5122
Ind. Eng. Chem. Res. 2007, 46, 5122-5130
MATERIALS AND INTERFACES Solid-State 29Si NMR Analysis of Cements: Comparing Different Methods of Relaxation Analysis for Determining Spin-Lattice Relaxation Times to Enable Determination of the C3S/C2S Ratio Christopher L. Edwards,†,‡ Lawrence B. Alemany,† and Andrew R. Barron*,†,§,| Department of Chemistry, Department of Mechanical Engineering and Materials Science, and Energy and EnVironmental Systems Institute, Rice UniVersity, Houston, Texas 77005
The determination of the relative concentrations of tricalcium silicate (C3S) and dicalcium silicate (C2S) within a cement sample is potentially important in the prediction of the setting properties of the cement. For model systems, we have demonstrated that the presence of a paramagnetic species in intimate (physical) contact with an otherwise diamagnetic silicon-containing mineral can dramatically decrease the longitudinal relaxation time T1 and consequently allow for the rapid data collection of the 29Si MAS NMR spectrum. Thus, it appears that, for cements, the T1 values of C3S and C2S will be significantly altered because of the paramagnetic Fe3+ present in the C4AF matrix that makes intimate contact with each of the silicate phases. The presence of the paramagnetic C4AF matrix thus allows for the rapid collection of spectra of cements. Experiments that might otherwise take weeks or even months now take mere hours. Three methods of NMR relaxation have been examined and compared: inversion recovery, variable-delay Bloch decay, and saturation recovery techniques. Analysis of an API class H cement demonstrated that the saturation recovery method for analysis of 29Si NMR data offers the best combination of accuracy, speed, and ease of processing. The saturation recovery method was used to determine the C3S/C2S ratios for eight different cements: three API class G cements and five API class H samples. Determination of C3S/C2S ratios by NMR spectroscopy provides an effective method of analysis for cements, owing to NMR spectroscopy’s direct measurement of the minerals in question without the assumptions of composition that are required with oxide analysis from X-ray fluorescence and application of the Taylor-modified Bogue equations. Given the importance of C3S and C2S (and hence their ratio) in defining the performance of cements, we propose that 29Si MAS NMR spectroscopy represents a viable method for cement characterization. Introduction Portland cement is the single most commonly used building material in the world today, despite having its origins in ancient Egyptian construction over 5000 years ago.1 At around $8.6 billion per year, the manufacture of cement in the United States is a significant component of the U.S. economy. Cement production can be classified by application into two primary groups: construction and energy services. The construction applications for cement consume the lion’s share of cement manufactured worldwide, but the cement produced for energy services applications is an integral part of meeting the world’s energy needs and requires tighter quality-control standards to meet that industry’s higher demands on control of the rheological properties of the fluid slurry state; the solid state; and especially the transition from the former to the latter, or the setting process.2,3 Four chief minerals are present in a cement grain: tricalcium silicate (C3S); dicalcium silicate (C2S); tricalcium aluminate * To whom correspondence should be addressed. E-mail: arb@ rice.edu. URL: www.rice.edu/barron. † Department of Chemistry. ‡ Present address: Halliburton Energy Services, Duncan, OK 73536. § Department of Mechanical Engineering and Materials Science. | Energy and Environmental Systems Institute.
(C3A); and calcium aluminoferrite (C4AF), which is more accurately considered a solid solution of dicalcium aluminate (C2A) and dicalcium ferrite (C2F). The calcium aluminoferrite forms a continuous phase around the other mineral crystallites.4 A typical example of cement contains 50-70% C3S, 15-30% C2S, 5-10% C3A, 5-15% C4AF, and 3-8% other additives. It is the hydration of these minerals that causes the hardening, or setting, of cement. In terms of cement composition, C3S and C2S have the primary influence on the long-term development of structure. Cements high in C3S hydrate more rapidly and lead to higher early strength. However, the hydration products formed, in effect, make it more difficult for hydration to proceed at later stages, leading to an ultimate strength that is lower than desired in some cases. Cements high in C2S hydrate much more slowly, leading to a denser ultimate structure and a higher longterm strength. On this basis, the accurate determination of the C3S/C2S ratio is important in the potential prediction of cement behavior. The “classical” calculation of the C3S/C2S ratio is based on oxide analysis from X-ray fluorescence (XRF) and application of the Taylor-modified Bogue equations.5 Unfortunately, this method makes several assumptions in the determination of the silicate concentrations, leading to questions of its accuracy and the development of alternative methodologies.6 Over the years, solid-state nuclear magnetic resonance (NMR) spectroscopy has been used in a wide variety of ways to
10.1021/ie070220m CCC: $37.00 © 2007 American Chemical Society Published on Web 06/23/2007
Ind. Eng. Chem. Res., Vol. 46, No. 15, 2007 5123 Table 1. Comparison of NMR-Active Isotopes for Common Elements in Cement isotope
spin
17O
5/
2
27Al
5/
2
29Si
1/
2
43Ca
7/
2
57Fe
1/
2
natural abundance (%)
magnetogyric ratio (10-7 rad T/s)
quadrupole moment (10-28 m2)
relative sensitivity
absolute sensitivity
0.037 100 4.7 0.145 2.19
-3.63 6.98 -5.32 -1.80 0.869
-0.026 0.149 0 -0.05 0
2.91 × 10-2 0.21 7.84 × 10-3 6.40 × 10-3 3.37 × 10-5
1.08 × 10-5 0.21 3.69 × 10-4 9.28 × 10-6 7.38 × 10-7
