total desorption of analytes or reaction by-products from the sensor surface, i.e. re-establishment of the clean surface conditions, during a recovery phase. Since reversibility is an essential aspect for any machine olfaction application, it is necessary to compromise, and hence it is normally necessary to use sensors which show only partial specificity for some of the target species. Gas sensors experience long term response drift often due to changes in the sensitive material; e.g. mechanisms that cause poisoning or changes in the material micro/nano structure after long operation at temperatures well above room temperature; or to degradation of the transducer element which is needed to support the gas sensitive material.

The response of an individual sensor to a given gaseous environment or odor generally consists of a current measured at a fixed voltage, a resistance of the active material or a resonant frequency. When this response is measured in a steady-state situation, which is often the case, a stand-alone feature per sensor is obtained at a time. To improve the poor selectivity performance of individual sensors, arrays of several sensors are commonly used, from which response vectors; i.e. a response feature per sensor within the array; are obtained and treated by employing pattern recognition engines (Brereton 1992). Increasing the dimensionality of response vectors; i.e. increasing the number of response features per measurement; seems logically to be a useful way to significantly improve the analytical capabilities of a machine olfaction instrument (Lorber & Kowalski 1988), provided that the new features response carry additional information.

Exploiting sensor dynamics is a useful way to obtain additional features, and this can be implemented through the modulation of either internal (e.g. operating temperature, bias voltage) or external (e.g. gas flow, pre-concentration or separation columns, etc.) parameters. Features extracted from out–of–equilibrium sensor responses carry information about the interaction between odorant molecules and the gas-sensitive materials (e.g. kinetics of adsorption, desorption, diffusion, and, eventually, reaction); therefore, such features can help with the identification or quantification of specific analytes or complex odors (Llobet 2006, El Barbri et al. 2008).

However, exploiting sensor dynamics is not the only existing strategy for making use of an extended set of features. This can be further combined with other possibilities that include the use of hybrid sensor arrays consisting of a number of different gas sensitive materials together with suitable transducer architectures (e.g. optical fiber, surface acoustic wave devices, microbalances, chemo-resistors) and geometries (electrode configuration, micro-hotplate, etc.), which employ different properties to convey information from the chemical domain into the electrical or optical domains. Göpel (1988) estimated that the full exploitation of all these combinations would lead to 1021 chemical sensor features. Although this figure is highly hypothetical since the number of useful parameters and realizable variations is for sure significantly lower, this shows that the number of potential features available to solve a given odor analysis problem can be extremely large.

Complete Chapter List

Search this Book:
Reset