RAMAN SPECTROSCOPY APPARATUS AND METHOD FOR CONTINUOUS CHEMICAL ANALYSIS OF FLUID STREAMS
Field of the Invention The present invention is directed to an apparatus and method for continuous chemical analysis of fluid streams, particularly fluid streams containing petroleum products, using Raman spectroscopy. The technique can also be applied equally well to aqueous streams and biological samples.
Technology Background
The refining and processing of crude petroleum into commercially useful petroleum products is a vital industry around the world. One of the most important petroleum products is the class of gasoline fuels. Gasoline fuels consist of a mixture of various hydrocarbon compounds. The concentration and chemical grouping of diese hydrocarbon compounds determines the resulting fuel properties such as octane number and Reid vapor pressure (RVP). Reid vapor pressure provides a volatility measurement of the gasoline. The octane number for a gasoline fuel is defined in terms of its knocking characteristics relative to a standard blend of isooctane (2,3,4-trimethylpentane) and Λ-heptane. Arbitrarily, an octane number of zero has been assigned to /z-heptane and a rating of 100 to isooctane. Thus, an unknown fuel having a knocking tendency equal to a blend of 90% isooctane and 10% «-heptane, by volume, is assigned an octane number of 90.
During the manufacture of various grades of gasoline, it is useful to monitor the final product to be sure it possesses me desired physical properties, such as octane number. However, there currently is not a quick and inexpensive system for continuously monitoring a gasoline fuel composition. Instead periodic samples are normally taken from the process stream and analyzed. Occasionally, a gasoline fuel blend sold at one octane rating actually has a higher octane rating. This is uneconomical, since higher octane gasolines are more valuable to the refinery than lower octane gasolines.
It is also useful to control the concentration of various hydrocarbon ingredients in the gasoline blend. For instance, it is desirable to reduce the concentration of benzene, a known carcinogen, in gasoline. Yet, benzene concentration is often difficult to measure using conventional analysis. It is also desirable to minimize me olefin concentration. Olefins are unsaturated hydrocarbons (containing C=C bonds) which are photoreactive and contribute to smog formation. Olefins are also hard to measure accurately using conventional techniques. Finally, because xylene is a valuable gasoline ingredient, excess xylene should be minimized. It
also would be valuable to identify and quantify the xylene isomers (para, meta, and ortho) present in the gasoline, but conventional analytical techniques cannot quickly distinguish between the xylene isomers. Thus, it would be a significant advancement in the art to permit continuous monitoring of the gasoline's chemical composition and physical properties. Concentrated sulfuric acid is used by refineries in the alkylation process. The sulfuric acid concentration needs to be carefully controlled between 90% and 98% during the alkylation process. Currently, refineries manually take individual "grab-samples. " Because of the unreacted hydrocarbon contamination in the acid, me samples require centrifuging to separate the acid. The acid is then titrated in the laboratory to determine exact acid content. These tests are typically run every four hours. Because of the time constraints with this type of testing, the refineries tend to run much higher acid concentrations man they would like to ensure that the process proceeds uninterrupted. Maintaining an excessively high acid concentration costs millions of dollars annually at each refinery.
When the acid concentration reaches 89%, it is removed from the process and hauled back to a chemical plant for recycling. This process of constantly removing, adding and transporting concentrated sulfuric acid is expensive and potentially dangerous. Because of the high cost of sulfuric acid, Amoco Oil Company calculates mat for each 0.5% drop in average acid content in their alkylation process, they could save $5,000,000 per year throughout their entire production system. It is estimated that a continuous, on-line, analysis of acid concentration would enable the average acid concentration to be reduced by about 1 %. Taking into account the additional savings from reduced transportation and handling costs, this total savings could average $12,000,000 per year for Amoco Oil. The sulfuric acid concentration problem is universal for the petroleum industry, and it has been estimated that a continuous, on-line chemical stream analysis could save me U.S. petroleum industry over $100,000,000 annually, with some of this savings returning to customers.
The alkylation process does not react the same with all alkene compounds; propylene compounds for instance, react much differently than do most other alkenes. The alkylation process needs a rapid, on-line control to properly react to changing feed stocks, to efficiently adjust acid levels, and economically produce petroleum products from a wide variety of incoming crude oil. Thus, mere is a significant need in the art for a process and mediod which allows on¬ line analysis and control of acid content of such process streams.
