Optimization of an automated FI-FT-IR procedure for the determination of o-xylene, toluene and ethyl benzene in n-hexane

The development and optimization of an automated flow injection (FI) manifold coupled with a Fourier transform infrared (FT-IR) detector for the determination of toluene, ethyl benzene and o-xylene in an n-hexane matrix is described. FT-IR parameters optimized were resolution and number of co-added scans; FI parameters optimized were type of pump tubing, carrier flow rate and sample volume. ATR and transmission flow cells were compared for the determination of o-xylene, the ATR cell was easier to use and gave better figures of merit, except for sensitivity, for which the transmission cell was twice as good. Multivariate calibration routines were applied to the FI-FT-IR data and the PLS1 algorithm gave relative root mean standard errors of crossvalidation (RRMSECVs) < 7% for all three analytes using mean-centred data and the first derivative for o-xylene.

During this period, there have also been several accounts of the application of multivariate calibration (MVC) routines to FT-IR data (as distinct from NIR data) for quantification. Examples of partial least squares (PLS) regression techniques applied to FT-IR data include: monitoring the epoxidation of indene to indene oxide [22], and the hydrolysis of penicillin V to 6-animopenicillanic acid [23]; the determination of methanol and methyl tertiary butyl ether in gasolines [24]; chlorinated hydrocarbons in water [25], and glucose, maltose and fructose in sugar syrups [26]; and comparison of PLS with multiple linear regression (MLS) and principal components regression (PCR) for quantifying latex in paper coatings [27]. There have, however, been very few publications that have combined the sample handling power of FI with the data treatment capabilities of MVC for organic analysis using FT-IR [1,2]. The combination has excellent potential for the characterization/quantification of organic process streams but does require an integrated and automated system for reliable data acquisition. This paper presents a structured approach to optimizing an automated FI-FT-IR system, with data treatment using MVC techniques. The determination of benzene, toluene and 0-xylene in hexane has been used as an industrially relevant model system to illustrate the approach.

Experimental
Reagents Solutions of 0-xylene (HPLC grade: Sigma-Aldrich, Gillingham, UK), toluene (AnalaR grade: Merck, Darmstadt, Germany) and ethyl benzene (GPR grade: Sigma, Poole, UK) were prepared volumetrically in nhexane (HPLC grade: Rathburn, Walkerburn, Scotland). Univariate calibration standards for 0-xylene, toluene and ethyl benzene (0-20% v/v) were prepared by addition of the appropriate volume of analyte into a volumetric flask (Grade A) and making up to the mark with n-hexane. Standards were prepared in a cool, unheated room to minimize volatilization. For multivariate data analysis, a full factorial design was used where the three analytes were represented at three concentrations (0, 10 and 20% v/v), producing 27 calibration standards.

Instrumentation
The FI manifold consisted of two peristaltic pumps (Ismatec Mini-S 820 for sample flow and Gilson Minipuls for carrier flow) and a low-pressure injection valve (Rheodyne 5020) operated manually or automatically (Universal valve switching module). Pump tubing of 1.3mm i.d. (Ismatec Viton) was used for sample and carrier flow. PTFE tubing of 0.8 mm i.d. was used for all flow lines. A Fourier transform mid-IR spectrometer (Bruker IFS-66) fitted with a closed cycle water-cooled globar source, Ge/KBr beamsplitter (6500-400cm-1), Michelson interferometer with air bearing scanner (maximum resolution 0.25 cm-1) and a mercury cadmium telluride detector (7000-420cm-) were used throughout. The instrument atmosphere was supplied by an air drier (Balston 75-52) at a flow rate of 3.4 min-1. Carbon dioxide was reported as being less than 2 ppm. One of two liquid cells was used to present the sample to the spectrometer: a Squarecol microflow through ZnSe ATR 330 gl cell (Graseby Specac); or a microflow through transmittance cell (Graseby Specac) equipped with KBr windows and 0.05 mm Pb spacer.
The ATR cell was used unless otherwise stated. Spectra were acquired using the instrument operating software (Bruker OPUS/IR-2 extended analytical package) with additional application packages (Bruker OPUS/ CHROM chromatographic coupling package and OPUS/3D data processing package) required for monitoring samples introduced via the FI manilbld.
Post-experimental data processing was performed using OPUS/IR-2, spreadsheets (Microsoft Excel v5.0, v7.0 and MicroCal Origin v3.78) and a multivariate data analysis package (Camo Unscrambler v6.1). Automation routines were written within OPUS/IR-2, and external apparatus control was performed via transistor-transistor logic (TTL) signals sent from the spectrometer via a custom-built interface. For automated data collection [1], a macro program was initiated and samples introduced to the FI manifold automatically. The macro was used to control external apparatus by switching one of four TTL lines to 0 V or + 5 V according to the device required. The carrier pump, sample pump and injection valve were switched on and off via an in-house built TTL to mains converter. A direct TTL signal was used to actuate the valve between the load and inject positions during analysis.
For automated measurements, a macro was written to coordinate sample injection with spectral acquisition. Upon initiation of the macro, the FI manitbld was flushed for 20s and a background spectrum of 50 coadded scans acquired, aher which the sample was injected into the carrier stream. After a delay of 18s, a sample spe'trtm of 50 co-added scans was acquired and integrated. This enabled a spectrum of optimum signalto-noise ratio to be acquired whilst the central portion of the injected sample passed through the flow cell. To determine the time delay required before sample injection, five repeat injections of a 5% v/v 0-xylene standard were performed using the procedure for continuous measurement. The delay was calculated as (time from injection to maximum Flagram response)-(half of total measurement time).
Each injection and measurement was replicated three times, background corrected against a 100% v/v n-hexane blank, and averaged to produce a mean spectrum. No smoothing was applied to raw spectral data and all samples were measured in a randomized order to reduce the risk of systematic error.