characterize cements and related materials.7 Given the major components of cement, it would be expected that this material should lend itself to solid-state NMR investigation (Table 1); however, despite the presence of NMR-active isotopes, there are significant challenges to the use of NMR spectroscopy for the characterization of cements. Whereas 29Si and 57Fe are nonquadrupolar nuclei (i.e., spin 1/2) and should therefore provide relatively sharp, well-resolved spectra, the natural abundance of 57Fe, and hence sensitivity, is extremely low. In contrast, 17O, 27Al, and 43Ca have large quadrupole moments, resulting in broad resonances with irregular lineshapes. Furthermore, 17O and 43Ca have very low natural abundances, further reducing the ease of data collection.8-11 Studies of 17O, 57Fe, and 43Ca suffer further still from the nuclei’s exceptionally low magnetogyric ratios, the latter two especially so. As a consequence of these factors, the two main thrusts of the field focus on either 27Al or 29Si NMR spectroscopy. Studies in 27Al NMR spectroscopy of cements fall into two general categories: those that observe the aluminum impurities in C-S-H12-14 and those that focus on the pure calcium aluminate phases.15 Work in 29Si NMR spectroscopy bifurcates along analogous lines, with studies tending to focus on either observing the C-S-H hydration product16,17 or the unreacted calcium silicates.18 Our prior research has used a combination of 27Al and 29Si NMR spectroscopies to investigate the interactions of cements with a range of setting retarders.19,20 One important element of cement NMR spectroscopy that has received little attention is the presence of paramagnetic Fe3+ ions in the C4AF matrix. Paramagnetism’s impact on NMR spectra is well-known, causing signal loss (although specialized techniques can be employed to observe the disappearing signal nearest the paramagnetic centers21), line broadening, and/or chemical shift alteration, as well as decreasing (in many cases dramatically) longitudinal (T1) relaxation times. On occasion, researchers have added paramagnetic species to their samples to maximize the signal-to-noise ratio achievable per unit time.22 Given the inherent presence of paramagnetic centers in cement (i.e., the Fe3+ ions in the C4AF matrix), this work acts on an interest in the potential for obtaining signal-enhanced 29Si NMR data from cements as compared to the pure minerals and using these data as a predictive tool in understanding cement kinetics. This signal enhancement does not come without a price, however. Paramagnetism not only increases possible pulse repetition rates but also broadens the resonances by forcing down the transverse relaxation time T2 and the apparent transverse relaxation time, T2*, which is inversely proportional to the halfwidth of a resonance. The present research focuses on the application of 29Si NMR spectroscopy to systems relating to cements, taking steps to exploit the paramagnetism present where possible. The impact of the paramagnetism on the T1 relaxation curve is discussed, and the use of 29Si NMR spectroscopy as an analytical tool is presented. Experimental Section Fumed silica was obtained from British Petroleum, Inc. The M(acac)3 compounds were obtained from Sigma-Aldrich, Inc.
C3S and C2S samples were obtained from Construction Technology Laboratories, Inc., Skokie, IL. All cements were obtained from Halliburton Energy Services, Duncan, OK. All materials were used as received. 29Si NMR Spectroscopy. Solid-state 29Si MAS NMR spectra were collected on a Bruker Avance 200 spectrometer at 39.76 MHz. Samples were measured using a 7-mm extended VT MAS probe with 7-mm long-barrel ZrO2 rotors and plugs and Kel-F fluoropolymer caps. Chemical shifts were referenced to hexamethylcyclotrisiloxane (δ ) -9.66 ppm) by sample replacement. All spectra were collected at a magic angle spinning speed of 6.00 kHz without high-power 1H decoupling. Spectra were acquired using a 90° 29Si pulse of 5.75 µs and a spectral width of 14 750 Hz unless otherwise specified. Data were acquired either with Bruker Biospin’s XWINNMR, version 2.6, or with TopSpin, version 1.3. Variable-delay Bloch decay spectra were collected for 20 480 scans over a spectral width of 15 kHz. Data were processed using TopSpin version 1.3. Fourier transforms were applied without line broadening, phase corrections were made manually, and a fifth-order polynomial was fit to the baselines across the F2 dimension. T1 relaxation times measured via the standard inversion recovery method were collected initially over a range of delay periods, τ, primarily varying over a millisecond time regime from 1 µs (approximating a value of 0 ms to accommodate pulse program timing considerations) terminating at 1500 ms for 20 000 scans at each value of τ with a relaxation delay of 6 s. After initial processing of the data, the times closest to exhibiting a null crossing of the C2S were identified, and subsequent inversion recovery runs were carried out over the proper time regime to identify the null-crossing time to the nearest 1 ms. Data were processed with Topspin v1.3. Fourier transforms were applied without line broadening, phase corrections were made manually, and a fifth-order polynomial was fit to the baselines across the F2 dimension. T1 relaxation times measured via the saturation recovery method were collected initially over a telescoping range of 16 delay periods, τ, primarily varying over a microsecond and millisecond time regime from 1 µs (approximating a value of 0 s to accommodate pulse program timing considerations) terminating at 3.0 s for 128 dummy scans and 40 000 acquisition scans at each value of τ, with a 400-pulse saturation train with a 2-µs interstitial period and no relaxation delay. A second aliquot of sample was analyzed under time regimes from 1 to 70 s for 5000 scans and from 70 µs to 10 ms for 200 000 scans. Otherwise, these additional saturation recovery experiments were performed under conditions identical to those used for the first run. Data were processed using TopSpin version 1.3. Fourier transforms were applied without line broadening, phase corrections were made manually, and a fifth-order polynomial was fit to the baselines across the F2 dimension. The data were then split into subspectra corresponding to individual values of τ for deconvolution analysis. Deconvolutions were carried out using a SIMPLEX routine integrated into the solids lineshape analysis tool in TopSpin v1.3 with eight Gaussian peaks and at least two optimizations of 1000
5124
Ind. Eng. Chem. Res., Vol. 46, No. 15, 2007
Figure 1. Standard Bloch decay NMR spectrum of a Texas Lehigh cement with (a) 10 240 scans with a relaxation delay of 60 s and (b) 102 400 scans with no relaxation delay.