Raman spectroscopy is an analytical technique which uses light scattering to identify and quantify molecules. When light of a single wavelength (monochromatic) interacts with a molecule, the light scattered by me molecule contains small amounts of light with wavelengths different from the incident light. The wavelengms present in me scattered light are characteristic
of the structure of me molecule, and the intensity of this light is dependent on the concentration of these molecules. Thus, the identities and concentrations of various molecules in a substance can be determined by illuminating the substance with monochromatic light and then measuring the individual wavelengths and their intensities in the scattered light. A continuing problem with Raman spectroscopy is me very low intensity of me scattered light compared to the incident light. Elaborate spectrometers, having high light gathering power and dispersion, high stray light rejection, and sensitive detectors, are required to isolate and measure the low intensity Raman scattered light. These instruments are costly and delicate, and are not well suited for use in industrial manufacturing or processing facilities. As a result, they have rarely been used outside of laboratory environments. Improvements in the fields of lasers, optical fibers, and filters enable one to remotely locate a fiber-optic probe from its laser light source and from its spectrometer.
It will be appreciated that mere is a need in the art for an apparatus and method for analyzing industrial fluid streams, particularly mose containing petroleum products, which provides quick and accurate results.
Summary of die Invention The present invention is directed to an apparatus and method for analyzing the composition of a fluid stream using Raman spectroscopy. The invention is particularly useful in continuously analyzing a fluid stream containing petroleum products, such as gasoline fuel or natural gas.
The apparatus includes a laser source for producing light having an excitation wavelengdi. Conventional laser diodes can be used as the light source. It is presently preferred to use an external cavity diode laser. The light output from the external cavity diode laser can be controlled to provide a broader wavelengdi, yet more powerful and stable light output ϋian typical single mode diode lasers. Almough current Raman spectrometers use very narrow line width lasers, the broader line widm obtained with the external cavity laser is adequate for the uses contemplated herein.
The light is introduced into a bundle of optical fibers containing an excitation optical fiber and a plurality of peripheral collection optical fibers. The fiber bundle is optically connected to a tubular Raman enhancement cell. The tubular Raman enhancement cell is configured to allow continuous sample fluid flow therethrough.
Light from the excitation optical fiber is coupled into the enhancement cell, and backscattered Raman signal (light) from the enhancement cell is coupled into me plurality of collection optical fibers. The optical fiber bundle is preferably separated from the flowing fluid
sample within the enhancement cell by a transparent fluid barrier (an optical window and\or lens assembly).
The tubular Raman enhancement cell is preferably lined with a material having an index of refraction less than the index of refraction of the fluid stream so that light reflects internally at the interface between the fluid sample and the enhancement cell liner. This eliminates light losses which would otherwise occur if the light passed through the liner and was reflected at the exterior (air interface) surface of the liner.
Scattered light from the enhancement cell enters the plurality of collection fibers for transmission to a Raman spectrometer. The exit end of the collection fibers is preferably aligned in a linear array so mat a linear optical signal is generated. The Raman spectrometer passes the linear optical signal through a filter to eliminate portions of the optical signal having the excitation wavelength. The linear optical signal is also passed through an optional optical slit. The optical slit typically has dimensions comparable to the linear array of collection optical fibers. The linear array of optical fibers can perform the same function as me optical slit by generating a linear optical signal. A commercially available charge coupled device converts me optical signal into a corresponding electronic signal to be analyzed by a computer and converted into a representation of me chemical analysis of me fluid stream.
A plurality of optical lenses are preferably provided to receive and convey me optical signal from the linear input array, through the Raman spectrometer, to the charge coupled device. A volume holographic transmission grating is used to disperse the signal, and an aberra¬ tion correction element is provided to correct optical aberrations introduced into the optical signal by the volume holographic grating element. A pair of 45 degree, right angle prisms is one - presently preferred aberration correction element. Omer optically transparent components such as wedges can be used for aberration correction. If necessary, the focusing function of the optical lenses can be performed with the holographic grating.
The use of an aberration correction element in combination with a volume holographic transmission grating can be used in conventional Raman spectrometers without an enhancement cell or optical fibers. In such cases, solid or non-fluid samples can be analyzed using conventional techniques. Through use of the optical fiber bundle, it is possible to locate the Raman enhancement cell in a remote location near an industrial process stream. The Raman spectrometer, laser source, and computer can be located in a distant, protected environment. The apparatus and method for analyzing the composition of a fluid stream using Raman spectroscopy is rapid and accurate. It has been shown to provide excellent analysis of fluid streams containing petroleum products. Of course, those skilled in the art will appreciate that the present invention can be
readily adapted for use in analyzing odier fluid streams such as aqueous streams and biological samples.