Procedures
For optimization of the FI-FT-IR manifold, the basic FT-IR parameters used were single-sided fast return acquisition of the interferogram with Blackman Harris 3 Term apodization and Mertz phase correction. The zero filling factor was set to 2 with a phase resolution of 32, and the aperture was opened to the maximum allowed without saturating the detector. Prior to acquiring a sample spectrum, a background spectrum of the nhexane carrier was acquired. Two measurement modes were employed: (i) continuous--where the throughput of the FI manifold was constantly monitored; and (ii) automated--where discrete spectra were acquired only when the injected sample was present in the flow cell.
For continuous measurement, OPUS/CHROM software was used to continuously acquire spectra and display them as a three-dimensional output (absorbance and wavenumber as a function of time) for the duration of the experiment. Quantitative data were acquired by integrating the absorbance profile of a C-H bending mode of each analyte (0-xylene--741cm -, toluene 728 cm-1, ethyl benzene 698 cm-), and plotting the integration value as a function of time to generate a FIagram. Continuous measurement was used to estimate the optimized parameters for automated measurements.

Choice of pump tubing
There are a number of materials available for use as pump tubes in peristaltic pumps. The choice of material is primarily dictated by operating life and chemical resistance to the carrier stream. Small samples of five types of commercially available pump tubing (Ismaprene Silicone Viton Tygon and Tygon Yellow were immersed in 10ml of n-hexane in aluminiumcapped bottles for 72 h and tested for changes in internal/external diameter and hardness (using the International Rubber Hardness Degrees scale (IRHD)) (table 1). Silicone was found to expand and soften in nhexane to give an IRHD value below the limit of detection of the instrument. Ismaprene (R) and Tygon (R) contracted, with the former showing a 25% reduction in hardness and the latter hardening beyond the upper measurable limit. These three pump tubing materials were therefore not suitable for pumping n-hexane. The remaining materials, Tygon Yellow and Viton were further assessed for suitability by monitoring their flow rates over 70 h of continuous pumping using an Ismatec Mini-S 820 peristaltic pump. The results showed that long-term deployment of Tygon Yellow (R) was not feasible as the pump tube walls collapsed within 5h. Viton pumped n-hexane at a constant rate for 70 h and was therefore selected as the ideal pump tubing for transferring n-hexane matrices in all subsequent experiments.

