Anal. Chem. 2008, 80, 1679-1688
Turbulence as a Source of Excessive Baseline Noise during High-Speed Isocratic and Gradient Separations Using Absorption Detection Deirdre Cabooter,† Fre´de´ric Lynen,‡ Pat Sandra,‡ and Gert Desmet*,†
Vrije Universiteit Brussel, Department of Chemical Engineering (CHIS-TW), and Pfizer Analytical Research Center, Universiteit Gent, Krijgslaan 281 S4-Bis, 9000 Gent, Belgium
The occurrence of turbulent flow conditions in the connection tubing or at the entrance of the detection flow cell has been identified as a potential source of excessive detector noise in the signal of the diode array UVabsorption detector flow cell of a commercial HPLC system used to perform high-speed separations on widebore columns (4.6 mm i.d.). This excessive noise was found to occur abruptly if the flow rate is increased beyond a certain critical value or during gradient runs, if the mobile phase viscosity falls below a certain critical value. Several detector cells of different dimensions were studied to investigate the dependency of the effect on the detector cell design and the system tubing. Evidence for the turbulent flow origin of this noise is that its onset always occurred at a given fixed value of the Reynolds number. This critical Reynolds number could also be used to predict the onset of detector noise during gradient elution runs. Second, evidence for the turbulent flow origin of the noise was that it could be reliably eliminated using a flow splitter after the column to reduce the flow rate below its critical value before entering the detector. The loss in separation efficiency and detection sensitivity accompanying this flow splitting solution was found to be so small that it does not weigh up against the huge advantage of the possibility to eliminate the excessive detector noise.
The pursuit of ever decreasing analysis times undoubtedly has been one of the major driving forces behind the recent developments in HPLC. Analysis time is a major concern in a whole range of high-throughput applications and on-line process control.1-3 High separation speeds are also essential in comprehensive liquid chromatography, where the separation in the second dimension * Corresponding author. Pleinlaan 2, 1050 Brussels, Belgium. Phone: (+)32.(0)2.629.32.51. Fax: (+)32.(0)2.629.32.48. E-mail: gedesmet@ vub.ac.be.. † Vrije Universiteit Brussel. ‡ Universiteit Gent. (1) Gerber, F.; Krummen, M.; Potgeter, H.; Roth, A.; Siffrin, C.; Spoendlin C. J. Chromatogr., A. 2004, 1036, 127-133. (2) Olsˇovska´, J.; Jelı´nkova´, M.; Man, P.; Kobeˇrska´, M.; Janata, J.; Flieger, M. J. Chromatogr., A. 2007, 1139, 214-220. (3) Stoll, D. R.; Paek, C.; Carr, P. W. J. Chromatogr., A. 2006, 1137, 153-162. 10.1021/ac701906j CCC: $40.75 Published on Web 01/31/2008
© 2008 American Chemical Society
is usually carried out at very high flow rates on short, low efficiency columns.4,5 Among the approaches to reach fast analysis is HPLC at elevated temperature (HT-HPLC),6-10 wherein the decreased viscosity of the mobile phase alleviates the pressure drop limitation to achieve high flow velocities and separation speeds. The raised temperature and the reduced mobile phase viscosity also combine into an increased molecular diffusivity. This leads to a reduced C-term band broadening, allowing the conservation of a high efficiency at high flow velocities. Another approach to achieve high separation speeds is the use of ultrahigh pressure in combination with small particle columns.11-15 Here the quadratic relation between the particle size and the C-term band broadening is exploited to maintain a high efficiency at high flow velocities. Common to these different high-speed separation approaches is that they involve the application of high flow rates, especially if performed in wide bore columns. This holds even more in the case of monolithic columns, another column type that is often used in high-speed analysis.16-19 Compared to a packed column operated at the same linear velocity, the higher external porosity of monolithic columns leads to a corresponding increase in flow rate. These high flow rates can introduce some unexpected sideeffects, as was observed during a series of high flow rate (4) Dugo, P.; Favoino, O.; Luppino, R.; Dugo, G.; Mondello, L. Anal. Chem. 2004, 76, 2525-2530. (5) Franc¸ ois, I.; de Villiers, A.; Sandra, P. J. Sep. Sci. 2006, 29, 492-498. (6) Chen, H.; Horvath, Cs. J. Chromatogr., A. 1995, 705, 3-20. (7) Greibrokk, T.; Andersen, T. J. Chromatogr., A. 2003, 1000, 743-755. (8) Thompson, J.; Carr, P. Anal. Chem. 2002, 74, 1017. (9) Vanhoenacker, G.; Sandra, P. J. Sep. Sci. 2006, 29, 1822-1835. (10) Yan, B.; Zhao, J.; Brown, J. S.; Blackwell, J.; Carr, P. Anal. Chem. 2000, 72, 1253-1262. (11) Jerkovich, A. D.; Mellors, J. S.; Thompson, J. W.; Jorgenson, J. W. Anal. Chem. 2005, 77, 6292-6299. (12) Patel, K. D.; Jerkovich, A. D.; Link, J. C.; Jorgenson, J. W. Anal. Chem. 2004, 76, 5777-5786. (13) de Villiers, A.; Lestremau, F.; Szucs, R.; Ge´le´bart, S.; David, F.; Sandra, P. J. Chromatogr., A. 2006, 1127, 60-69. (14) Majors, R. E. LCGC North Am. 2005, 23, 1248-1255. (15) Nguyen, D. T.-T.; Guillarme, D.; Heinisch, S.; Barrioulet, M. P.; Rocca, J.L.; Rudaz, S.; Veuthey, J. L. J. Chromatogr., A. 2007, 1167 (1), 76-84. (16) Hjerte´n, S.; Liao, J. L.; Zhang, R. J. Chromatogr. 1989, 473, 273-275. (17) Svec, F.; Fre´chet, J. M. Anal. Chem. 1992, 64, 820. (18) Gusev, I.; Huang, X.; Horvath, Cs. J. Chromatogr., A. 1999, 855, 273-290. (19) Minakuchi, H.; Nagayama, H.; Soga, N.; Ishizuka, N.; Tanaka, N. Anal. Chem. 1996, 68, 3498.