iterations each. Linewidths and intensities of all peaks were free to be altered by the optimization independently, but no Lorentzian character was permitted. The SIMPLEX routine otherwise used default parameters. After the first lineshape was attempted by visual inspection and optimized, the parameters were copied into subsequent data sets to be used as a starting point for optimization after adjustments for peak intensities by visual inspection. Seven of the eight peaks were attributed to C3S, whereas the remaining peak at -71.4 ppm was attributed to C2S solely. Skibsted et al. used 13 peaks for their analysis of similar lineshapes, with relative intensities of the 12 C3S Gaussian and/or Lorentzian peaks fixed, but the work presented herein found seven peaks to be sufficient to fit the C3S portion of the lineshape in the samples studied.18 Areas of fitted peaks were calculated by multiplying the halfwidths (in hertz) of the fitted peaks by their respective heights (in self-consistent arbitrary units). The C3S content was calculated from the sum of the seven peak areas attributed to C3S. To compare data from runs with differing numbers of scans, the areas from subspectra with matching values of τ were divided, and the average ratio was used as a scaling factor that was then applied to the data set of more extreme τ values. Preparation of M(acac)3-Coated Silicates. Samples of fumed silica (10 g) were loosely poured into a lidless standard glass Petri dish, covered with 1% (m/v) solutions of M(acac)3 (M ) Fe, Cr) in acetone to a loading of 1% M(acac)3 by mass, and then covered with a Kimwipe laboratory tissue. The solutions evaporated under ambient laboratory conditions, depositing the M(acac)3 onto the particles. Similar protocols were used for C3S and C2S substrates.
Whereas trace paramagnetic impurities in the silicate phases might be present, the effect is overwhelmed by the presence of the paramagnetic C4AF matrix. The C2S component of the signal consists of a single symmetric resonance at approximately 71.4 ppm; the remainder of the lineshape is due to the C3S present. This particular spectrum was collected for 10 240 scans with a relaxation delay of 60 s, a typical relaxation delay for a spectrum that would be collected for a standard diamagnetic silicate. The total acquisition time was 7 days, 2 h, 49 min, and 14 s; practical instrument maintenance considerations prohibit collecting spectra for longer durations. Figure 1b shows the 29Si NMR spectrum of the same Texas Lehigh cement sample as shown in Figure 1a, but collected with no relaxation delay whatsoever. The lack of use of such a delay allows the nuclei to relax only during free-induction-decay (FID) acquisition. Were the sample diamagnetic, the resonance would saturate after a handful of scans, and the spectrum collected would rapidly decrease in signalto-noise ratio as noise built up but no signal was generated in subsequent scans. As is apparent from the spectrum, the lineshape is virtually identical to that in Figure 1a, but in this spectrum, 10 times as many scans were collected, all in a mere 1 h, 31 min, and 46 s. The surprisingly high degree of similarity between the spectra in parts a and b of Figure 1 was the catalyst for the interest in NMR spectroscopy as a potentially novel tool for cement analysis. First, it suggests that 29Si NMR spectra of cements can be collected rapidly with no relaxation delay, which would go a long way toward overcoming the mediocre magnetogyric ratio and poor natural abundance of 29Si. Second, because the Fe3+ is present in the C4AF matrix that surrounds the C3S and C2S mineral particles, these minerals feel the effects of the paramagnetic centers. The ramifications of the paramagnetism present due to the C4AF in cements ultimately come down to a reduction of the longitudinal relaxation time, T1. This reduction in T1, which is normally quite long for a typical diamagnetic solid when compared to comparable solution-state relaxation times, allows for a far more rapid pulse repetition rate, which, in turn, builds signal at a much higher rate, thereby allowing for a greater acquisition of signal-to-noise ratio per unit time. Experiments that might otherwise take weeks or even months, as shown in Figure 1, now take mere hours. To qualitatively verify the impact of the paramagnetism on the silicates within cements, we prepared a series of model samples by the addition of a paramagnetic additive to the surface of silica, which is not inherently paramagnetic. Samples of fumed silica (SiO2) were coated with M(acac)3 (I, M ) Fe3+, Cr3+, acac ) acetylacetonate) at a loading of approximately 1% by mass. Iron(III) and chromium(III) were chosen for M for their high paramagnetism and ease of dissolution. The compounds were dissolved in acetone, and the silicate powder substrates were coated with the M(acac)3 solution (see Experimental Section).
