Conventional infrared wavelength generation technologies such as Non-Dispersive Infrared (NDIR) spectrometry and Fourier Transform Infrared (FTIR) spectrometry present significant downfalls when it comes to the measurement of hydrocarbon gases in the process environment. NDIR spectrometry is susceptible to false positives from spectrally overlapping species and is not capable of achieving simultaneous characterisation of more than one chemical compound. FTIR spectrometry, on the other hand, provides high-resolution results but it lacks ruggedness whilst being rather expensive.

In contrast, Encoded Photometric Infrared (EP-IR) spectrometry is capable of simultaneously resolving several chemical compounds even if they share a similar chemical composition and subsequent spectroscopic features.

This article demonstrates the capability of EP-IR spectrometry to provide adequate spectral information to model the simultaneous measurement of methane, ethane and propane in conditions mimicking the presence of ethane and propane in pure stream of methane. An Aspectrics MultiComponent5000 analyser was used for this analysis. Its specifications are:

• 128 photometric channels.


• Nominal spectral range: 2.5–5.0microns (4000-2000cm-1).


• 100 scans per second.


• 60-second RMS noise @ maximum energy (~3.8microns) less than 20 µAU (using an Aspectrics Verifier IR photometric testing fixture).

The analyser contains patented encoder disk technology that enables the encoding of full spectral information 100 times per second, resulting in 10milli-second true refresh rate of the chemical information. Moreover, dedicated on-board processors allow users to monitor processes in real time, even as often as 100 times per second. Quantitative analysis software enables real-time characterisation of multiple chemicals in process streams. The software also communicates analytical results and instrument health information to central network locations, and facilitates full automation of calibration and calibration maintenance in production mode.

Featuring a pathlength of 2.4metres and an internal volume of 100 mL, excluding transfer lines, Aspectrics’s multipass I-Cell gas cell was also used in this experiment. Cell materials include calcium fluoride windows and gold-coated mirrors. No temperature control was needed on this model and all measurements were completed at room temperature.

All data were collected using the Aspectrics Commander data collection software package. A boxcar setting of 6000 ensured that each spectrum collected results from the normalised co-addition of the previous 5999 spectra and itself. With the scanning speed of 100 scans per second of the MultiComponent5000 analyser, this resulted into an actual integration time of 60seconds.

A down sample setting of 600 was adopted, meaning that only one in every 600 scans generated was actually saved. This corresponds to the response rate of the analyser of one new spectrum every

10 seconds. In order to account for the time for the sample to fill the gas cell, a waiting time of 200–400 seconds was set between the collection of each sample in order to allow for stabilisation of the gas mixture concentration in the gas cell and avoid sample carry-over contamination.

In order to gather sufficient numbers of calibration, validation and prediction data points, around 100 spectra were collected for each set of gas mixtures, requiring approximately 10 minutes for each gas mixture. The Commander software package allows users to control both the spectrometer for data collection and the Environics4000 gas blender for delivery of samples.

Seven calibration standards were used, consisting of certified mixtures of methane, ethane and propane. Concentrations for ethane and propane ranged from

0 to 200ppm (180ppm effective after gas blender delivery due to Mass Flow Controller (MFC) limitation) and 0 to 150ppm (135ppm effective), respectively. Five data points were created using orthogonal design in order to describe maximum variation in ethane and propane concentrations with a minimum number of standards. Concentrations of methane were the complement to 1000ppm (900ppm effective) in order to mimic a 1:1000 dilution of 0–20percent ethane and 0–15percent propane in a stream of pure methane. Spectra of monoblend methane, ethane and propane diluted in dry nitrogen were also added to the set of calibration standards.

The use of certified mixtures of methane, ethane and propane allowed the use of only one MFC at a time. This was of particular importance as the precision of the reference method for this calibration was that of the MFC of the gas blender, namely

one per cent relative to the set concentration point according to Environics published specifications.

Five validation standards were used consisting of certified mixtures of methane, ethane and propane. Concentrations for ethane and propane were carefully selected to be as different as possible from those in the set of calibration standards in order to test instrumental response linearity as well as accuracy in validation. Concentration values ranged from 46 to 134ppm for ethane and from 33 to 103ppm for propane, while maintaining an orthogonal design to avoid co-linearity. Concentrations of methane were the complement to 1000ppm (900ppm effective) in order to mimic a 1:1000 dilution of 4.6–13.4percent ethane and 3.3–10.3percent propane in a stream of pure methane.

Just as in the case of the calibration standards, the use of certified mixtures of methane, ethane and propane required only one MFC to be used at a time.

Twenty-three prediction samples were used, divided into two sets: firstly, a set of 18 samples with all concentrations in methane, ethane and propane within the range for which the EP-IR had been calibrated; and secondly a set of five samples in which the methane concentration was deliberately set outside the range for which the instrument had been calibrated, while maintaining ethane and propane concentrations within the calibration concentration ranges.

Initially, it was decided to generate random mixtures of methane, ethane and propane using three mono-blend tanks and a tank of dry nitrogen as solvent. However, such set-up would have required the use of four MFCs, each of which has a built-in precision of one per cent of the set point. In order to reduce the error on the reference measurement of each prediction blender sample, only one MFC for all zeros and only two MFCs for the random mixture were used. This was accomplished by using one of the certified mixture tanks as a solvent and a second MFC, allowing blending in some mono-blend gas.

The use of the above protocol allowed using no more than two MFCs at a time, each of which had a manufacturer specified precision of onepercent relative to the set concentration point. Considering the additive nature of errors of the measurement, the precision of concentrations delivered by the gas blender was therefore kept at ±2percent relative to the set point at most.

The company’s Chemobuilder chemometrics software package was used to develop, optimise and validate the calibration equations as well as to predict the concentrations in the three gases in the gas blender mixtures. Data treatment was completed in four distinct steps. Firstly, the starter calibration was developed establishing analytical performance baseline and identifying/selecting spectral regions of interest. Then a full calibration model was developed creating a PCR-based (Principal Component Regression) quantitative model for the simultaneous measurement of methane, ethane and propane in a mixture without a false positive.

The full calibration model was then validated by realising the (RM) SEP accuracy of measurement as well as confirming Limits Of Determination (LODs) and instrumental response linearity within the calibration range.

The final step was split into two different scenarios which were adopted. The first scenario was that all concentrations of methane, ethane and propane in the mixtures were within the range of concentration for which the instrument was calibrated and the second scenario showed some concentrations of methane, ethane and propane in the mixtures were not within the range of concentration for which the instrument was calibrated.

A Principal Component Analysis/Regression (PCA/PCR) quantitative algorithm was applied onto automated polynomial baseline corrected absorbance spectra using dry nitrogen spectra as background.

Results

The EP-IRMC5000A delivered excellent analytical results in calibration and validation (Table1 and 2) for the simultaneous measurement of methane, ethane and propane in gas phase. Most impressive was the precision values obtained for all measurements, whether analysing calibration, validation or gas blender prediction samples. Accurate calculations were dependant on the precision of the MFCs used in the gas blender.

The experiment demonstrated that the chemometrics methods are accurate. Using an orthogonal experimental design and certified gas mixtures to minimise reference method precision errors, a very accurate model was developed. The characterisation was by 1*s (RM)SEC for all compounds of less than 0.32percent relative to the maximum concentration in the range of calibration standards, and by 1*s (RM)SEP for all compounds of less than 0.36percent relative to the maximum concentration in the range of validation standards.

Bertrand S Lanher is Director of Applied Science & Technology, Aspectrics Inc For more information, visit www.aspectrics.com