Analytical Chemistry, Vol. 80, No. 5, March 1, 2008 1679
experiments on a state of the art commercial instrument. During these experiments, a strong baseline noise suddenly appeared beyond a certain flow rate. Switching back and forth between low and high flow rates, the noise could be suppressed and generated with a perfect repeatability. In the process of trying to determine the origin of this sudden noise, it was found to occur under various conditions of mobile phase composition and for a variety of different detector cells. Given the strong dependency of the phenomenon on the flow rate, the hypothetical explanation for this observation was that a turbulent flow was developed in the detector cell. According to the theory of hydrodynamics, a flow becomes turbulent if the ratio of inertial to viscous forces exceeds a certain critical value.20-23 Expressing these forces in terms of the system variables leads to the dimensionless Reynolds number (Re):
Re )
uFdref η
(1)
wherein u is the fluid velocity (m/s), F the fluid density (kg/m3), η the dynamic viscosity (kg/m s), and dref a characteristic dimension (in m). Whereas in a laminar steady flow, the velocity vectors in each point of the flow domain remain invariable with the time, these vectors tend to fluctuate heavily around a given mean if the flow becomes turbulent. Usually, this turbulent flow is organized in a series of superimposed whirls (so-called eddies) which strongly increase the mass transport and mixing rates over those encountered in a laminar flow. The selection of the characteristic length dref that needs to be used in eq 1 can be quite difficult in the case of a complex geometry, but in the case of a circular tube, dref can be simply taken equal to the tube diameter. Doing so, it is then a wellestablished fact that for cylindrical straight pipes with a constant cross-section and smooth tube walls turbulence sets in if Re is of the order of 2000-3000, regardless of the fact whether the fluid is a gas or a liquid. In the case of rough tube walls or the presence of sudden or gradual bends and contractions, turbulence sets in at a lower Re typically ranging between 500 and 2000.20,21 This follows from the theory of hydrodynamic stability20 and is due to the fact that turbulence is a phenomenon that needs to be triggered. In a perfectly smooth tube, this triggering is of a fundamental nature, as it comes from inevitable small thermal vibrations at the molecular scale. In other cases, this triggering originates from the presence of little surface roughness elements or from sudden contractions or sudden bends. Another notable point is that, once a flow has entered a flow domain with turbulence promoting conditions, it still takes a certain length before the turbulence is fully developed.20,22 Similarly, it also takes a finite length for a turbulent flow entering a flow domain with laminar flow promoting conditions to damp out and to become effectively laminar. The present study has been set up to check the abovementioned turbulent flow-induced detector noise hypothesis. For (20) Schlichting, H. Boundary-Layer Theory; McGraw-Hill: London, U.K., 1958. (21) Granger, R. A. Fluid Mechanics; Dover Publications Inc.: New York, 1995. (22) Incropera, F. P.; De Witt, D. P. Fundamentals of Heat and Mass Transfer, 2nd ed.; Wiley: New York, 1985. (23) Hlushkou, D.; Tallarek, U. J. Chromatogr., A. 2006, 1126, 70-85.
1680
Analytical Chemistry, Vol. 80, No. 5, March 1, 2008
this purpose, two main types of experiment were conducted: (1) changing the mobile phase composition to change its viscosity and density; the onset of the noise should always occur at a flow rate that corresponds to a single critical value of the Reynolds number and (2) splitting-off part of the flow after the column to bring the flow rate below its critical value should eliminate the noise. Apart from the above checks, the dependency of the effect on the detector cell design and the system tubing was investigated by testing several detector cells of different dimensions and by changing the dimensions of the connection tubing. The effect of the employed pump was tested as well. The occurrence of turbulent flow conditions as a potential source of noise during absorption detection has to the best of our knowledge thus far only been suggested as a side-remark in a paper by Gerber et al. treating the practical aspects of fast RPHPLC chromatography.1 They, however, did not verify the turbulent flow assumption and performed no experiments to determine its geometrical causes. EXPERIMENTAL SECTION Chemicals and Columns. Acetonitrile (HPLC gradient grade) and water (Chromasolv HPLC, gradient grade) were purchased from Sigma-Aldrich (Steinheim, Germany). The acetonitrile (ACN) contained maximally 0.0005% of nonvolatile matter and had a transmittance of minimally 75% at 195 nm, minimally 93% at 200 nm, and minimally 99% at 230 nm. The water contained maximally 0.001% of nonvolatile matter. Propylparaben was purchased from Sigma-Aldrich (Steinheim, Germany) and dissolved in pure ACN at a concentration of 50 µg/mL (ppm). Toluene (Fischer Scientific, Leicestershire, U.K.), ethylbenzene (Sigma