Description of me Drawings Figure 1 is a schematic view of an apparatus for analyzing the composition of a fluid stream using Raman spectroscopy within the scope of the present invention.
Figure 2 is a cross sectional view of a bundle of optical fibers used to transmit light having an excitation wavelength and to receive scattered light.
Figure 3 is a cross sectional view of the linear array of collection optical fibers. Figure 4 is an enlarged cross sectional view of one end of a Raman enhancement cell showing connection of an optical fiber bundle and fluid entrance into the enhancement cell.
Figure 5 is a cross sectional view of an aberration correction element and volume holographic grating widi in the scope of the present invention.
Figure 6 is a graph of the Raman spectrum of unleaded gasoline obtained with an apparatus according to the present invention.
Figure 7 is a graph of me Raman spectrum of various octane grades of gasoline fuel showing the changes in me key octane determinant peaks as the octane rating changes.
Figure 8 is a graph of the Raman spectrum around 1400 cm'1 and 1440 cm'1 of different types of fuels which can be used to indicate change in the laser power. Figure 9 is a graph of me Raman spectra of two different types of gasoline product, "cat gas" and unleaded premium (ULP), and the relative location of the olefin peaks at 1625 cm"1 to 1725 env1.
Figure 10 is a graph of the Raman spectra of the gasolines analyzed in Figure 9 widi an expanded peak in the region from 1625 cm"1 to 1725 cm'1 as the olefin concentration is increased from 0.9% to almost 29%.
Figure 11 is a graph of Raman intensity vs. total olefin concentration for the curves shown in Figure 10, indicating a linear relationship between Raman signal strength and the total olefin concentration.
Figure 12 is a graph of me Raman spectrum of a predistillate fraction rich in xylene illustrating the individual Raman peaks for the various xylene isomers, m-, o-, and p-xylene. Figure 13 is a graph of me Raman spectra for increasing concentrations of benzene in unleaded premium gasoline.
Figure 14 is a graph of Raman intensity vs. the benzene concentration for me curves shown in Figure 13, indicating a linear relationship between Raman signal strength and the benzene concentration in me gasoline sample.
Figure 15 is a graph of die Raman spectra for two different octane rated gasolines and the subtracted Raman spectrum of me two Raman spectra.
Figure 16 is a graph of me Raman spectra for two fuel samples from the same manufacturer having the same octane rating and me subtracted Raman spectrum of the two Raman spectra.
Figure 17 is a graph of me subtracted Raman spectra for various brands of gasoline subtracted from a single brand of gasoline.
Figure 18 is a graph of the Raman spectra for different types of fuels indicating die ability to identify the type of fuel from its unique Raman spectrum. Figure 19 is a graph of the Raman spectrum for pure sulfuric acid, with a major Raman peak at 920 cm 1.
Figure 20 is a graph of the Raman spectrum showing me sulfuric acid Raman peak present in a sample of alkylation acid.
Figure 21 is a graph of Raman intensity vs. acid percentage indicating a linear relationship between die Raman signal strength and the actual acid concentration in the alkylation acid process stream.
Detailed Description of the Preferred Embodiments Reference is now made to the figures wherein like parts are referred to by like numerals throughout. With particular reference to Figure 1 , a schematic representation of an apparatus 10 for analyzing the composition of a fluid stream using Raman spectroscopy is illustrated. The apparatus includes a laser source 12 for producing light having an excitation wavelengdi. Although conventional laser diodes can be used as the light source, it is presently preferred to use an external cavity diode laser. External cavity diode lasers can provide a stable, broader wavelength and higher power output signal than typical single mode laser diodes.
Currently preferred output power of the laser source is about 1 watt. The excitation wavelength is preferably in the range from 670 nm to 850 nm for use in Raman spectroscopy. The precise wavelengdi used may vary depending on me material being analyzed. For liquids which tend to fluoresce, a higher wavelengdi, such as 850 nm, is preferred. For gas phase materials, shorter wavelengths, such as 670 nm or 788 nm, are preferred. Because the Raman spectra shifts further to die red (longer wavelengths) and gas phase materials have larger Raman shifts, a shorter wavelengdi is preferred for gases so mat die Raman shift does not exceed die detection limit of me CCD detector (usually about 1 μ).