Results and Discussion 29Si NMR Spectroscopy of Cement and Model Systems. A basic 29Si NMR MAS spectrum of a cement sample from Texas Lehigh Cement Company, LP, is shown in Figure 1a. The spectrum consists of a single peak with a complex lineshape, arising from the overlapping resonances of the C3S and C2S silicate moieties, broadened by the paramagnetic C4AF matrix.
Figure 2 shows the 29Si MAS spectra of the coated and uncoated fumed silica. As is evident in the spectra, the resonance
Ind. Eng. Chem. Res., Vol. 46, No. 15, 2007 5125
Figure 2. 29Si NMR spectra of fumed silica (a) with Fe(acac)3, (b) with Cr(acac)3, and (c) without added paramagnetic material.
broadens in the presence of the paramagnetic relaxation agent; however, it is the T1 values (measured via the inversion recovery method) that show the greatest effect (see below). As expected, the T1 relaxation values decrease significantly with the addition of a paramagnetic surface species. Whereas fumed silica has a measured T1 value of 62.72 s, the addition of about 1% surface Cr(acac)3 and Fe(acac)3 results in T1 values of 2.24 × 10-4 s and 1.74 × 10-4 s, respectively. Similar effects are seen for the addition of Cr(acac)3 and Fe(acac)3 to other silicates including C3S and C2S. Although the relative magnitude of the reduction in T1 is consistent with the number of unpaired electrons in the M(acac)3 species (i.e., Fe d5> Cr d3), the relative size of the metal complex crystallites and the concentration on the surface of the fumed silica are variables that were not addressed. We have demonstrated that the presence of a paramagnetic species in intimate (physical) contact with an otherwise diamagnetic silicon-containing mineral allows for rapid data collection of the 29Si MAS NMR spectrum. Judging from the model systems, it appears that, for cements, the T1 values of C3S and C2S are significantly altered because of the paramagnetic Fe3+ present in the C4AF matrix that makes intimate contact with each of the silicate phases.4 The presence of the paramagnetic C4AF matrix thus allows for the rapid collection of spectra of cements. Additionally, we can propose that the paramagnetic material can potentially act as a probe of the heterogeneity of the intermingled silicate system. Whereas paramagnetic species are known to eliminate the Zeeman interaction within ∼1 nm, they are also known to influence the relaxation behavior of nearby nuclei outside the 1-nm “dead zone” with an inverse r3 dependence.22 Thus, the smaller a diamagnetic silicon-containing particle within a paramagnetic matrix, the greater the relaxation effect over the whole volume of the particle. In this regard, we propose that alteration of the relaxation of the observed 29Si nucleus can be used as a probe of the relative particle sizes or homogeneities of diamagnetic mineral particles (i.e., C3S and C2S) within a paramagnetic matrix (i.e., C4AF) in a composite material (i.e., a cement grain). Techniques such as X-ray fluorescence and X-ray diffraction effectively measure the content of the entire particle; we propose that NMR relaxation analysis can potentially focus on the collective surfaces of the silicates, the hydration of which forms the backbone of solidified and hardened cements. Figure 3 provides a schematic representation of the relative portions of a particle studied in a paramagnetic matrix. We note that the average size of a silicate grain within the C4AF matrix is 10-15 µm, and thus, an insignificant fraction of the grain is within the dead zone. Comparison of Methods of NMR Relaxation Analysis. Because the early formation of C-S-H (the primary factor
dictating solidification and strength development in cements) depends predominantly on the quantity of C3S available for reaction, gaining useful information about the amount of C3S able to hydrate is of value. Because cement manufacturing is controlled primarily by regulation of silicon content and the ultimate fate of silicon is in either C3S or C2S, knowledge of the C3S/C2S ratio puts the amount of C3S alone in the context of the whole cement. Judging from the premise that an accurate (as well as fast and simple) measurement of the C3S/C2S ratio is important in understanding the potential reaction chemistry of cements, there is naturally an interest in determining the optimum methodology for this data collection. We propose that 29Si MAS NMR spectroscopy offers such a methodology. Of several alternative data collection and analysis methods, we chose both the inversion recovery and saturation recovery methods and variable-delay Bloch decay experiments for investigation. The first technique to be considered was the inversion recovery method. This method utilizes a 180° pulse to put the sample’s net magnetization vector at equilibrium onto the -z axis in the rotating frame of reference. This inversion pulse is followed by the variable delay period (τ) that gives the single varying time to relax back to the +z axis. The magnetization relaxes back to equilibrium according to the equation I(τ) ) I0[1 - 2 exp(-τ/T1)], where I0 represents the maximum signal intensity and I(τ) is the intensity at a particular variable delay period (τ). Figure 4 shows a graphical representation of the impact of τ on the intensity of a given peak. At very short values of τ, the vector is inverted; as τ increases, the negative vector diminishes, reaches a null point when τ equals T1 ln(2), and then grows in the positive direction to reach an effective maximum at τ ≈ 5T1. At the end of the delay period, a 90° pulse is applied to move the z component of the net magnetization vector into the xy plane of the