Aldrich, Steinheim, Germany), phenanthrene (Merck, Hohenbrunn, Germany), and perylene (Aldrich Chemical Co., Milwaukee, WI) were all dissolved in 20/80 vol %/% H2O/ ACN at a concentration of 40 µg/mL (ppm) and used in the gradient experiments. During the course of the study, two columns of similar dimensions were used: an Agilent Zorbax Stable Bond column C18 (50 mm × 4.6 mm, particle size 3.5 µm) was purchased from Agilent (Palo Alto, CA) and a Waters Xbridge C18 column (50 mm × 4.6 mm, particle size 3.5 µm) was purchased from Waters (Wexford, Ireland). The results obtained with both columns were perfectly interchangeable, in agreement with the physical expectations. Apparatus and Methodology. All experiments were conducted at a temperature of 30 °C. The measurements were performed on a HPLC Agilent 1200 system equipped with a binary pump (Agilent Technologies, Waldbronn, Germany) and a diode array detector. The system was operated with the Agilent Chemstation software. Absorbance values were measured at 254 nm (all isocratic runs) and 214 nm (the gradient runs in A Practical Example: Gradient Elution section) at a data acquisition rate of 80 Hz. The injection volume was 2.0 µL. To investigate the influence of the employed pump, a series of experiments replacing the binary pump by a preparative pump (Smartline 1000, Knauer, Berlin, Germany) equipped with a 50 mL pump head were conducted as well. For every investigated combination of mobile phase composition and detector cell, the flow rate was always increased from
Table 1. Dimensions of the Used Detection Flow Cells and Critical Reynolds Range Based on the Diameter of the Inlet Tubing, the Diameter of Each Flow Cell Was Calculated from Its Path Length and the Volume as Stated by the Manufacturer
flow cell
volume (µL)
path length (mm)
micro microhigh pressure semi-micro
2 1.7 5
3 6 6
diameter (mm)
diameter inlet tubing (mm)
critical Re range
0.92 0.60 1.03
0.12 0.12 0.12
800-1000 800-1000 800-1000
fractions of acetonitrile and water, respectively; and Rorg and Rwater are empirical parameters. The density (F) of the different mobile phases was calculated as
Fm ) xorgForg + xwaterFwater
(3)
where ηorg, ηwater, and ηm are the viscosities of acetonitrile, water, and the mixture, respectively; φorg and φwater are the volume
where Forg, Fwater, and Fm are the densities of acetonitrile (771.2 kg/m3), water (995.7 kg/m3), and the mixture, respectively; and xorg and xwater are the mole fractions of acetonitrile and water in the mixtures, respectively. Different micro- and semi-microflow cells were studied: a highpressure microflow cell (volume, 1.7 µL), a normal microflow cell (volume, 2 µL), and a semi-microflow cell (volume, 5 µL). All these flow cells have capillary inlet tubing that goes through a heat exchanger. This capillary tubing leading into the flow cells is part of the standard configuration of the instrument and has a length of 290 mm and is made from stainless steel. Two types of i.d. were available (i.d. ) 120 or 170 µm). The flow cells themselves are claimed to have been manufactured in such a way that the cross section is uniform and the volume is very precise.25 Therefore, it is assumed that the diameter of the flow cell can be calculated from its volume and path length, approximating the internal flow path as a cylindrical tube. The most important geometrical properties of the flow cells are given in Table 1. During most of the experiments, a zero dead-volume connection piece was present to connect the outlet tubing of the column (PEEK, 50 mm length, 125 µm i.d.) with the inlet tubing of the detector cell. It could, however, be verified that the presence of this piece did not affect the flow rate at which the turbulent noise occurred. Flow Rate Accuracy and Flow Splitting. The flow rate accuracy of the Agilent 1200 binary pump was evaluated at low (0.15 mL/min) and high (1, 2, and 4 mL/min) flow rates. For this purpose, pure acetonitrile was pumped through the system with a column installed. The volume of the mobile phase was collected in a 10 mL measuring cylinder at the detector outlet. The measurements were performed at 30 °C. The pumping accuracy was always better than 98.8% at all flow rates. For the preparative pump, the pumping accuracy was always better than 97.8% at all flow rates. To split the flow, a metal T-piece (Upchurch Scientific, Oak Harbor, WA) was installed after the column but prior to the zerodead volume connection piece. Connecting the T-piece with the
(24) Li, J.; Carr, P. Anal. Chem. 1997, 69, 2530-2536.
(25) Mueller, J.; Mueller, B. U.S. Patent 7,005,090, Feb. 28, 2006.
Figure 1. Baseline noise recorded with the 1.7 µL flow cell. (a) Typical baseline signal (0.06 mAU) when no significant noise is present, recorded at a flow rate of 1.75 mL/min, corresponding to a Reynolds number of 746, (b) baseline-noise determined at a flow rate of 2 mL/min (0.4 mAU), corresponding to a Reynolds number of 853, (c) baseline-noise at a flow rate of 3 mL/min (2.5 mAU), corresponding to a Reynolds number of 1279. To indicate how the noise was read out from the chromatograms, arrows have been added to parets (b) and (c). Mobile phase ) pure ACN.