The external cavity diode laser may be a conventional Littrow or Littman external cavity laser configuration or odier similar configuration. It is currently preferred to construct die
external cavity diode laser widi me components coupled widi an Invar material. Invar is a nickel-iron alloy having a low coefficient of thermal expansion. Invar coupling provides proper alignment of the diode laser components when temperature at a range of temperatures.
The light from the laser source preferably has a line width of at least 50 GHz, widi typical a line widdi between 60 and 120 GHz. For many applications, such as analysis of liquid petrochemicals, a line width as high as 120 GHz can be used. For analysis of gas phase petrochemicals, a line width as high as 300 GHz can be used.
The light is introduced into a bundle of optical fibers 14 containing a center, excitation optical fiber 16 and a plurality of peripheral collection optical fibers 18. One currently preferred optical fiber bundle 14 configuration is illustrated in Figure 2. Such optical fiber bundles are commercially available from manufacturers such as Fiberguide Industries, Inc., Stirling, New Jersey. The fiber bundles preferably contain at least 30 or more fibers, and more preferably from 40 to 50 fibers. The individual fibers have a small size to maintain good spectral resolution. Optical fibers having a diameter in the range from about 50 microns to 100 microns are suitable, with die presently preferred size being about 100 microns. Typically, only one fiber is required as me excitation optical fiber, although it is possible to use more mat one fiber to carry the excitation light. The remaining optical fibers are collection optical fibers to receive backscattered Raman signal (light).
The fiber bundle is optically connected to a tubular Raman enhancement cell 20. The Raman enhancement cell 20 is configured to allow continuous fluid flow therethrough, illustrated by arrows A and B. A fluid stream inlet 22 and a fluid stream outlet 24 are located adjacent die ends of die enhancement cell 20 to allow fluid flow through die enhancement cell 20. Experimental tests indicate diat die fluid flow rate does not affect die Raman spectroscopy results and fluid flow can go in either direction. An optical connector 26 located at one end of the en- hancement cell permits connection of the optical fiber bundle 14 to die enhancement cell 20.
As shown best in Figure 4, the Raman enhancement cell 20 preferably includes die close coupling of the adjacent bundle of optical fibers 14 to couple the light from the excitation optical fiber 16 into the enhancement cell 20 and to allow backscattered Raman light from the enhancement cell to enter the plurality of collection optical fibers. A transparent fluid barrier 30 is preferably located between die bundle of optical fibers 14 and me fluid flow (shown by arrow A) within the enhancement cell 20. An index matching fluid 60 is put between the fiber bundle tip and die barrier to improve optical coupling efficiency. Suitable index matching fluids are commercially available. They are typically clear viscous fluids, such as mineral oil or glycerol, which fill any voids between two optical surfaces. Optical epoxy can also be used, but it would result in permanent attachment of the fiber bundle to die fluid barrier.
The tubular Raman enhancement cell 20 is preferably lined with a wavequide core (liner) 32 having an index of refraction less dian die index of refraction of the fluid stream. If the index of refraction of die liner is low enough, much of die light will reflect internally at die interface between die fluid sample and die enhancement cell liner 32, as shown in Figure 4. Of course, the enhancement cell can be lined with a transparent material, such as glass, which will cause reflection to occur at the exterior surface of the liner. Because light loses some of its intensity as it passes through the liner, it is presently preferred to line die enhancement cell with a low index of refraction material. Fluorinated polymers, such as polytetrafluoroethylene (PTFE), PFA, copolymer of tetrafluoroethylene and hexafluoropropylene (FEP), AF and Tefzel, are currently preferred because they have indexes of refraction from below about 1.29 to 1.35. Preferred index of refraction is one mat is less dian the fluid being analyzed. The enhancement factor for the cell is dependent upon the refractive index of die sample solution in the cell. Different solution refractive indices result in different critical angles and different effective sample path lengths.
The length and widm of the tubular Raman enhancement cell 20 can vary. Generally, a longer enhancement cell will yield a larger enhancement effect up to the point where additional cell length will have less and less effect because of internal losses in the liquid medium and die wavequide core. The tubular Raman enhancement cell 20 preferably has a length in die range from about 0.1 m to about 4 m, widi a preferred lengdi ranging from 1 to 2 meter, depending upon die degree of sensitivity desired. An enhancement cell length of one to two meters yields excellent amplification of the Raman signal while still being easy to fabricate and install.