rotating frame, allowing for observation and acquisition of a scan. Inversion recovery is the preferred method for analyzing the T1 relaxation times of most ordinary (i.e., homogeneous and diamagnetic) samples, as the magnetization vector is under tightest control and thus tends to give the most accurate measurements. The downside of inversion recovery in the general case comes from the requirement of a relaxation delay of at least 5 times the expected value of T1, resulting in long waiting periods between acquisitions and imposing a practical limit on the amount of signal that can be acquired in a reasonable time frame. The conventional alternative to inversion recovery for measuring T1 relaxation times is the saturation recovery method. In this method, the sample signal is saturated by a pulse train, a rapid succession of pulses applied to eliminate the net magnetization vector. The pulse train is then followed by a varying delay period τ during which the vector recovers from saturation back toward equilibrium on the +z axis. The vector relaxes back to equilibrium according to the equation I(τ) ) I0[1 exp(-τ/T1)]. Figure 5 shows a graphical representation of the impact of τ in a saturation recovery experiment on the intensity of a given peak. At the end of the recovery period, a 90° pulse is applied to put the z component of the net magnetization vector into the xy plane of the rotating frame, allowing for observation and acquisition of a scan. The saturation recovery method has the advantage of generating peaks of only non-negative intensity. This characteristic becomes particularly advantageous when dealing with overlapping signals, where the guarantee of positive intensity from all observable signals eliminates the complication of overlapping positive and negative peaks, which often produce undesirably complex lineshapes. Additionally, this
5126
Ind. Eng. Chem. Res., Vol. 46, No. 15, 2007
Figure 3. Detection fractions of a particle (dark green) versus the remainder of the particle (red) (a) when studied by X-ray fluorescence or X-ray diffraction and (b) when evaluated by NMR T1 relaxation analysis, allowing for more relaxation to occur from left to right.
Figure 4. Graphical representation of the impact of the variable delay (τ) on the intensity of a given peak in an inversion recovery experiment for a typical diamagnetic, homogeneous material. Figure 6. Inversion recovery spectrum of cement sample CAP.
Figure 5. Graphical representation of the impact of the variable delay (τ) on the intensity of a given peak in a saturation recovery experiment for a typical diamagnetic, homogeneous material.
Figure 7. Subspectrum of the inversion recovery data in Figure 6 illustrating the superposition of the positive C3S signal and the negative C2S signal for the CAP cement sample.
method provides a clearer expression of the impact of paramagnetic species on the relaxation behavior of a heterogeneous material, in the same line of analysis as that used by Hunter and co-workers.23 The primary disadvantage of this method comes from the difficulty in establishing a pulse train sufficient to completely saturate the net magnetization vector prior to the recovery period. Relaxation behavior can also be explored more obliquely through a variable-delay Bloch decay experiment, wherein standard Bloch decay spectra are acquired (after sufficient dummy scans are run to establish a steady state of spin relaxation) with varying relaxation delays. In this way, the spectra can approximate the data collected in the saturation recovery runs, with the sum of the relaxation delay and duration of the FID approximately equating to the saturation recovery values of τ. This method has the advantage of acting as a screening tool for streamlined analysis if a given saturation recovery spectrum is found to be particularly effective at predicting hydration behavior. The greatest disadvantage of this method of characterization is its inability to evaluate relaxation values less than the duration of the FID.
To demonstrate the application of these three relaxation analysis techniques, an exemplar sample of cement was chosen for analysis. Capitol Aggregates, Ltd.’s Cement Division (CAP) is an API class H cement produced in the San Antonio (Texas) area. Its general relaxation behavior is not unlike that of any cement and can readily serve to illustrate the capabilities of these techniques for studying cements. Inversion Recovery Method. Figure 6 shows the pseudo2D inversion recovery spectrum of a sample of CAP cement. The spectrum shows, at long and short values of τ, the (inverted) signal and typical overlapping lineshape expected of oil well cements (Figure 1b). However, at intermediate values of τ, C3S has begun to pass its null point, whereas C2S remains in the negative intensity regime. The result is a lineshape with both positive and negative intensity simultaneously, greatly complicating further analysis and extraction of useful data. A subspectrum illustrating this phenomenon is shown in Figure 7. In an effort to overcome this limitation, one might begin with the assumption made by Skibsted et al.18 that, at the point when the C2S signal is nulled, the C3S signal has fully relaxed. From this assumption, one can isolate the spectrum at the null-crossing
Ind. Eng. Chem. Res., Vol. 46, No. 15, 2007 5127 Table 3. Processed Deconvolution Data from CAP Saturation Recovery Experiments
Figure 8. Subspectrum subtraction for the estimation of C3S and C2S contents for the CAP cement sample: (c) Fully relaxed subspectrum; (b) C2S-nulled subspectrum; and (a) difference, showing C2S and residual C3S.