0.25 to 5 mL/min in steps of 0.25 mL/min. After each new flow rate increment, the baseline noise was recorded over a time window of 2 min, after a 3 min stabilization period. The average peak-to-valley value was read out as shown in Figure 1. To check the influence of the mobile phase composition, this was modified from 100% ACN to 20% ACN in steps of 10%. The viscosity of the different employed mobile phases was estimated with a correlation, established by Li and Carr:24
ηm ) φorgηorg exp(φwaterRwater) + φwaterηwater exp(φorgRorg) (2)
Analytical Chemistry, Vol. 80, No. 5, March 1, 2008
1681
waste, the flow leaving the detector cell was split into a certain ratio by adjusting the length of the employed split tubing (PEEK, 125 µm i.d.). Two split ratios (respectively, splitting off 1/2 and 2/ of the total flow) were considered. The accuracy of the flow 3 rate during these flow splitting experiments was checked at different flow rates and was always better than 95%. To investigate the effect of the flow splitting on efficiency, sensitivity, and retention, 50 µg/mL propylparaben was injected on the Xbridge column with no split, a 1/2-split, and a 2/3-split applied at flow rates ranging from 1 to 5 mL/min and with a mobile phase consisting of 100% ACN. Plate counts were determined at half-height. The obtained data were subsequently corrected for the system contribution by replacing the column with a zero-dead volume union and repeating the same injections. The corrected plate heights were determined as
H)
σt2 - σt,ext2 L )L N (t - t )2 R
(4)
R,ext
wherein tR is the solute retention time and tR,ext the time spent by the solute in the system. σt2 represents the total peak variance and σext2 the external variance. Experiments Conducted With On-Capillary Detection. To investigate whether the turbulent flow profile fully developed in the inlet tubing itself or rather at the interface between the inlet tubing and the detector cell, experiments were conducted on a HP 3D CE instrument (Hewlett-Packard, Waldbronn, Germany) with diode-array detection (DAD). DAD signals were recorded at 254 nm. The capillary cassette was thermostatted at 30 °C. Fused silica capillaries (Polymicro Technologies, Phoenix, AZ) of varying diameters (50, 100, and 150 µm i.d.) were used. The detection window was simply formed by removing a small part of the polyimide coating from the capillary. Because of the on-capillary detection, no junction was present between capillary and detection window. Pure ACN was pumped through the capillaries at different flow rates using an external pump (both the Agilent 1200 pump and the preparative pump). RESULTS AND DISCUSSION Onset of Noise as a Function of the Flow Rate. Figure 1 shows the phenomenon that was typically observed in the present study with the micro- high-pressure flow cell (1.7 µL). Below a certain critical value, all flow rates lead to a normal (i.e., on the order of 0.06 mAU) degree of noise (Figure 1a). A slight increase in flow rate (only a step of 0.25 mL/min in a 4.6 mm bore column) then suddenly leads to a clear increase of the baseline noise to 0.30 mAU (Figure 1b). This noise subsequently increased in magnitude if the flow rate was further increased, typically leading to noise levels on the order of 2.0-2.5 mAU (Figure 1c at 3 mL/ min). Figure 2 shows a typical evolution of the observed noise as a function of the flow rate ((). The reported noise values have been determined as indicated in Figure 1. Figure 2 clearly shows how the noise level remains flat up to a certain value and then abruptly increases in one single flow rate increment step of 0.25 mL/min toward a noise level that is typically at least 5 times larger. This behavior is representative for all conducted flow rate series experiments (cf. the graphs shown further on). In overlay (0), 1682 Analytical Chemistry, Vol. 80, No. 5, March 1, 2008
Figure 2. Baseline noise versus flow rate for the 1.7 µL flow cell ((, left-hand y-axis). The Reynolds numbers corresponding to each flow rate are shown as well (0, right-hand y-axis). The arrows indicate how the Reynolds number at a certain flow rate and corresponding to a certain noise level should be read out from the graph. Mobile phase ) pure ACN.
Figure 2 also shows the Re associated with each flow rate, using eq 1 and taking the diameter of the detector cell inlet tubing as the reference length (dref). The critical Reynolds number associated with the turbulent flow transition can be read out by drawing a vertical arrow at the critical flow rate and then reading out the corresponding Re on the y-axis on the right-hand side of the graph. In the present example, the critical Re is read out as 850. This value is smaller than the traditional Re ) 2000-value usually cited in text books21,22 as the transition point to turbulence, but it clearly falls within the reported range of Re (500-2000) wherein the flow through a cylindrical tube can become turbulent in the case of a severe internal wall surface roughness.20 The assumption that the present case provides an example of a flow system wherein the surface roughness has lowered the critical Reynolds number certainly makes sense, since stainless steel tubing is reputed to develop some severe surface roughness.21 Of course, it could also still be that the observed lowering of the critical Re is due to the presence of some sharp internal bends inside the detector flow cell or due to the sudden tube expansion at the interface between the connection capillary and the flow cell. This issue is further discussed in the Experiments Without Detection Cell section). Use of Flow Splitting To Eliminate the Onset of Noise. Because the Reynolds number is directly proportional to the velocity through the detector cell and hence also the flow rate, it was assumed that it should be possible to decrease or even eliminate the noise by decreasing the flow rate going through the outer connection tubing and the flow cell below the value corresponding to the critical Re. Figure 3 shows the evolution of the observed baseline noise as a function of the total flow rate in the case of a 0, a 1/2-, and a 2/3-flow split. Without flow splitting, the flow turns turbulent at a flow rate of 2 mL/min. When the flow is split in half, the noise only sets in at a total flow rate of 4 mL/min, i.e., exactly 2 times as high as in the absence of a flow split. The Reynolds number based on the reduced flow rate going through the connection tubing and the detector is again 850, i.e., the same value as in the nonsplitted case. For the 2/3-flow split (2 and 4), the Reynolds number never reaches the critical Re, not even at the highest possible flow rate of 5 mL/min. It is hence
Figure 3. Baseline noise versus flow rate for the 1.7 µL flow cell (2, 9, (, left-hand y-axis) and corresponding Reynolds numbers (4, 0, ), right-hand y-axis). The different curves correspond to different ratios of flow splitting that were applied: 0-flow split ([), 1/2-flow split (9), and 2/3-flow split (2). The arrows indicate the Reynolds number that corresponds to the onset of the turbulent regime. Mobile phase ) pure ACN.
