 |
||||||
 |
 |
 |
 |
|||
 |
||||||
|
Petroleum
|
|||
|
Raman Spectroscopy Defined (Click picture for a high quality version) |
Olefin Monitoring |
||
Process Instruments Inc. manufactures an optical fiber-based Raman scattering instrument for real-time, on-line analysis of finished gasoline, gasoline component streams, diesel fuel, jet fuel, and kerosene. Raman analysis is well suited to gasoline blending for several reasons: 1.) Raman scattering requires no sample conditioning, 2.) The high resolution Raman spectra of gasoline allow the various chemical components to be distinguished and quantified, 3.) Working in the NIR region allows for fiber-optic coupling of the sample stream probe and the Raman analyzer, and 4.) Working with optical fibers allows for multiplexing many individual Raman probes to one single instrument, reducing instrument calibration and simplifying overall maintenance. Continuous in-situ monitoring of the component or final product streams can provide almost instantaneous feedback on important changes in the blending process which can lead to real-time process control, improved finished product quality, reduced overall cost, and minimized waste. |
|||
|
|||
Raman scattering is similar to infrared absorption in that they are both vibrational spectroscopies that probe the discrete vibrations of molecules and molecular bonds. The Raman effect relies on the scattering of the incident photons, whereas infrared occurs through light absorption. These vibrational techniques have different selection rules for the types of molecular vibrations that can be observed, and so they are often considered complementary. The figure to the right displays a typical on-line Raman spectrum of a gasoline stream with the different peaks from many of the major chemical components indicated. The peak frequency shift (i.e. the location along the X-axis) yields the sample composition, and the peak intensity yields the concentration of that particular component. A key advantage with the Raman effect is that it is a linear process. Therefore, a sample’s Raman band for an individual molecule will be twice as intense when the sample contains twice as many of those molecules. Using the on-line spectra, chemometric data analysis models can be created to determine the different sample parameters. A model is created by correlating the on-line spectral response to laboratory data, and generally this approach requires fewer than 100 samples to create a robust model. |
 |
||
|
|||
The figure on the right compares NIR gasoline data and data obtained with our Raman analyzer. NIR has very broad bands arising from overtones and harmonics that contain only subtle differences between different gasoline blends whereas Raman spectroscopy provides excellent spectral resolution of fundamental vibrations allowing for minimum component overlap and maximum component specificity. Raman has the advantage over infrared in that there is no interference from water so the probe can be directly inserted into a process stream. Infrared requires a sample conditioner. A single Raman system can be multiplexed up to 18 channels/streams. Because of the poor resolution of NIR analyzers, it is difficult to transfer chemometric models between instruments. However, models transfer easily with our Raman instruments and so additional models for new blends or components can be developed on our laboratory instrument and then be directly loaded onto our on-line process Raman analyzer. |
 |
||
|
|||
|
A schematic representation of the components of our Raman analyzer system is displayed. We use our frequency-stabilized, high power laser (Patent No. 6,100,975) for sample excitation, and the laser can be configured for specific applications to operate anywhere in the wavelength range from 640 to 850 nm. The laser output is then coupled into an excitation fiber that can be up to 200 meters (650 ft) in length. Our Raman Probe (described in more detail below) is located at the other end of the excitation fiber. Sample excitation and collection occurs from the distal end of the Raman probe, and the Raman signal is coupled into a collection fiber where it is transmitted to our spectrograph (Patent No. 6,028,667). Detection of the individual Raman frequencies is preformed via a CCD array. Our PROspectTM software generates the Raman predictions from the acquired spectra and also controls the entire electronic interface connections used to operate the Raman analyzer. One advantage with this configuration is that we are only transmitting light to and from the sample and do NOT utilize any electrical power at the sample probe and so the probes are intrinsically safe.
