|
|
|
Rapid identification of organic contaminants in pretreated waste
water using AOTF near-IR spectrometry
Abstract:
A near-infrared analyzer
utilizing state-of-the-art acousto-optic tunable filter (AOTF)
spectrometry with digital wavelength control and a high D*
extended-range InGaAs TE-cooled detector provides excellent
wavelength repeatability (better than 0.02 nm) and very high
signal-to-noise ratio. Because the AOTF dispersive element is
completely solid-state (no-moving parts), as is the entire
spectrometer, the instrument is small, rugged and very reliable.
Using this spectrometer, methods employing chemometrics have been
developed and applied to measure organic contaminants such as
gasoline and a variety of jet fuels in water. Qualitative
identification of contaminants was achieved with discriminant
analysis software developed specifically for this task. Both the
technique of grouping sample spectra into specific clusters based of
Mahalanobis distances and that of matching each spectrum with the
most representative member of the appropriate group of calibration
spectra were used to identify contaminants. After initial
classification, appropriate qualitative chemometric calibrations may
be applied to more accurately assess the level of contamination.
This instrument will be used to evaluate ground water supplies.
Keywords:
Infrared Sensing,
Spectroscopic Analyzers, Multicomponent, Water Quality Monitors,
Composition Monitors, Process
Introduction
Certain
industrial or military installations may be susceptible to the
unexpected or accidental introduction of organic contaminants in
waste water streams. In some cases, these contaminants can be
destructive to the water treatment facilities. At Edwards Air Force
Base, dumping of fuel oils (diesel, jet fuel, etc.) has proven to be
fatal to the biological organisms used in their waste water
treatment system. However, chemically similar materials, such as
vegetable oil, have no detrimental effects. A method for real-time
analysis of waste water, ultimately capable of automatically
diverting contaminated streams when appropriate, is desirable.
Near-IR spectroscopy offers an attractive solution for the rapid,
routine detection and identification of organic substances in water
at non-trace contaminant levels. Combinations and overtones of
fundamental molecular vibrational modes are primarily responsible
for the absorptions that occur in the near-IR. Because the
absorptivities are substantially lower than those of the fundamental
frequencies observed in the mid-IR region, the use of longer path
lengths (1 to 10 mm) is possible. Near-IR spectroscopy can be
successfully used for routine quantitative and qualitative analysis
of bulk materials in numerous instances where conventional mid-IR
spectroscopy fails due to practical sampling considerations.
The molecular absorptions observed in the near-IR spectrum involve
strong bonds between relatively light atoms, as exist in water and
are typical of organic materials. Relatively high absorptivities of
water in selected regions of the near-IR spectrum, crude assessment
of overall water purity can be achieved by simply measuring the
magnitude of absorption bands in the vicinity of 1450 and 1940 nm.
Organic substances (hydrocarbons, solvents, etc.) can easily be
detected using near-IR spectroscopy. Detection is based upon the
characteristic absorptions of these materials occurring at specific
wavelengths throughout the near-IR spectrum. Computerized
discriminant analysis algorithms can be employed for identification
of the particular contaminant or contaminants involved. It may also
be possible to determine the contaminant concentration levels for
contaminants that are miscible in water.
Experimental
Mixtures of water, hydrocarbons and organic solvents were
spectroscopically analyzed to establish the feasibility of
contaminant level detection and contaminant identification. Spectra
were collected with the Brimrose Luminar 2000 near-IR spectrometer.
This spectrometer provides excellent wavelength repeatability and
very high signal-to-noise ratio. Utilizing state-of-the-art
acousto-optic tunable filter (AOTF) spectrometry with digital
wavelength control and a high D* extended-range InGaAs TE-cooled
detectors. The AOTF dispersive element is completely solid-state
(no-moving-parts), as is the entire spectrometer. Therefore, the
instrument is small, rugged and very reliable. Spectra were
collected over a 1200-2200 nm wavelength range at 2 nm resolution.
100 scans were averaged at a rate of approximately 10 scans per
second for each sample spectrum. Sampling was performed by immersing
the reflectance probe with its transreflectance attachment
(effective path length of 3 mm) directly into the liquid. Spectra of
pure jet fuel, contaminated jet fuel, water, and various other
organic materials were collected in order to identify characteristic
spectral features. A series of water and JP8 jet fuel mixtures were
sampled over a wide concentration level range, as were various
mixtures of water and methanol in acetone and ethanol.
Figure 1 shows the NIR absorbance spectrum of pure jet fuel that was
obtained. The large absorption peak occurring between 1600 and 1800
is very useful for spectroscopic identification and quantitation.
Less prominent spectral features in the 1200 and 1400 nm wavelength
regions might also be useful for longer path lengths. Figure 2 shows
the absorbance spectra of several different concentrations of jet
fuel and water. The concentrations of jet fuel in the water are
relatively high. Water absorption peaks in the vicinity of 1450 and
1940 nm are clearly visible. Absorptions corresponding to the jet
fuel between these peaks can be easily seen. This region provides a
useful window that can be utilized for the identification and
quantitative determination of jet fuel in water. Figure 3 shows
spectra of water with small amounts (one or more drops) of jet fuel
added. The spectral features of water predominate, although slight
differences can still be readily observed. The high signal-to-noise
that can be achieved enables extraction of analytically useful
information from subtle spectral differences. This information can
be extracted via computer upon applying appropriate statistical
algorithms.