not surprising to see that the 2/3-split ratio effectively allows avoiding any noise increase over the entire range of possible flow rates. The data shown in Figure 3 clearly provide another confirmation of the hypothesis of a critical Re determining the onset of the detector noise, in turn suggesting the turbulent flow origin of this noise. Given the potential practical application of this flow splitting method as a means of detector noise reduction, the effect of the presence of the T-piece flow splitter on the retention, efficiency, and detector sensitivity has been investigated. The results are shown in Table 2. In the case of a 1/2-split, the T-piece only has a very small influence: the retention time increases with some 2% because the system volume is slightly larger, the efficiency drops with less than 5%, and the detection sensitivity is maximally 6% lower than without flow split. Making a 2/3-split, the effect on the system performance is slightly larger: the retention time increases by 2.5%, the efficiency drops by 12%, and there is a loss in sensitivity of maximum 11%. The major cause of the loss in efficiency and the increased residence time of course is the additional dead volume of the T-piece used to split the flow. This could be witnessed from the fact that, if correcting the obtained plate height values for the extra-column contribution, the loss in efficiency was reduced to only 5% instead of the 12% difference between the noncorrected plate height values. It must be noted that, when repeating the experiments with a zero-dead volume instead of the column to correct for the system contribution, the increase in noise occurred at the same flow rate as observed with the column installed. This observation was made with and without the flow-splitter installed. This implies that the column itself does not contribute to the development of the turbulent regime as observed in the detector cell. It should be remarked that the efficiency losses given above were obtained by comparing the performances of the system below the critical flow rate. The cited values hence only relate to the loss in efficiency caused by the presence of the flow splitting T-piece under conditions where no noise occurs. It should be obvious that this loss in efficiency is, however, negligible compared to the loss in efficiency caused by the presence of the noise.
The latter can be completely detrimental for the efficiency, especially if low-concentration peaks are present in the chromatogram. Since the T-piece allows effectively eliminating this noise, the overall effect of the T-piece is of course highly positive. Onset of Noise as a Function of the Mobile Phase Viscosity. To investigate the influence of the nature of the liquid on the observed detector noise, the mobile phase composition was varied between 100% ACN and 20% ACN in steps of 10%. For each mobile phase, a complete flow series sequence was run (very often in duplicate or triplicate). Knowing that the viscosity varies from η ) 0.320 × 10-3 kg/(m s) at 100% ACN, reaches a maximum at η ) 0.895 × 10-3 kg/(m s) at 20% ACN, and subsequently decreases to a value of η ) 0.846 × 10-3 kg/(m s) at 5% ACN, varying the mobile phase rate under the assumption that the onset of the noise occurs at a fixed Re implies that the flow rate corresponding to this onset should also vary considerably. Equation 1 in fact shows that the existence of a fixed transition Re requires the observed critical flow rate to vary in a linearly proportional way with the mobile phase viscosity. In the present case, the critical flow rate should go from 2.0 mL/min at pure ACN to 4.5 mL/min at 20% ACN. This is indeed what was observed, as is shown in Figure 4. In this figure, the noise values have been plotted directly versus the corresponding Re because this provides a much more compact representation than the flow rate plots shown in Figures 2 and 3. As can be noted, the detector noise always sets on at roughly the same critical Re (around 800-1000), while the actual difference in critical flow rate between the pure ACN and the 20% ACN amounts up to a factor of 2.25. Figure 4 hence provides another clear argument for the turbulent nature of the observed detector noise, because it demonstrates that the noise sets in at a given critical Re, independent of the employed liquid. Of course, there is a certain variation around the exact Re above which the turbulent noise sets in. This has multiple reasons. First, given that the flow rate was increased in steps of 0.25 mL/ min, a variation of some 10% on the critical Re is inherent to the conducted set of measurements. Second, there is also some uncertainty about the viscosity values used to calculate the Re. Comparison of the data coming from the employed viscosity correlation and that coming from our own set of independent measurements, this uncertainty can be assumed to be of the order of some 10%. A similar value was cited by the authors themselves.24 Third, and perhaps most important, turbulence is of a stochastic nature itself, leading inevitably to some statistical fluctuation of the transition Reynolds number. This stochastic nature arises from the extreme dependence between the onset of the turbulent fluctuations and the occurrence of thermal or mechanical vibrations induced by the system or the surroundings. In the present series of experiments, this was reflected by the fact that sometimes a significant increase in noise (i.e., larger than 3 or 5 times the normal noise level) was noticed during one or more flow rate steps, while the noise observed after the following increment step(s) was again clearly lower, before finally taking off toward the highest noise levels at the subsequent steps. The data series for the 90% ACN mobile phase shown in Figure 9 (b and 9), further on, for example, displays such a trend. The occurrence of this “pre-noise” was however highly unpredictable and unrepeatable. Analytical Chemistry, Vol. 80, No. 5, March 1, 2008
1683
Table 2. Effect of Flow Splitting on Efficiency, Sensitivity, and Retention, Determined by Injecting 2 µL of a 50 µg/mL Propylparaben Sample, Dissolved in Pure ACN, On the Xbridge Column 1/
no split
2/
2-split
3-split
F (mL/min)
tR (min)
N
S (mAU)
tR (min)
N
S (mAU)
tR (min)
N
S (mAU)
1 2 3 4 5
0.573 0.292 0.197 0.151 0.122
5800 5600 5110 4888 4500
247 246 236 227 216
0.580 0.295 0.200 0.152 0.124
5638 5565 5182 4733 4298
242 242 223 215 204
0.586 0.298 0.202 0.154 0.125
5175 4913 4484 4460 3996
225 223 211 204 195
Figure 4. Baseline noise versus Reynolds number for the 1.7 µL flow cell. The different curves correspond with different mobile phases: 100% ACN (b), 80% ACN/ 20% H2O (2), 60% ACN/ 40% H2O ([), and 40% ACN/ 60% H2O (9).