Effect of instrument resolution
The spectra of 5% v/v solutions of the individual components recorded at a resolution of 2cm -1 showed (figure 11 the C-H deformation bands for ethyl benzene (698cm-), 0-xylene (741 cm-l) and toluene (728cm--1).
The spectra of a 5% v]v mixture recorded at resolutions of 2, 4, 8, 16 and 32cm -1 are shown in figure 2. At a resolution of 2 cm-1, the three deformation bands were well resolved and there was some low-level spectral detail in the window 1150-1000cm-l. Reducing the resolution to 4cmdecreased the separation between the major 0xylene (741 cm-1) and toluene (728cm-l) bands, but the combined ethyl benzene and toluene band (696cm-l) was still clearly separated. Reducing the resolution further to 8, 16 and 32cm -l degraded the quality of spectral resolution to the point where, at 32cm-1, all three deformation bands were indistinguishable. A resolution of 4cmwas subjectively estimated to provide adequate spectral information for the purposes of calibration.
In terms of measurement time, a resolution of 4 cm -l was much faster (8.1 s) than a resolution of 2 cm -(14.2 s) in acquiring a spectrum of eight co-added scans at a modulation frequency of 10.0 kHz. Increasing the modulation frequency to 160.0 kHz reduced the measurement time considerably to 0.6 and 0.9 s for 4 cm -1 and 2 cm--1, respectively. A resolution of 4cmand a modulation frequency of 160kHz were therefore used in all subsequent experiments to maximize the number of coadded scans acquired during a measurement event. To assess spectral noise as a function of the number of coadded scans, three replicate spectra of 5% v/v mixture solutions in n-hexane were acquired by co-adding 1, 10, 50 100, 250, 500 and 1000 scans at a resolution of4cm-l. Maximum peak-to-peak noise was determined in the range 2100-2000cm-where there was no absorbance from the analytes, leaving a signal due only to fluctuations in the instrument and flow rate of the carrier stream. As expected, noise and reproducibility (shown as + one standard deviation) decreased with increasing number of co-added scans, as shown in figure 3. No significant reduction in noise was observed with increasing the number of co-added scans above 10. Fifty coadded scans gave an RSD of 3.2% and this was selected as the optimum condition for discrete spectra.
To study the effect of instrument resolution on spectral noise, three replicate 50 scan spectra of 5  with coarser resolution (figure 4). A resolution of 4cmwas therefore selected as the optimum condition.

Effect of FI parameters
In FI with spectrophotometric detection of chemically derivatized analytes decreasing the flow rate decreases sample dispersion and increases residence time in the FI manifold, which improves sensitivity for slow reactions.
With FT-IR detection of organic species, however, derivatization is not normally required. In addition, the residence time of the sample in the beam path directly affects the number of sample scans that can be acquired, which in turn affects the signal-to-noise ratio. Flow rate should theretbre be optimized so that the sample is resident in the flow cell long enough to achieve the required signal-to-noise ratio.
Five replicate 20gl injections of a 5% v/v 0-xylene standard were manually injected into an n-hexane carrier stream at flow rates of 0.5, 1.0, 1.5 and 2.0 ml min-. The response was monitored continuously based on measurement of the peak height of the 0-xylene band at 741 cm-. Spectra of eight co-added scans were continuously acquired, and all sample spectra were ratioed against a background spectrum of 100 co-added scans of 100% v/v n-hexane. The reproducibilities (n 5) were in the range 0.4-1.7% and there was no significant difference in signal response with varying carrier flow rate for a 20gl sample volume. This is contrary to expectation for transmission cells [28,29] and was due to the small volume of the injected sample (20 gl) relative to that of the ATR flow cell (330gl). Hence, a carrier flow rate of 2.0 ml minwas used to maximize throughput.
Five replicates of 25, 100, 150, 200 and 300 gl volumes of a 5% v]v 0-xylene standard were manually injected into an n-hexane carrier stream pumped at a flow rate of 2.0 ml min-. The response increased linearly with sample loop volume, but the larger volumes took longer to return to the baseline. The increase in response was also due to the large volume of the ATR flow cell compared with the range of sample volumes studied, i.e. larger sample volumes covered a larger area of the ATR element. The optimum sample volume selected was 220 gl which was a compromise between sensitivity and fast return to baseline.

FI-FT-IR calibration for toluene, ethyl benzene and o-xylene
With the data acquisition and FI parameters sequentially optimized, the analytical performance of the FI-FT-IR system was assessed by examining the calibration data for the analytes in n-hexane. Five replicate injections of 2001al of a range of toluene (0.04-20% v/v), ethyl benzene (0.04-20% v/v) and 0-xylene standards (0.05-20% v/v) were manually injected into an n-hexane carrier flow stream at a flow rate of 2.0 ml min-. The response was measured continuously as described above.
The linear calibration parameters are given in table 2 with limits of detection (LOD) defined as three times the standard deviation of the lowest standard because blank injections could not be detected on the carrier baseline.
The large relative standard errors (13-18%) observed at low concentrations could be minimized by co-adding a higher number of scans to improve the signal-to-noise ratio. However, increasing the number of scans in timeresolved measurements limited the temporal resolution of the FIagram response. The preferred approach to solving this problem is therefore to automate the FI-FT-IR manifold. A further advantage of automation is that computer memory consumption is significantly reduced; a time-resolved measurement takes up to 5 Mb of hard disk space, whereas discrete single spectrum measurements require only 10 kb.
Automated FI-FT-IR calibration for toluene, ethyl benzene and o-xylene A macro was written to control external apparatus via a 50 pin serial D connector and a 5 V TTL signal. Five replicates of 200 gl of a range of toluene, ethyl benzene and 0-xylene standards (0-20% v/v) were injected into an n-hexane carrier stream pumped at 2.0 ml min-. All sample spectra were ratioed against a background spectrum of 500 co-added scans of the carrier stream, 100% v/v n-hexane. A higher degree of repeatability (< 7% error) was attained for standards of concentrations greater than 0.05% v/v with the automated manifold, demonstrating that repeatability improved compared to the manual approach. The calibration results are shown in table 2, and the limit of detection of 0.004% v/v for 0xylene in n-hexane compares favourably with previously reported values of 0.03% v/v [14] and 0.01% v/v [15] using a manual FI-FT-IR manifold equipped with a transmission flow cell.