τa
aliquot
C3S areab
C2S areab
C3S/C2S area ratio
100 µs 300 µs 700 µs 1 ms 1 ms 3 ms 3 ms 7 ms 7 ms 10 ms 10 ms 30 ms 70 ms 100 ms 300 ms 1s 1s 3s 3s 7s 10 s 30 s 70 s
1 1 1 1 2 1 2 1 2 1 2 1 1 1 1 1 2 1 2 2 2 2 2
202 290 037.4 611 906 609.3 1 332 946 612 1 770 911 561 10 535 829 174 4 468 713 517 25 683 154 738 9 552 959 170 47 182 092 925 9 552 959 170 58 114 533 149 16 012 957 734 20 539 462 783 21 930 141 454 25 264 567 156 25 708 504 866 2 031 695 637 25 731 329 115 1 993 516 859 2 005 679 975 2 065 699 229 2 033 206 930 2 007 320 582
9 265 950.67 58 015 759.08 104 992 819.6 164 778 021.7 893 374 367.6 327 496 167 2 482 057 772 858 812 883.4 4 443 310 535 858 812 883.4 6 010 770 237 1 695 664 777 2 634 809 641 3 207 932 589 4 156 251 695 5 011 852 595 397 032 550.5 5 192 454 848 420 325 357.5 420 353 390.4 420 717 131.5 407 086 244.9 435 101 172.2
21.831 546 98 10.547 248 18 12.695 597 82 10.747 255 88 11.793 296 92 13.645 086 47 10.347 524 96 11.123 446 51 10.618 680 05 11.123 446 51 9.668 400 364 9.443 468 989 7.795 425 698 6.836 222 659 6.078 690 371 5.129 541 299 5.117 201 688 4.955 522 94 4.742 794 656 4.771 413 816 4.909 947 978 4.994 536 061 4.613 457 078
a First aliquot with central τ values; second aliquot with extreme τ values. Area calculated by multiplying peak intensity (arbitrary units) by peak half-width (Hz) without rescaling to connect intermediate and extreme τ regimes.
b
Figure 9. Saturation recovery spectrum of cement sample CAP. Table 2. Raw Data from Deconvolution of the Saturation Recovery Subspectrum CAP (3 s) peak ID
peak attribution
chemical shift (ppm)
peak intensity (arbitrary units)
peak widtha (Hz)
1 2 3 4 5 6 7 8
C3S 1 C3S 2 C3S 3 C2S C3S 4 C3S 5 C3S 6 C3S 7
-67.381 -68.994 -70.166 -71.166 -72.145 -73.231 -74.543 -76.948
27 606 352 35 592 161 59 278 642 91 273 950 65 171 374 49 386 078 36 647 253 17 194 682
187.56 83.28 62.99 54.91 59.01 70.42 99.23 167.06
a
Half-width of Gaussian peak.
point for C2S (which is representative of the C3S content) and subtract it from the fully relaxed spectrum in order to estimate the amount of C2S present in the sample. Figure 8 provides an illustration of this process. This method, although ingenious in concept, is ultimately flawed, however, as the C2S component is nulled at a point prior to the C3S component’s full relaxation, as seen in Figure 7. Thus, inversion recovery is an unacceptable model for relaxation analysis of cements in the general case. Overlapping positive and negative peaks in subspectra make deconvolution extremely difficult at best and impossible at worst. Methods to work around the need for deconvolution by
exploitation of differential relaxation times of C2S and C3S are unreliable at best. Saturation Recovery Method. In light of the complications observed in the inversion recovery spectra, consideration now moves to the method of saturation recovery. Figure 9 shows the pseudo-2D saturation recovery spectrum of CAP. It is noteworthy that the subspectra contain only non-negative peaks and that the lineshape changes as the relative contributions of C3S and C2S change with increasing recovery period τ. As this data are relatively well-behaved compared to the inversion recovery data, a more detailed analysis becomes readily accessible. To this end, deconvolution analysis was applied to each subspectrum to isolate the C2S signal from the C3S signal. Whereas Skibsted et al.18 used deconvolution analysis on fully relaxed cement samples and observed correlation with quantification via XRD, this work applied deconvolution throughout the saturation recovery analysis to track the separate relaxation regimens of the two minerals. Table 2 contains an example of the output of the deconvolution optimization used in further processing. Figure 10 shows the subspectra of the saturation recovery spectrum focused on the region of greatest change in signal. Owing to practical constraints on time, data corresponding to extremely low values of τ are often too low in signal-tonoise ratio to detect in the same number of scans as data of intermediate value, and data corresponding to higher, multisecond values of τ are too time-intensive to take the number of scans desired at intermediate values of τ. To attempt to collect these data in a practical time frame, the extreme values of τ are collected in their own separate experiments, with values overlapping the intermediate range serving as a rescaling tool to combine the data sets. To correlate the data from one set to another, overlapping values of τ are selected and used to find average ratios between extreme and intermediate data sets. The collected data are given in Table 3. Figure 11 shows a plot of the rescaled data illustrating the unification of the series. Having obtained a full saturation recovery curve of cement, deconvolved
5128
Ind. Eng. Chem. Res., Vol. 46, No. 15, 2007
Figure 10. Subspectra of the saturation recovery data for sample CAP (first aliquot) at τ values of (a) 100 µs, (b) 300 µs, (c) 700 µs, (d) 1 ms, (e) 3 ms, (f) 7 ms, (g) 10 ms, (h) 30 ms, (i) 70 ms, (j) 100 ms, (k) 300 ms, (l) 1 s, and (m) 3 s, illustrating the superposition of the fitted peaks forming the C3S (blue) and C2S (red) contributions resulting from deconvolution (summation in green). Table 4. Summary of Data Analysis Methodologies method
advantages
disadvantages
inversion recovery
tight control over net magnetization vector, simple pulse sequence
saturation recovery
no negative signals, simple conceptualization of data, accessible to deconvolution simplest experimental setup, translates best into standardized testing
variable-delay Bloch decay experiments