Practical Example: Gradient Elution. One of the consequences of the above-noted influence of the mobile phase viscosity is that it can be expected that, during gradient elution separations, the turbulence-induced noise will abruptly appear as soon as the gradient has reached the point where its mobile phase viscosity goes below the critical value given by the critical Re. To demonstrate that this turbulence-induced noise can indeed spoil real life applications, a high-speed gradient analysis of a mixture toluene, ethylbenzene, phenanthrene, and perylene was conducted. The gradient ran from 20% to 100% ACN over 20 min at flow rates of 4 and 4.5 mL/min. Figure 5 shows the gradient run at a flow rate of 4 mL/min. Without flow split (Figure 5a), a steady increase of the noise is seen over the baseline, becoming more pronounced after approximately 8 min, corresponding to a mobile phase composition of 50% ACN and hence to a Reynolds number of 920. The noise becomes especially disturbing for the fourth peak, preventing its proper integration. Splitting the flow in half (Figure 5b) allows running the gradient analysis almost entirely, without any disturbing noise occurring. Indeed, the increase in noise is only noticed after about 20 min, i.e., when the mobile phase consists of 100% ACN and the Reynolds number equals 850. To perform the analysis without any noise occurring, only one-third of the flow can be allowed through the flow cell (Figure 5c). Indeed, in this case the highest Reynolds number never exceeds a value of 710 and the turbulent flow transition cannot occur. Figure 6 shows the gradient run at a flow rate of 4.5 mL/min. At this flow rate, the Reynolds number amounts to 871 already at 20% ACN. It is hence no surprise to see for the zero flow split1684 Analytical Chemistry, Vol. 80, No. 5, March 1, 2008
Figure 5. Gradient run on the Agilent column at a flow rate of 4 mL/min. The mobile phase composition was varied from 20% ACN/ 80% H2O to 100% ACN in 20 min; 2 µL of a mixture consisting of toluene (1), ethylbenzene (2), phenanthrene (3), and perylene (4), each at a concentration of 40 µg/mL was injected on the column. Different flow split ratios were applied: (a) 0-flow split, (b) 1/2-flow split, and (c) 2/3-flow split. The arrows indicate the onset of the turbulent regime.
case that the turbulence-induced noise is already present at the start of the gradient and remains so throughout the entire run (Figure 6a). The integration of all four peaks is hampered and especially peak three and four are overwhelmed by the noise. For the 1/2-flow split, the noise appears around 17.5 min (Figure 6b). The mobile phase present in the column at that time roughly has an ACN-percentage of 90%, having a viscosity of 0.391 × 10-3 kg/
Figure 6. Gradient run on the Agilent column at a flow rate of 4.5 mL/min. The mobile phase composition was varied from 20% ACN/ 80% H2O to 100% ACN in 20 min; 2 µL of a mixture consisting of toluene (1), ethylbenzene (2), phenanthrene (3), and perylene (4), each at a concentration of 40 µg/mL was injected on the column. Different flow split ratios were applied: (a) 0-flow split, (b) 1/2-flow split, and (c) 2/3-flow split. The arrows indicate the onset of the turbulent regime.
(m s) and leading to a value of Re ) 842, in perfect agreement with the above observed critical Re range. Figure 6c shows that only the 2/3-split is strong enough to prevent the onset of the turbulence-induced noise throughout the entire chromatogram. The above experiments show that the possibility to calculate the Reynolds number not only allows to predict the onset of the turbulent noise but also to determine which split ratio is needed to avoid any turbulence showing up during the gradient and hindering a proper integration of the peaks. Influence of System Geometry and Detector Cell Size and Shape. To examine the influence of the flow cell design and to gain insight in the main geometrical parameter causing the observed turbulence in the flow cell, different flow cells were compared. For each cell, the flow rate was varied for different mobile phase compositions (now ranging from 100% to 60% ACN) and the noise was monitored. Figure 7a shows the course of the noise in the 1.7, the 2, and the 5 µL detector cells, for three
Figure 7. Baseline noise versus Reynolds numbers for three different flow cells: 1.7 µL flow cell (black, solid symbols, left-hand y-axis), 5 µL flow cell (black, open symbols, left-hand y-axis), and 2 µL flow cell (gray, solid symbols, right-hand y-axis) at a mobile phase composition of 100% ACN (b), 90% ACN/10% H2O (9), and 80% ACN/20% H2O (2). (a) Reynolds number calculated with dref ) dtub, (b) Reynolds number calculated with dref ) d cell.
different mobile phase compositions each, and plotted versus the Reynolds numbers based on the diameter of the connection tubing. As can be noted, the onset of the noise occurs in the same critical Re in all three detector cells (again Recrit = 900). This implies that the major source of the turbulent noise must be something all flow cells have in common. Given the three cells have a variable path length and internal diameter, the single correspondence between the three cells is the diameter of the connection tubing. This justifies the use of the connection tubing diameter as the relevant characteristic dimension in eq 1. Indeed when plotting the noise level versus the Reynolds number associated with the diameter of the flow cell’s diameter, the onset of the turbulence now no longer coincides (Figure 7b). The 5 and 2 µL flow cells still nearly perfectly coincide, but this is due to the fact that they have similar internal diameters. Experiments with a much larger flow cell (13 µL standard flow cell) showed a deviation from the otherwise consistently observed Re = 900-rule. In this case, the onset of the turbulent noise only occurred around Re = 1050. Given that in this cell the inner diameter is much larger than in the three other cells (approximately 10 mm), it can be assumed that the ratio of the incoming connection tubing and the inner cell diameter is so large that the sudden decrease in fluid velocity is so large that the Analytical Chemistry, Vol. 80, No. 5, March 1, 2008
1685
Figure 8. Baseline noise versus Reynolds number obtained with the binary pump (b) and with the preparative pump (4). (a) 1.7 µL flow cell and (b) 5 µL flow cell. Mobile phase ) pure ACN.