The enhancement cell preferably has a diameter in die range from about 0.005 to 0.2 inches. For any given cell length, the smaller the wavequide diameter, the greater the enhancement effect. The preferred diameter is determined more by what size wavequide is die easiest to fabricate and produce and to what extent die Raman signal needs to be enhanced. A presently preferred diameter is about 0.03 inches, mainly because of the ease of fabrication and the ability to maintain a reasonable level of flow. Too small a diameter, such as 0.005 inches, will yield excellent Raman signal enhancement, but the pressure required to obtain a reasonable flow through the cell will be high. Too large of a diameter, i.e., greater than 0.2 inches, will greatly reduce the Raman signal enhancement effect for a practical lengdi of waveguide. The tubular Raman enhancement cell 20 may be rigid or flexible. In operation, acceptable performance is obtained widi die enhancement cell in eimer linear, curved or coiled configurations. If die enhancement cell is coiled, it is preferred to maintain the bend radius greater than '/. inch to minimize signal loss caused by bending die waveguide.
In use, scattered light from the enhancement cell enters the plurality of collection fibers for transmission to a Raman spectrometer 36. As shown in Figure 3, the exit end of the
collection fibers is preferably aligned in a linear array 38 so diat a linear optical signal is generated. The Raman spectrometer 36 passes the linear optical signal through a filter 40 to eliminate portions of the optical signal having the excitation wavelength. A suitable filter must have good laser line rejection, such as a holographic or dielectric filter. Suitable dielectric filters are manufactured by optical filter companies such as Omega Optical. The holographic filter is made by companies such as Kaiser Optical, Ann Arbor, Michigan and Ralcon Development, Paradise, Utah.
The linear optical signal is also passed through an optical slit 42 to remove stray light. The optical slit 42 typically has dimensions comparable to the widdi and height of the linear array 38 of collection optical fibers. The narrower die slit the better the spectral resolution. However, this increased resolution also reduces the overall signal throughput. The optical slit preferably has a widdi in die range from 10 microns to 150 microns. In a currently preferred embodiment, the slit ranges from 7 mm x 25 μm to 7 mm x 100 μm. The linear array of optical fibers can perform die same function as die optical slit by generating a linear optical signal. In this manner, the slit and associated optics can be replaced by the linear array of optical fibers.
A commercially available charge coupled device ("CCD") 44 converts the optical signal into a corresponding electronic signal to be analyzed by a computer 46 and converted into a representation of die chemical analysis of the fluid stream. The computer 46 can also be used to control die laser source 12. A plurality of optical lenses 48 are provided to receive and convey the optical signal from the linear array 38, through the Raman spectrometer 36, to the charge coupled device 44. The speed of the optical lenses can vary, but faster speeds are generally preferred in order to capture more light. The improved performance obtained by faster lenses must be balanced by die increased cost of die lenses. Optical lenses having a speed in die range from f/1.0 to f/2.8 have been found to be suitable.
A volume holographic transmission grating 52 is used to disperse the signal. If necessary, die focusing function of the optical lenses can be performed widi the volume holographic grating. An aberration correction element 54 is preferably provided to correct optical aberrations introduced into the optical signal by the volume holographic grating 52. One presently preferred aberration correction element 54, illustrated in Figure 5, is a pair of 45° right-angle prisms 54. Odier optically transparent components such as wedges can be used for aberration correction. The aberration correction elements could be joined together and placed in front of or behind the holographic grating element. It is important that the aberration correction element contain a sufficient thickness of optically transparent material to correct the aberrations produced by die volume holographic transmission grating.
The configuration shown in Figures 1 and 5 is currently preferred because it requires less space and allows die closer coupling of the optical elements, i.e. the grating, lenses, and CCD array. It also allows for die use of higher resolution gratings. The closer coupling allows for greater light throughput for any given lens size. The volume holographic transmission grating 52 can be fabricated with various degrees of spectral resolution with equivalent grooves/mm from several hundred grooves/mm up to 2400 grooves/mm. The preferred resolution is on die order of 1500 grooves/mm to 2400 grooves/mm. More grooves per mm increases die spectral resolution, but also decreases the optical signal diroughput. Increased spectral resolution also increases die aberration mat must be removed to obtain improved resolution. The choice of grating resolution depends upon die resolution required, die range covered by die CCD detector, and die degree of signal diroughput desired. The volume holographic transmission grating 52 is preferably sandwiched between die two optically transparent aberration correction elements 54.