Comparison of A TR and transmittance flow cells
The two alternatives for presenting samples to an IR spectrometer from an FI manifold are ATR and transmittance flow cells. Previous results [2] have shown that whilst transmittance flow cells have a higher throughput of source energy than ATR cells, they are difficult to assemble and maintain. ATR flow cells are more mechanically robust and optically stable due to their fixed path lengths. A micro flow-through transmittance cell was incorporated into the FI manifold in place of the ATR cell, and the manifold optimized for flow rate and sample loop volume. Five replicates of 20 gl of a 5% v/v 0-xylene standard were manually injected into an nhexane carrier stream at flow rates of 0.5, 1.0, 1.5 and 2.0 ml min--. In addition, five replicates of 25, 100, 150, 200 and 300lal of a 5% v/v 0-xylene standard were manually injected into an n-hexane carrier stream at a flow rate of 0.5 mlmin-. The response in both experiments was measured continuously. Absorbance was inversely related to flow rate, with a maximum mean observed signal of 0.31 AU at 0.5 ml min-. The repeatability error of sample response (n--5) was less than 0.7% in all instances. The observation of the maximum response occurring at the minimum carrier flow rate was expected as sample dilution due to axial dispersion in the FI manifold was minimized. This is in contrast to the flow through ATR cell where flow rate had no significant effect on sample response. Overall, the ATR cell was the more user-friendly accessory, being easier to dismantle, clean and re-assemble. The transmittance cell suffered from leaks if the entrance and exit ports were not correctly aligned with the flow-through KBr chamber (alignment was best achieved with the insertion of a thin wire guide through the ports prior to tightening). Optical alignment is crucial with most ATR accessories for maximizing sensitivity. Adjustments have to be made in all three dimensions and fine-tuning performed until no further gains can be made in signal throughput, which is time consuming and labour intensive. In contrast, the transmittance cell requires the minimum of alignment. ATR cells are highly reproducible once aligned due to their fixed path length, but reproducibility in transmittance cells is difficult to achieve as the path length changes with the tension applied to the fastening nuts of the demountable cell.
The ATR cell used in this investigation suffered from poor flushing efficiency due to the orientation of the ZnSe crystal and the entrance and exit ports. No such effect was observed in the simpler transmittance cell. The choice of flow cell is ultimately dependent on the sample matrix, e.g. acid media cannot be used with ZnSe ATR elements due to the production of toxic gaseous H2Se.
For analytical purposes the ATR cell is preferred due to better reproducibility and ease of use.   for ethyl benzene were observed for three PC models based on mean centring only. The best 0-xylene model required second derivatization and only two PCs to produce an RRMSECV of 4.0%. The results show that rapid FT-IR data acquisition of liquid samples is possible through the use of FI, and that the results can be used for quantitative prediction via univariate and multivariate calibration.

Conclusions
The optimized and automated FI manifold provided a rapid and reproducible means of liquid sample intro-duction to FT-IR. The manifold used for the determination of toluene, ethyl benzene and 0-xylene in n-hexane was optimized for the following parameters: pump tubing (Viton), resolution (4cm-1), number of co-added scans (50), modulation frequency (160kHz), flow rate (2 ml min -1 for the ATR cell, 0.5 ml minfor the transmission cell) and sample loop size (200tl). Each of the analytes was univariately calibrated in binary mixtures (0-20% v/v) with n-hexane. Limits of detection were 0.014% v/v for toluene, 0.010% v/v for ethyl benzene and 0.004% v/v for 0-xylene, and the error of repeatability was < 7% for concentrations greater than 0.05% v/v. The analytical performance of an ATR and transmission flow cells were compared using an 0-xylene calibration, and the ATR cell was more user-friendly and analytically superior, except for sensitivity, which was approximately half that of the transmission cell.
The use of PLS1 for the multivariate calibration of mixture spectra of toluene, ethyl benzene and 0-xylene in n-hexane acquired by automated FI-FT-IR resulted in RRMSECVs of <8%. Derivatization was required to enhance the resolution of the acquired spectra, at the cost of much poorer signal-to-noise ratio, for the prediction of 0-xylene.