incoming turbulent flow decays more rapidly and therefore only yields a visible noise at a larger flow rate. This assumption is, however, difficult to validate because the difference in design between the standard flow cell and the other, smaller flow cells is not only the larger cell diameter of the former but also the very long inlet tubing that is coiled around the standard flow cell in a specifically designed groove. When experiments are done without the zero dead-volume piece connecting the column outlet tubing and the detector inlet tubing, directly connecting the detector to the column, no significant difference in critical Re could be observed. This hence also allows ruling out the presence of this zero dead-volume connection as a potential source of the turbulent noise. Influence of the Employed Pump. To investigate the influence of the employed pump, the standard binary pump of the instrument was replaced by a preparative pump. Again, a mobile phase consisting of 100% ACN was pumped through the system at different flow rates, and the onset of the turbulent regime was monitored for the 2 and 5 µL flow cells. Figure 8 compares the evolution of the noise for both pump types in the 2 µL (Figure 8a) and the 5 µL flow cell (Figure 8b). Again no significant influence of the employed cell could be noted. The pump itself, however, seems to have a small but significant influence on the observed noise pattern. Whereas with the binary pump the noise level remains flat until the flow rate reaches the critical Re, the preparative pump already displays a slight and 1686
Analytical Chemistry, Vol. 80, No. 5, March 1, 2008
gradual noise increase before this critical Re is reached. The biggest jump in detector noise, however, still occurs around the same critical Re as for the binary pump. In the turbulent regime, the preparative pump showed large fluctuations in pressure, often of 3-4 bar, while the pressure of the binary pump remained stable. Possibly, the higher noise level of the preparative pump in the lower Re-region is due to the fact that it generates some slight pressure fluctuations, in turn enhancing the transition to turbulence. Influence of the Diameter of the Connection Tubing. As a final test for the assumption that the observed turbulence originates from the stainless steel tubing preceding the detector cell, it was investigated what the effect of changing the inlet tubing would be. This is of course the most direct test for our assumption that the turbulences are generated in the connection capillary preceding the detector cell. To our knowledge, only two formats of capillary inlet tubing with heat exchanger as used for the Agilent flow cells are available: the 120 µm i.d. inlet tubing with a length of 290 mm, used for the 1.7, 2, and 5 µL flow cells, and the 170 µm i.d. inlet tubing with a length of 590 mm, used for the 13 µL flow cell. Because only the 13 µL flow cell incorporates a special groove where the long inlet capillary can be wound on, it was impossible to mimic the exact same configuration on any of the other available detector cells. However, we managed to install it on the 5 µL flow cell and coil the inlet tubing around the flow cell, be it less tight than possible on the 13 µL cell. Three mobile phase compositions (ranging from 100 to 80% ACN) were studied with the new inlet tubing (170 µm i.d.) on the 5 µL flow cell and the onset of the increased noise was compared with the former results obtained on the 5 µL flow cell, using its own inlet tubing (120 µm i.d.). Figure 9 shows the increase in noise versus the Reynolds number for both configurations at mobile phase compositions of 100% ACN and 90% ACN. When the Reynolds number is calculated using the diameter of the inlet tubing as dref (Figure 9a), both configurations lead to an increase in noise at Re = 800-1053, which still corresponds with the former values obtained on the other flow cells. With the use of the diameter of the detector cell as dref (Figure 9b), the resulting critical Reynolds numbers are not correlated (Re = 100 for the 120 µm i.d. tubing and Re = 150160 for the 170 µm i.d. tubing). Although the combination of Figures 7 and 9 irrefutably shows that the onset of turbulence is related to the connection tubing diameter (a change of detector cell does not influence the value of the critical flow rate, whereas a change of the connection tubing diameter causes a linear proportional shift of the critical flow rate), it cannot be ruled out that the true origin of the observed turbulence is situated at or close to the interface between the connection capillary and the detection cavity. Given the estimated dimensions of the latter (cf. Table 1), this interface is marked by a sudden (large) expansion of the diameter and it is known from the theory of hydrodynamics26 that such a sudden transition gives rise to a boundary layer separation and the creation of recirculation vortices in the low velocity corners of the cavity, in turn leading to the creation of a series of time-dependent flow instabilities (Hopf bifurcations), becoming increasingly complex with increasing Re. It now so turns out that the onset of this series of events also (26) Rani, H. P.; Sheu, Tony W. H. Phys. Fluids 2006, 18, 084101.
Figure 9. Baseline noise versus Reynolds numbers for the 5 µL flow cell, using the normal 120 µm i.d. inlet tubing (open symbols) and using the 170 µm i.d. inlet tubing (solid symbols) and at a mobile phase composition of 100% ACN (b) and 90% ACN/10% H2O (9). (a) Reynolds number calculated with dref ) dtub, (b) Reynolds number calculated with dref ) dcell.
depends strongly on the diameter of the inlet tube diameter,26-28 making it is impossible to distinguish from the experiments reported in Figure 9 whether the turbulence that is observed is generated inside the connection tube or immediately after the sudden expansion at the flow cell inlet. Critical Re associated with the transition to a turbulent flow pattern after a sudden diameter expansion in a rectilinear system have been reported to range between Re ) 150 (short and wide cavity)28,29 and 750 (shallow but long cavity).26 These values certainly are of the same order of magnitude as the critical Re observed in the present study. This agreement can however not be used to pinpoint the source of the turbulent flow to the sudden expansion at the entrance of the flow cell because none of the geometries investigated in literature really correspond to the exact shape of the presently employed detector cells. From ref 25 we know that the flow inside the 3-D geometry of the flow cell is much more complicated than in the simplified geometries investigated in refs 26-28, as the flow inside the detection cell also makes a number of turns and in fact undergoes more than one diameter expansion. Flow simulations could provide an outcome here, but given the strong (27) Nie, J. H.; Armaly, B. F. Int. J. Heat Mass Transfer 2004, 47, 4713-4720. (28) Kolsek, T.; Jelic, N.; Duhovnik, J. Appl. Math. Modell. 2007, 31, 23552373. (29) Maurel, A.; Ern, P.; Zielinska, B. J. A.; Wesfreid, J. E. Phys. Rev. E 1996, 54, 3643-3651.
