SAIL/ES/Cf- 77 9g70
Active voltammetric microsensors with neural signal processing
Michael C. Vogt* and Laura R. Skubal
Argonne National Laboratory, Argonne, IL 60439
ABSTRACT
Many industrial and environmental processes, including bioremedation, would benefit from the feedback and control
information provided by a local multi-analyte chemical sensor. For most processes, such a sensor would need to be rugged
enough to be placed in situ for long-term remote monitoring, and inexpensive enough to be fielded in useful numbers. The
multi-analyte capability is difficult to obtain from common passive sensors, but can be provided by an active device that
produces a spectrum-type response.
Such new active gas microsensor technology has been developed at Argonne National Laboratory. The technology couples
an electrocatalytic ceramic-metallic (cermet) microsensor with a voltammetric measurement technique and advanced neural
signal processing. It has been demonstrated to be flexible, rugged, and very economical to produce and deploy. Both narrow
interest detectors and wide spectrum instruments have been developed around this technology. Much of this technology's
strength lies in the active measurement technique employed. The technique involves applying voltammetry to a miniature
electrocatalytic cell to produce unique chemical "signatures" from the analytes. These signatures are processed with neural
pattern recognition algorithms to identify and quantify the components in the analyte.
The neural signal processing allows for innovative sampling and analysis strategies to be employed with the microsensor. In
most situations, the whole response signature from the voltammogram can be used to identify, classify, and quantify an
analyte, without dissecting it into component parts. This allows an instrument to be calibrated once for a specific gas or
mixture of gases by simple exposure to a multi-component standard rather than by a series of individual gases. The sampled
unknown analytes can vary in composition or in concentration; the calibration, sensing, and processing methods of these
active voltammetric microsensors can detect, recognize, and quantify different signatures and support subsequent analyses.
The instrument can be trained to recognize and report expected analyte components (within some tolerance), but also can
alarm when unexpected components are detected. Unknowns can be repeat-sampled to build a reference library for later post
processing and verification.
Keywords: voltammetry, polarography, microsensor, sensor, chemical, intelligent, neural network,
1. A BACKGROUND ON GENERAL SENSOR STRATEGIES AND SUPPORT
1.1 Active vs. passive sensors
Typical passive industrial gas sensors attempt to measure gas presence and/or concentration as a function of a single
parameter such as resistance. This is the case in a semiconductor such as tin dioxide (SnO2), where the presence of a
hydrocarbon such as methane (CH4) will react with the SnO2 to cause a measurable rise in electrical conductivity.' While
this type of resistance measurement "can" be performed over a scanned spectrum of applied voltage, it typically is not. The
single reading that is produced is very susceptible to modification by many other external factors such as temperature,
pressure, humidity, and/or rate of change of any of these factors. The added complexity of implementing a voltage sweep vs.
a single applied potential is usually avoided solely for economical reasons although the additional information gained by
actively applying a swept voltage and producing a non-linear spectrum response curve is more beneficial. Similar limitations
are seen with strictly passive optical sensors, which monitor only the attenuation of a single, narrow-focused, wavelength
band.2 Passive technologies were developed when dedicated microelectronics were expensive, unlike modern Application
Specific Integrated Circuits (ASICs) and microcontrollers. These technologies now artificially burden chemical phenomena
by attempting to produce linear or near-linear output voltages as a function of a passive chemical reaction. This severely
limits the phenomena available for modern chemical sensor design.