|
 |
||
|
- Reduce Octane Give Away |
 |
||
|
|||
|
|
 |
||
Our multiplexer (Patent No. 6,859,581) is designed to interface a single Raman instrument to 18 individual Raman probes located at a different process locations or sites. The instrument has a fast response, minimal signal loss, and analyzes each channel/site in 60 to 120 seconds. This allows measuring all of the 18 different locations and feeding forward the on-line analysis to the controller every ~ 30 minutes. Multiplexing many individual Raman probes to one single Raman instrument reduces instrument calibration and simplifies the overall maintenance. Our PROspectTM software is designed to automatically load and use unique models for each of the different multiplexer channels. The multiplexer can also be set up to provide an ASTM validation channel (E1655 & D6122) that is used to monitor the overall system performance. |
 |
||
|
|||
The on-line immersion Raman probe interfaces directly into the process stream allowing one to monitor directly at the process conditions. The probe is intrinsically safe since it does not require any electrical connections, and only light is transmitted via fiber-optic cables to and from the probe. Since our instrument works in the NIR region, the fiber-optic cables can be up to 200 meters (650 ft) in length for sampling streams located at a remote location in the refinery. For comparison MID IR instruments are limited to fibers that can only be a few meters in length. The probes can be manufactured from a variety of materials including 316-SS, carbon steel, alloy-20, hastelloy, etc. The probes are also designed to be retractable for easy access and cleaning in low pressure applications (< 550 psi). High pressure applications require a welded ANSI flange seal. Sample excitation and collection are conducted through a sapphire window located on the immersed end of the probe. The probes are rated up to 80 OC and 550 psi. Optional high temperature and pressure probes are also available. |
 |
||
The in-line Raman probe and flow cell is used to monitor a slip-stream sample line. The Raman probe is protected from the sample stream by a 1 mm thick sapphire window that is rated up to 550 psi. Typical flow rates are 20 to 50 cc/min. The sample cell has minimal dead space and can be easily cleaned if necessary. Conventional stainless steel compression fittings allow quick connection with a conventional refinery sampling stream. Just like the immersion Raman probes the in-line unit is intrinsically safe with no electrical connections and can be used with fiber-optic cables that are up to 200 meters (650 ft) in length. The in-line probe is encased in a NEMA 12 or NEMA 4x enclosure that protects it from the outside environment. |
 |
||
|
|||
|
|
 |
||
Our 42 station auto sampler allows refinery lab personnel to collect spectra and model predictions automatically from multiple samples located in 40 ml scintillation vials that have an open top screw cap with septa seals. The autosampler uses an internal small air compressor to pressurize the vial and push the sample through the sample flow cell. Continuous positive sample pressure ensures that any volatile components remain in the sample during the Raman measurement cycle. Our PROspectTM software allows the user to specify the type of sample in each vial so that the appropriate Raman prediction model is used for each sample and the parameter predictions are output to a refinery’s LIMS system. The main advantage with the autosampler is that lab personnel are not required to conduct each measurement individually and would be available for other tasks. |
 |
||
|
|||
| Typical on-line results for different gasoline parameters from a model that was built with < 150 samples are displayed to the right. The minimum and maximum parameter ranges are shown in the middle column. The final column is the standard deviations between the laboratory minus the Raman predicted values. Even with a model containing a limited number of samples, the Raman predicted values are nearly equal to the standard laboratory measurements. |
 |
||
|
|||
| Butane is often added to gasoline to increase its Reid vapor pressure (RVP). The bottom spectrum is from butane. Spectral differences between a summer (low RVP) and winter (high RVP) blend of gasoline are displayed as the top two spectra. The winter blend has more butane. Our Raman analyzer measures the concentrations of the different components and the PROspectTM software then uses a chemometric model to predict the on-line RVP corresponding to each collected spectrum. TVP (True Vapor Pressure) can also be measured. |
 |
||