dependency of the critical Reynolds number values reported in literature on the width and length of the tube after the expansion,26-29 such simulations would only be relevant if every detail of the flow cell would be known with the greatest accuracy (which is not the case) and would require an extensive experimental validation study involving in situ measurements of the velocity field, since the outcome of the simulations would depend highly on the selected turbulence models. Experiments Without Detection Cell. To try to determine whether the turbulent flow fully develops in the inlet tubing (capillary) or rather at the interface between the inlet tubing and detector cell, a series of experiments was conducted without a detection flow cell, using a capillary CE system with on-capillary detection. With the use of the 1200 pump to generate the flow rate and a 100 µm i.d. capillary, a Re of 2558 was reached at a flow rate of 5 mL/min. Even for a perfectly smooth tube (assuming that fused silica is smoother than for example metal), turbulence should set in at a Re of approximately 2000. We could, however, not see any difference in baseline noise when comparing the high-flow rate experiments with lower-flow rate experiments. To reach even higher Re, we subsequently coupled the preparative pump to the CE device, now equipped with a 150 µm i.d. capillary (a larger diameter was used to reduce the backpressure in the capillary). Pumping pure ACN through the capillary allowed increasing the flow rate up to 15 mL/min, corresponding to a Re of 5117 for which the flow should certainly be turbulent. Again, however, no increase in baseline noise was observed. As a final test, we tried to mimic the abrupt transition in diameter as it occurs when the inlet tubing reaches the inlet of the detector cell. The ratio in diameters of the inlet tubing (120 µm) and the micro-high-pressure flow cell (600 µm) is 0.2. Therefore, a 50 µm i.d. capillary was coupled to a 250 µm i.d. capillary (ratio ) 0.2) using a zero-dead volume union, and the detection window in the 250 µm i.d. capillary was positioned as close as possible to the transition between the two capillaries. Pure ACN was pumped through the construction at a flow rate of maximum 2.2 mL/min (at higher flow rates the maximum pressure of the system was reached). A flow rate of 2.2 mL/min corresponds with a Re of 2502 when using the diameter of the narrowest capillary (50 µm) as dref (same calculation was done for the detector cell, i.e., using the narrowest diameter as dref). Again, no increase in baseline noise was observed at the highest flow rate, compared to the lower flow rates. It can therefore not be concluded from these types of experiments whether turbulence fully develops in the inlet tubing, at the connection of the column to the detector cell, or at the interface between inlet tubing and detector cell. An explanation for the absence of a baseline noise increase in the capillary setup could be the shorter optical path length (100150 µm) in the capillary compared to the path length in the DAD detector cells (3-10 mm). Vortexes of recirculating fluid and/or matter might not be visible over such a small distance. Other Observations. Although the present study has allowed one to relate the occurrence of the detector noise to a turbulent flow transition, the precise mechanism translating the turbulent eddies stirring up the mobile phase liquid in the detector cell into the observed noise is yet unclear. We can only speculate the noise Analytical Chemistry, Vol. 80, No. 5, March 1, 2008
1687
is caused by the light scattering effect of some microscopic solids or gas bubbles present in the mobile phase. However, with the use of the purest available ACN and water (both containing less than 0.001% of volatile matter) and the use of a degassing system, there seems no way to escape from it. The presence of microscopic “islands” of nonmixed mobile phase components can be ruled out as a potential light scattering source, as the level of noise was as high with pure ACN as it was with the mixtures. When the detector wavelength varied from 210 to 230 nm and 254 nm, it was found that the noise occurred at all UV wavelengths without any special difference in the magnitude of the noise. In the visible range (400 nm), the increase in noise was, however, much smaller, if not nonexistent. Possibly this observation can shed some more light on which liquid constituent precisely leads to the observed noise. CONCLUSIONS Performing high-speed separations with wide bore columns (4.6 mm diameter) can lead to an excessive baseline noise in the signal of diode array UV-vis absorption detector cells. The explanation for this noise is that the required high flow rates (over 2 mL/min for the 4.6 mm diameter column) lead to a transition to turbulent flow conditions in the narrow connection tubing running into the detector or at the sudden transition to the wider flow path in the detector flow cell. Evidence for the turbulent flow origin of this noise is that its onset always occurred at a given fixed value of the Reynolds number, independent of the employed mobile phase composition. In the present study, with the use of stainless steel capillary tubing, this critical Re roughly equaled 900 if using the diameter of the connection tubing as the characteristic distance. This critical Re
1688
Analytical Chemistry, Vol. 80, No. 5, March 1, 2008
was such a reliable noise source indicator that it could even be used to predict the onset of detector noise during gradient elution runs. The critical Re was the same for all investigated detector cells with a volume in the range of 1.7-5 µL. With the use of a pump with a smaller pressure stability, a small “pre-noise” was observed in the range just below the critical Re. In the 13 µL flow cell, where the volume is so big that it can more easily damp out the turbulence in the entering flow, the critical Re was slightly larger, lying around 1050. The use of a flow splitter, put between the column and the detector to reduce the flow rate below its critical value before entering the detector, provides a highly effective and simple method to omit the occurrence of the turbulence-induced noise. The small losses in efficiency and sensitivity that are caused by the presence of this flow splitter are much smaller than the advantage of the noise elimination, because the latter can be completely detrimental for a correct integration or even the detection of small concentration peaks. ACKNOWLEDGMENT D.C. is supported through a specialization grant from the Instituut voor Wetenschap en Technologie (IWT) of the Flanders Region (Grant No. SB/1279/00). Ken Broeckhoven and Isabelle Franc¸ ois are thanked for helpful discussions. Agilent Technologies is acknowledged for the kind donation of the 1200 instrument.
Received for review September 11, 2007. Accepted November 26, 2007. AC701906J