the signal into individual components, performed derivative analysis on the regions of greatest change in the individual components, and standardized for disparate relative intensities among different acquisition regimes, a full characterization of the relaxation behavior has now been found. Bloch Decay Experiment. In the interest of finding a simple experiment that can be used readily by a technician to evaluate
overlapping positive and negative peaks, spectrum algebra thwarted by too similar T1 values, very time-intensive optimization of saturation pulse train required, narrower range of signal intensities possible fundamental lower limit on accessible relaxation regimes
the properties of a particular sample of cement, we develop here the concept of the variable-delay Bloch decay experiment as a screening tool to identify simple experiments with useful delays. Figure 12 shows the pseudo-2D variable-delay Bloch decay spectrum of CAP. It is noteworthy that, unlike the saturation recovery spectrum, this data set never fades into the baseline noise as the delay time approaches zero. The time required to
Ind. Eng. Chem. Res., Vol. 46, No. 15, 2007 5129
Figure 11. Rescaled extreme τ data sets combined with intermediate τ data for CAP. The three series are color coded for short (green), medium (red), and long (blue) τ data for C3S and C2S.
Figure 12. Variable-delay Bloch decay spectrum of cement sample CAP.
collect the FID (in this case, ∼30 ms) for each subspectrum provides a firm limit to the lowest observable relaxation period. Determination of C3S/C2S Ratio. Predictions of the set kinetics, reaction to inhibitors, and properties of the set cement are presently not available. Instead of predictions, a series of experiments must be performed on each sample to obtain the property information required. It would be desirable to have a simple analytical method (or short series of methods) that would allow for the prediction of the chemistry of individual samples of cement from various sources. In this regard, we have demonstrated that the saturation recovery method for analysis of 29Si NMR data offers the best combination of accuracy, speed, and ease of processing. Figure 13 shows a schematic representation of the analysis volume as a function of relaxation time for 29Si NMR experiments using the saturation recovery method. As the relaxation time is increased, the volume fraction of the silicate particle (crystallite) being “observed” increases until the entire particle is being sampled. Although the integration for each mineral does not provide an absolute concentration or weight percentage, the processed peak areas attributed to the C2S and C3S contents do
Figure 13. Scaled signal intensity for C3S and C2S as a function of NMR relaxation (τ) showing the detection fractions of a particle (dark green) versus the remainder of the particle (red).
allow for an accurate derivation of the C3S/C2S ratio. Using the data from the longest relaxation times (i.e., after the intensity curve in Figure 13 plateaus) provides the relative concentrations of C3S and C2S. Thus, the C3S/C2S ratio can be calculated by a consideration of the intensity ratio at τ values above which the slope in the plot of intensity of the various saturation recovery parameters plotted versus τ is essentially 0 (e.g., τ g 1 s in Figure 11). In all cases for cements we have studied, this involved τ above 1 s. We propose that the errors associated with our NMR method are limited to measurement errors and variability of the sample as taken for NMR spectroscopy. The saturation recovery data were measured on at least two samples taken at random from the cement batch. The two samples were chosen on different days (often 60-120 days apart). The overlap of the low and central data and the high and central data shown in Figure 11 suggests that variability between samples is low. For example, from Table 3, a variation due to sample choice of about 2-3% can be calculated. An estimate of the error inherent in the NMR measurements can be determined by a consideration of the C3S/ C2S ratio calculated for a particular sample at τ values above which the slope in the plot of the intensity of the various saturation recovery parameters plotted versus τ is essentially 0 (cf., Figure 11). Using the data from Table 3, an error of 4% is calculated. To ascertain the generality of our method and compare the results with conventional XRF-based oxide analyses, various cement samples were analyzed by 29Si NMR experiments using the saturation recovery method with τ between 1 and 70 s. In the present work, eight different cements were studied. Each sample was stored in a dry box under argon prior to NMR analysis. Three of the samples were API class G cements from foreign countries: Brazil (BRZ), Russia (RUS), and Thailand (THA). The remaining five samples were API class H cements from domestic cement manufacturers: Lafarge (Joppa plant) (JOP), Texas Industries (TXI), El Toro (ELT), Southdown (SDN), and Capitol (CAP). Class G and H cements both use the same specifications for elemental composition and restrict grinding aids to gypsum, but they differ with respect to fineness, with class G cements being the more finely ground of the two. A summary of the results is provided in Table 3 and in Tables S2-S8 in the Supporting Information. This work was motivated in large part by the lack of utility of conventional XRF-based oxide analyses in predicting cement performance properties. Figure 14 shows a plot of C3S/C2S ratios
5130
Ind. Eng. Chem. Res., Vol. 46, No. 15, 2007
Figure 14. Plot of the “active” C3S/C2S ratio (as determined by MAS 29Si NMR spectroscopy) versus the classical C3S/C2S ratio (as determined by XRF) for the cements studied herein.