The volume holographic grating 52 is used because of its high efficiency in dispersing light into the needed spectrum and also because it can be used widi very fast optical lenses. Overall signal diroughput is increased by several orders of magnitude compared widi conventional spectrometers using conventional reflective optics and gratings. This increased efficiency allows for enhanced sensitivity in detection of Raman signal and greater speed of detection. A currently preferred volume holographic transmission grating is commercially available from Kaiser Optical Systems, Ann Arbor, Michigan or Ralcon Labs, Paradise, UT. The prisms, optical lenses and all optical interfaces are preferably coated widi anti-reflection coatings to minimize light losses through the spectrometer.
Because the Raman spectrometer illustrated in Figure 1 relies on fiber-optic coupling of the Raman signal to the spectrometer, die volume holographic transmission grating 52 is preferably designed for optimum efficiency with die randomly polarized light carried by die optical fiber bundle 14. The grating is preferably designed to balance die diffraction efficiency for die "s" and "p" polarization modes. This allows the grating to pass more Raman signal when confronted widi a randomly polarized signal. Most holographic gratings are not designed to work efficiently in die near infrared region. Refractive index modulation in die grating is preferably optimized for near infrared operation. The emulsion thickness of the grating 52 is preferably designed minner to obtain a flatter diffraction efficiency curve over die near infrared spectral region (840 to 950 nm).
The prisms 54 have anodier function in addition to correcting signal aberration caused by die volume holographic transmission grating 52. When the grating is sandwiched between two optical wedges (prisms), the grating can operate at higher effective groove densities and provide
higher spectral resolution. The sandwiched prism design allows die holographic transmission grating 52 to have a grating resolution greater than 2300 grooves/mm. Normally, holographic transmission gratings operating in air alone have a practical resolution limit on the order of 1700 grooves/mm. The increase in resolution afforded by die disclosed design offers improved signal diroughput for faster integration times and improved Raman signal resolution. Raman peaks mat are very close togedier (resolution of 4.5 cm 1) can be effectively detected and measured according to the disclosed design.
Several experiments were performed using a Raman spectroscopy apparatus as shown in Figure 1 to analyze various petroleum products. The results of these tests are illustrated in Figures 6-18. The enhancement cell had a lengdi between 0.31 and 2 meters and a diameter between 0.06 and 0.03 inches. A laser diode generating a excitation wavelength between 780 and 800 nm was used. The laser diode had a power output between 50 and 100 mW. The integration times ranged from 60 to 90 seconds.
Additional experiments were performed to analyze sulfuric acid and process streams containing sulfuric acid. The results of diese tests are illustrated in Figures 19-21. The Raman spectroscopy apparatus was identical to diat described above, except diat an excitation wavelengdi of 847 nm was used.
Octane determination
Figure 6 identifies die location and relative intensities of die Raman peaks for the major constituents of gasoline. Toluene, m-, o-, and p-xylene, ethyl benzene, and benzene make up most of the important aromatic constituents. Isooctane is a major ingredient whose concentration can be adjusted to readily increase or decrease octane rating. The aliphatics, such as n-heptane and n-hexane, have Raman peaks that are much broader man the aromatics, and their peaks are illustrated in die region from 890 to 940 cm'1. By measuring the relative intensities of die "octane enhancers" such as m-, o- and p-xylene, isooctane, ediyl benzene and toluene to die intensities of die aliphatics, one can develop a model mat will determine the octane rating of me fuel. It is necessary to be able to distinguish diese compounds from the aliphatics that reduce octane ratings radier dian increase die octane rating. The aromatics and isooctane tend to be die most important compounds in determining octane rating and their Raman peaks are very narrow and distinct in mis "finger print" region of the frequency scale, i.e., 700 to 1025 cm '.
Commercially available chemometric analysis software such as MATLAB, sold by The Madi Works, Inc., Natick, Massachusetts, or GRAMS/386, sold by Galactic Industries Corporation, Salem, New Hampshire, can be used to develop die octane rating model. The combustion properties reflected in octane number can be derived from a variety of compositions of branched and aromatic hydrocarbons. Thus, the calibration model development must include a
full spectrum analysis to capture die range of structures that relate to octane number. Since die composition of fuels can vary considerably, die algorithm must have robust predictive capabilities. For mis task, it is currently preferred to first use the principal component regression approach of partial least squares or "PLS. " This approach extracts a set of factors that capture the variation in the calibration spectra diat relate to variation in die property being predicted while ignoring variation that is not related to this property. This attribute of PLS allows for a wider variation in compositions of unknowns that could be tolerated in classical least squares or more traditional principal component regression approaches.