| Typical on-line partial least squares (PLS) analysis of the Reid vapor pressure. The plot shows Raman predicted values vs. the measured laboratory values. The standard error of prediction for the PLS model with outliers removed was 0.187. |
 |
||
| A histogram of the distribution of differences between the laboratory and Raman predictions. |
 |
||
|
|||
| Spectra of a regular and premium gasoline comparing the concentrations of various octane rich components. The premium gasoline has more toluene and iso-octane (high octane components) while the regular has more ethyl-benzene (low octane components) and very little iso-octane. |
 |
||
| Typical on-line partial least squares (PLS) analysis of the research octane number (RON). The plot shows Raman predicted values vs. the measured laboratory values. The standard error of prediction for the PLS model with outliers removed was 0.252. |  |
||
|
A histogram of the distribution of differences between the laboratory and Raman predictions. |
 |
||
| Typical on-line partial least squares (PLS) analysis of the motor octane number (MON). The plot shows Raman predicted values vs. the measured laboratory values. The standard error of prediction for the PLS model with outliers removed was 0.264. |  |
||
| A histogram of the distribution of differences between the laboratory and Raman predictions. |
 |
||
| Typical on-line partial least squares (PLS) analysis of the road octane number (ROAD). The plot shows Raman predicted values vs. the measured laboratory values. The standard error of prediction for the PLS model with outliers removed was 0.197. |  |
||
|
A histogram of the distribution of differences between the laboratory and Raman predictions. |
 |
||
|
|||
|
Typical on-line partial least squares (PLS) analysis of the 10% distillation temperature (10%). The plot shows Raman predicted values vs. the measured laboratory values. The standard error of prediction for the PLS model with outliers removed was 2.07. |
 |
||
|
A histogram of the distribution of differences between the laboratory and Raman predictions. |
 |
||
|
Typical on-line partial least squares (PLS) analysis of the 90% distillation temperature (90%). The plot shows Raman predicted values vs. the measured laboratory values. The standard error of prediction for the PLS model with outliers removed was 2.05. |
 |
||
|
A histogram of the distribution of differences between the laboratory and Raman predictions. |
 |
||
|
|||
|
Typical on-line partial least squares (PLS) analysis of the vapor / liquid ratio (V/L). The plot shows Raman predicted values vs. the measured laboratory values. The standard error of prediction for the PLS model with outliers removed was 1.09. |
 |
||
| A histogram of the distribution of differences between the laboratory and Raman predictions. |
 |
||
|
|||
|
Spectra of regular and premium gasoline comparing the benzene concentrations. The benzene peak is observed at ~ 992 cm-1 and the regular gasoline has more benzene (2.13%) compared to the premium (1.1%). The spectra also show the 1003 cm-1 toluene peak which is at a higher concentration in the premium blend. |
 |
||
| Typical on-line partial least squares (PLS) analysis of the percent benzene concentration (BENZ). The plot shows Raman predicted values vs. the measured laboratory values. The standard error of prediction for the PLS model with outliers removed was 0.033%. |
 |
||
|
A histogram of the distribution of differences between the laboratory and Raman predictions. |
 |
||
|
|||
| Spectra of regular and premium gasoline comparing the olefin concentrations. The main olefin peaks are observed in the 1640 to 1680 cm-1 region, and the regular gasoline has more olefin (12.6%) compared to the premium (0.8%). |
 |
||
| Typical on-line partial least squares (PLS) analysis of the percent olefin concentration. The plot shows Raman predicted values vs. the measured laboratory values. The standard error of prediction for the PLS model with outliers removed was 0.23%. |
 |
||
|
A histogram of the distribution of differences between the laboratory and Raman predictions. |
 |
||
|
|||
|
Typical on-line partial least squares (PLS) analysis of the api gravity (API). The plot shows Raman predicted values vs. the measured laboratory values. The standard error of prediction for the PLS model with outliers removed was 0.09%. |
 |
||
|
A histogram of the distribution of differences between the laboratory and Raman predictions |
 |
||
|
|||
|
Typical on-line partial least squares (PLS) analysis of the sulfur concentrations. The plot shows Raman predicted values vs. the measured laboratory values. The standard error of prediction for the PLS model with outliers removed was 5.7 ppm. |
 |
||
| A histogram of the distribution of differences between the laboratory and Raman predictions. |