The
utility of the spectrometer for performing rapid qualitative
material identification was demonstrated. Figure 4 shows
superimposed spectra obtained for several different liquids using
the spectrometer with reflectance probe and transflectance
attachment (overall path length approximately 2 mm). The water and
ethanol spectra, although substantially different, both exhibit
significant O-H absorptions in the 1400-1500 nm and 1850-2000 nm
regions. Chemical similarities of jet fuel and vegetable oil (both
hydrocarbons) are reflected in certain similar spectral features,
although significant spectral differences are also readily apparent.
The region between 1650 and 1800 nm yields useful chemical
information about the C-H bonding for a particular substance.
Organic materials and solvents (ethanol, acetone, fuel oils, etc.)
can be easily distinguished from inorganic materials such as water.
Figure 5 shows the large differences in the near-IR spectrum of
water in comparison to hydrocarbons such as gasoline and jet fuel.
Figures 6 and 7 show that, although the spectra of different
hydrocarbons are somewhat similar, definite spectral differences
exist enabling discrimination. These spectral differences are most
apparent in the 2050-2200 nm (Figure 6), 1600-1850 nm (Figure 6),
and 1350- 1450 nm (Figure 7) spectral regions. The entire near-IR
spectrum can be effectively used for qualitative identification of a
wide array of materials.
An
automated system was required for analysis of the near-IR spectral
information and identification of unknown substances although
spectroscopic differences were apparent . An identification software
package was tested for this purpose. This software package is
capable of utilizing spectral information to discriminate between
sample types and spectrally match materials within a group. The
software is Microsoft¨ Windows¨ based,
providing high computer platform compatibility and user friendly
operation. Spectra were assembled into calibration, or training sets
by logically grouping them into distinct categories based on known
chemical differences between corresponding samples. This information
was used to train the software.
The
identification software enables characterization of unknown
materials on various levels. Most fundamentally, the software allows
for discrimination between different groups, or types of materials.
The software may be trained to distinguish between water and
different alcohols, hydrocarbons (jet fuel, vegetable oil, etc.),
organic solvents, or other organic materials based upon their
respective spectral features, for example. This is accomplished by
first performing a calibration, whereby spectra from numerous
samplings of these materials are first collected. These spectra are
used to train the system by creating multiple group files, each of
which specify the spectra that will define the spectral
characteristics for a certain substance. Once created, various
parameters associated with these group files can be viewed and/or
edited. Descriptions of the specific spectra that comprise a group
can be viewed and individual group members may be selectively
deleted. Once a system of substance groups is developed, the
spectral similarities within a particular group and the spectral
differences between groups are utilized for future prediction of
unknown samples. Addition of new material types or deletion of
irrelevant material is easily performed by the addition or deletion
of group files. Discrimination between groups is achieved by
measuring distances between a particular "unknown" sample spectrum
and the defined groups consisting of spectra collected from "known"
materials as measured in multidimensional space in Mahalanobis
distance units. If the spectrum of a spectroscopically analyzed
material is found to be sufficiently far away from any of the
defined groups, the material is classified as a truly unknown
material, the identity of which requires some other manner of
determination.
The
identification software is capable of operating in a real-time
sample I.D. monitoring mode in conjunction with the spectrometer.
Identification analyses are performed at a user-defined sampling
rate, averaging a user-specified number of scans per sample spectrum
to make the analysis, with other scan parameters set to conform with
the calibration data. Real time sample identification using this
program was demonstrated upon training the identification software
to distinguish between different materials, including water, fuel
oils, vegetable oil, and organic solvents (as plotted in Figure 4).
Utilizing the spectrometer with a reflectance probe and
transflectance attachment (total path length approximately 2 mm),
the system was found to be capable of reporting material identity in
real time at a sampling rate of only 2 seconds. Various liquids,
including jet fuel and water in a "layered" mixture, could be
reliably distinguished at this sampling rate. Calibration robustness
has been positively established by dismantling/reconfiguring the
system and operating the system at different sites and on different
days.
Furthermore, the software is capable of identification of a specific
material according to the members that comprise a particular group.
For example, a group may be defined for fuel oils, comprised of
spectra of diesel, jet fuel, kerosene, etc. In this manner, the
software can, on a primary level, identify an unknown material as
belonging to the fuel oil group. On a secondary level, by matching
of the unknown material spectrum with members of the fuel oil group
calibration set, the particular type of fuel oil that the unknown
sample represents can also be determined. This capability may be
used as a semi- quantitative tool. Different concentrations of a
particular component may make up the group for a certain material
type. The material identification performed for spectra within that
group could determine which component concentration level is most
representative of a particular "unknown" sample.
Conclusions
The
technique of grouping sample spectra into specific clusters based on
Mahalanobis distances and that of matching each spectrum with the
most representative member of the appropriate group of calibration
spectra were used to identify contaminants. Overall, the approach
was found to work well for both identification of potential
contaminants as well as subcharacterization within a particular
substance category. Additionally, coupling of the qualitative
material identification capability of the spectrometer and
identification software with the excellent quantitative performance
of the spectrometer should enable spectral data for a previously
unknown sample to be first used for sample identification and
subsequently applied to analytical calibrations appropriate for the
specific material type in order to determine relevant component
concentrations. It is anticipated that this instrument will be used
to evaluate waste water streams in and around Edwards Air Force
Base.
Acknowledgments
The authors gratefully acknowledge
the SBIR funding from the U.S. Air Force: Contract No.
F04611-94-C-0108, managed by Ronald W. Mahlum, Larry Tolley and
Robert Busch, Environmental Management Office, Edwards Air Force 
Figures |