Model development and calibration will proceed using standard samples which have been octane graded by standard knock engine techniques; samples from a variety of refineries will be brought into die calibration set. Cross-validation (leaving out the standard being tested) should be used to check die predictive capabilities of die model over die calibration set; many tests of die model and method should be performed during instrument calibration while operating on-line, in parallel widi a knock engine. Continuous quality testing of die model is carried out during analysis; lack-of-fit criteria testing of the model is carried out during analysis; lack-of-fit criteria (excess spectral error, structured residuals) are examined for all unknown samples widi outliers preferably stored for further evaluation. Once the model is developed and tested, suitable analysis software is developed based on die model.
Figure 7 illustrates how Raman measurements can identify die changes in die key octane determinant peaks as the octane rating changes in fuel. With higher octane rating the octane enhancer isooctane increases in concentration, while toluene concentration decreases. Each petroleum company has its own special blending practices to produce their higher octane rated gasolines. The Raman spectra need to be generated for each company's gasoline and diat information put into a mathematical model diat can men be used to determine the octane rating for that refiner's products. A model for one company's fuel system may or may not work directly for anodier fuel system. The models may have to be made individually for each blending pattern.
Figure 8 illustrates the existence of Raman peaks in the Raman spectrum around 1400 and 1440 cm"1 that change very little widi different types of fuels as well as widi different octane ratings. These peaks are made up of die Raman spectra of many of me gasoline components, thus diese peaks stay relatively constant widi changing individual gasoline constituents. Aldiough diese peaks are not useful for analyzing a particular aliphatic or aromatic component, they can be used as an internal reference to ensure that the relative excitation energy has not changed. Because die laser diode power output can fluctuate, any changes in laser power will show up as increases or decreases in die individual Raman peaks. Examining the intensities of die Raman
peaks mat change very little widi concentration, allows one to use this information to indicate if and when the laser power is changing. The use of these particular peaks does not mean that a routine calibration procedure is not needed; it does mean that die calibration may not have to be performed as often. Olefin Determination
Figure 9 illustrates die Raman spectra of two different types of gasoline product. The top curve is the Raman spectrum for a standard unleaded premium gasoline, die lower curve is for a gasoline product called "cat gas. " This particular product has a very high olefin content, typically 25 to 30%. It is shown here to illustrate the relative location of the olefin peaks (encircled region) at 1625 cm'1 to 1725 cm 1. The unleaded gasoline has a fairly low olefin concentration, around 6.5%, so die Raman spectrum for the olefin in me unleaded gasoline is much smaller. It turns out that most of the olefins (C=C double bond) compounds have Raman peaks in the region of 1625 cm"1 to 1725 cm'1. There may be as many as 50 to 100 different olefin species present in gasoline. It is not necessary to measure each individual olefin component, rather die total olefin concentration as a group is sufficient. The fact that Raman spectra for most of the olefins have at least one peak in this region allows us to measure the area under this curve and determine die total olefin concentration.
Third party chemometric analysis software can be used to develop a calibration model for determining olefin concentration in a manner similar to determining octane rating described above. Once die calibration model is developed and tested, suitable analysis software is developed based on this model.
The multivariate calibration of die model would preferably utilize a maximum likelihood or classical least squares approach to estimate olefin concentrations. A two vector design matrix, confined to wavenumber regions near the olefin peak(s), would be constructed with a "pure component" olefin vector and a baseline vector. The vectors for the calibration model can be determined by least squares from a series of gasoline standards containing known olefin content, while several functional forms for the baseline are tested for quality of fit. If more than one olefin scattering band is included in die model, men each band can be individually tested for its prediction uncertainty by cross-validation, and the results pooled and weighted by their inverse variances in die final calibration model.
The olefin concentration can be reported by application of die above model in a multivariate least squares step that determines me olefin concentration and baseline amplitude in a sample spectrum. Despite the fact that die olefin analysis requires scattering data only over a few Raman bands, the multichannel nature of me CCD-detected spectrum provides quality assurance on die validity of die model. The magnitude of the residual difference between the
best fit model and the measured data would be tested for error in excess of the shot noise in die data; excess error and structure in the residuals (as assessed by an autocorrelation analysis) would indicate die presence of an interfering peak diat was not present in die samples used to develop die model; if several olefin bands are used in die analysis, dien alternate models could be applied where a poorly-fit band is dropped from the analysis.