 |
||
| A Ad-Hoc trend of total sulfur measurements comparing On-line Raman with On-line UV Fluorescence. Excellent agreement is observed with the two different techniques. |  |
||
|
|||
|
Spectra of a low-sulfur and an ultra-low diesel fuel. The down arrows in the figure indicate several distinct low sulfur spectral bands that are not apparent in the ultra low spectrum. |
 |
||
| Typical on-line results for a low sulfur diesel stream. The Raman predicted values accurately matched standard laboratory measurements even though some of the models contained a limited number of samples (last column). |
 |
||
|
|||
| Our instruments have been used for monitoring blending component streams in gasoline refineries to avoid the production of off-spec gasoline. The measured parameters for a naphtha stream are displayed to the right. The low standard deviations between the laboratory minus the Raman predicted values show that our Raman analyzer can reliably model critical parameters. |
 |
||
| The measured parameters for a stabilized bottoms stream are displayed to the right. The low standard deviations between the laboratory minus the Raman predicted values show that our Raman analyzer can reliably model critical parameters. |
 |
||
|
The measured parameters for a reformate stream are displayed to the right (Slide 1 of 2). The low standard deviations between the laboratory minus the Raman predicted values show that our Raman analyzer can reliably model critical parameters. |
 |
||
|
The measured parameters for a reformate stream are displayed to the right (Slide 2 of 2). The low standard deviations between the laboratory minus the Raman predicted values show that our Raman analyzer can reliably model critical parameters. |
 |
||
|
|||
|
Our Raman analyzer has been used to monitor different input and output hydrotreater streams. A schematic representation is displayed to the right. In the next four hydrotreater stream figures the low standard deviations between the laboratory minus the Raman predicted values show that our Raman analyzer can reliably model critical parameters. |
 |
||
|
Measured parameters from the input and output feed of a heavy FCC stream through a hydrotreater (Slide 1 of 2). |
 |
||
|
Measured parameters from the input and output feed of a heavy FCC stream through a hydrotreater (Slide 2 of 2). |
 |
||
|
Measured parameters from the input and output feed of a light FCC stream through a hydrotreater (Slide 1 of 2). |
 |
||
| Measured parameters from the input and output feed of a light FCC stream through a hydrotreater (Slide 2 of 2). |
 |
||
|
|||
|
Spectrum of a typical aviation fuel, Jet-A. |
 |
||
|
|||
|
|
 |
||
|
|
 |
||
|
|||
| Our frequency-stabilized, high power diode laser (Patent No. 6,100,975) has an estimated lifetime of 6 to 7 years continuous operation. To minimize system downtime, a secondary backup laser is built into each industrial system that will become active if any problems are ever detected with the primary laser. |
 |
||
|
|||
|
The four-week study looked at two different 50/50 toluene-acetonitrile solutions on two separate multiplexer channels. Determination of the mean, FWHM, and amplitude of a Raman band was accomplished by fitting each peak to a Gaussian/Lorentzian Equation. The figure to the right displays the fitted mean for toluene’s two strongest Raman bands at 786 and 1003.5 cm-1 during the four week stability experiment. Each of the fitted means had a standard deviation (SDV) of ~ 0.015 cm-1, and this demonstrates that our instrument has the ability to distinguish small spectral band shifts down to a few hundredths of a wavenumber. |
 |
||
|
|||
|
The figure to the right shows different spectral regions of a gasoline spectrum obtained at three different temperatures (0, 23, and 48 C). All of the Raman peak positions and amplitudes remain constant for the different temperatures. This demonstrates that Raman measurements are nearly independent of the sample temperature. Predictions from an on-line Raman probe in an exterior process stream will not be altered by changes in the outside environment’s temperature. The temperature independence is another advantage when comparing IR vs. Raman since IR spectroscopy varies more significantly with the sample temperature |
 |
||
|
|||
| To illustrate our Raman analyzer's precision, data were obtained from an on-line Raman analyzer performance channel over a three month period where each parameter’s mean and standard deviation (using 6,700 predictions) are displayed to the right. The standard deviations from each parameter are consistently two to three orders of magnitudes lower then the mean value. |
 |
||
|
|
 |
||