Figure 10 illustrates die Raman spectra of this region (1625 to 1725 cm"1) as the olefin concentration is increased from 0.9% to almost 29%. The areas under these curves can be plotted (see Figure 11) and die data indicate the linear relationship between Raman signal strength and the total olefin concentration. These data can be acquired in seconds, rather than hours as widi existing Florescence Indicator Analysis (F.I. A.) technology currently being used in refineries.
Xvlene Determination
The mree isomers of xylene, meta, ortho, and para, are very valuable as starting materials for petrochemicals. For this reason, diey are often removed from gasoline and sold to chemical processing plants. The concentrations of the three individual xylene isomers are difficult to measure quickly and to separate from other contaminating aromatics, such as benzene, toluene, and ediyl benzene. Xylenes are used in such large volume that their rapid and accurate measurement is essential. Current chromatography techniques require nearly fifteen to twenty minutes to analyze for the relative concentrations of the individual isomers. In large chemical processing plants this analysis is just too slow.
Figure 12 illustrates the individual Raman peaks for the various xylene isomers, m-, o-, and p-xylene in a predistillate fraction rich in xylene. These components can easily be differentiated from the other commonly occurring aromatics, such as emyl-benzene, toluene and benzene. Benzene Concentration Determination
Because benzene is considered a carcinogen, problems surrounding locating its extent, concentration, and safe containment or removal are of utmost concern. EPA Clean Air rules currently in effect require petroleum refiners to reduce benzene levels in gasoline to below 1 %. Traditional process chromatography is normally too slow, requiring 15 to 20 minutes per test. Infrared absoφtion (IR) requires too short of a cell padi lengdi, and Near-IR is not sensitive enough. Recent EPA regulations will force a broad range of industries to cut emissions of benzene that pose cancer risks to workers frequently in contact with petroleum products.
Figure 13 illustrates the Raman spectra for increasing concentrations of benzene in unleaded premium gasoline. Figure 14 illustrates die linear relationship between die area under die Raman benzene peak and the benzene concentration in die gasoline sample.
Fuel Differentiation Application
Figure 15 illustrates the subtraction of two different octane rated gasolines. Since die Raman spectra are different for different fuel formulations, their difference spectra can be used to verify or substantiate tiiat indeed two fuel systems are the same or different from one another. It is not easy for petroleum manufacturers to readily determine if a fuel sample was manufactured by diem or their competitor. The ability to compare complete Raman spectra make it relatively easy to test two different fuels and determine how similar they are to one another.
Figure 16 illustrates mis same subtraction process applied to two samples of fuel that are in fact the same. The subtracted spectrum is basically a flat line. This type of analysis can be used in pipelines to rapidly determine when one company's fuel has stopped flowing and anotiier company's fuel has begun, or when one fuel type has been replaced by anodier. In large transmission lines, many different fuel types and brands are carried togetiier, one right after another. Density measurements are often used to determine when one particular fuel is different from another. The density of diesel is sufficiently different from gasoline that this particular test can be used. However, die density does not change sufficiently between two different brands of gasoline to readily determine the cutoff point. The analysis of cutoff point in a pipe line could easily be determined by Raman analysis according to die present invention.
Figure 17 illustrates die comparison of the subtracted Raman spectra for various brands of gasoline subtracted from a single brand, in this case an Amoco sample. For most of the samples, it was easy to determine that die samples were different, that a Conoco gasoline was different from die Amoco gasoline, etc. Two of the fuels tested have straight lines, indicating diat tiiey are identical to die Amoco fuel. It was verified diat die Texaco and Sinclair fuel samples tested were made by Amoco. Those two refiners purchased tiieir fuel from a local Amoco refinery rather than transporting their fuel from distant facilities. Figure 18 illustrates the Raman spectra for different types of fuel systems, high and low sulfur diesel, Jet A, Jet JP4, and unleaded premium gasoline. It is apparent that their Raman spectra are unique and can be used to identify me type of fuel. Alleviation Acid Concentration Determination
Figure 19 illustrates the Raman spectra for pure sulfuric acid, widi a major Raman peak at 920 cm'1. Figure 20 illustrates this same sulfuric acid Raman peak present in a sample of alkylation acid. The Raman spectra for this and odier sulfuric acid Raman peaks can be used to determine the actual acid concentration in the alkylation acid process. The acid content is currently determined by conventional titration techniques. The area under the sulfuric acid Raman peaks has a linear relationship to actual acid concentration in die alkylation acid process as illustrated in Figure 21.