determined by MAS 29Si NMR spectroscopy versus the calculated classical C3S/C2S ratios based on oxide analysis from X-ray fluorescence and application of the Taylor-modified Bogue equations. As can be seen from Figure 14, the two parameters show little relationship. Oxide analysis appears to provide a lower C3S/C2S ratio than determined by NMR analysis. Furthermore, oxide analysis suggests that all of the cements have C3S/C2S ratios of 2-5, whereas our NMR method suggests that the actual range is significantly broader. Conclusions The introduction of paramagnetic species into an otherwise diamagnetic system can dramatically decrease the longitudinal relaxation time T1. In the case of cements, the T1 values of C3S and C2S can change because of the Fe3+ present in the C4AF matrix that makes intimate contact with the silicate phases. Three methods of NMR relaxation have been examined and compared herein. Inversion recovery, although desirable for most compounds, has severe limitations when examining cements. Variable-delay Bloch decay experiments offer a simple experimental setup but obscure behavior most characteristic of surface behavior. Saturation recovery provides the best balance of theoretical simplicity and spectral tractability with minimal disadvantage. Table 4 contains a summary of the advantages and disadvantages of the three methodologies. Determination of C3S/C2S ratios by NMR spectroscopy provides an effective method of analysis for cements, owing to NMR spectroscopy’s direct measurement of the minerals in question without the assumptions of composition that are required for oxide analysis from X-ray fluorescence and application of the Taylor-modified Bogue equations. Given the importance of C3S and C2S (and hence their ratio) in defining the performance of cements, we propose that 29Si MAS NMR spectroscopy represents a viable method for cement characterization. Acknowledgment Financial support for this work was provided by Halliburton Energy Services and the Robert A. Welch Foundation. The Bruker Avance 200 NMR spectrometer was purchased with funds from ONR Grant N00014-96-1-1146. Supporting Information Available: Figures and tables that provide more detailed information regarding 29Si NMR data, as well as inversion and saturation recovery of C3S, C2S, and SiO2. (PDF.) This material is available free of charge via the Internet at http://pubs.acs.org. Literature Cited (1) Gadalla, M. Pyramid Illusions; Bastet Publishing: Erie, PA, 1997.
(2) Smith, D. K. Cementing; Society of Petroleum Engineers: Richardson, TX, 1990. (3) Baker, R. A Primer of Oil Well Drilling, 5th ed.; Petroleum Extensions Service: Austin, TX, 1996. (4) Taylor, H. F. W. Cement Chemistry; Thomas Telford Publishing: London, 1997. (5) ASTM C 856, Standard Practice for Petrographic Examination of Hardened Concrete. In Annual Book of ASTM Standards 4.02, Concrete and Aggregates; ASTM International: West Conshohocken, PA. (6) Hall, C.; Scrivener, K. L. Oilwell Cement Clinkers-X-ray Microanalysis and Phase Composition. AdV. Cem.-Based Mater. 1998, 7, 2838. (7) Wieker, W.; Hubert. C.; Heidemann. D. Recent results of solid-state NMR investigations and their possibilities of use in cement chemistry. Proc. Int. Congr. Chem. Cem., 10th 1997, 1, 2. (8) Emery, J.; Massiot, D.; Lacorre, P.; Laligant, Y.; Conder, K. 17O NMR in room temperature phase of La2Mo2O9 fast oxide ionic conductor. Magn. Reson. Chem. 2005, 43, 366. (9) Trokiner, A.; Le Noc, L.; Yakubovskii, A.; Mykhalyov, K. N.; Verkhovskii, S. V. 43Ca NMR study of bismuth-based High-Tc Superconductors. Z. Naturforsch. 1993, 49a, 373. (10) Lin, Z. J.; Smith, M. E.; Sowrey, F. E.; Newport, R. J. Probing the local structural environment of calcium by natural-abundance solid-state 43Ca NMR. Phys. ReV. B 2004, 69, 224107/1. (11) Dupree, R.; Howes, A. P.; Kohn, S. C. Natural abundance solid state 43Ca NMR. Chem. Phys. Lett. 1997, 276, 399. (12) Schneider, J.; Cincotto, M.; Panepucci, H. 29Si and 27Al highresolution NMR characterization of calcium silicated hydrate phases in activated blast-furnace slag pastes. Cem. Concr. Res. 2001, 31, 993. (13) Faucon, P.; Charpentier, T.; Bertrandie, D.; Nonat, A.; Virlet, J.; Petit, J. C. Characterization of calcium aluminate hydrates and related hydrates of cement pastes by 27Al MQ-MAS NMR. Inorg. Chem. 1998, 37, 3626. (14) Faucon, P.; Charpentier, T.; Nonat, A.; Petit, J. C. Triple-quantum two-dimensional 27Al magic angle nuclear magnetic resonance study of the aluminum incorporation in calcium silicate hydrates. J. Am. Chem. Soc. 1998, 120, 12075. (15) Skibsted, J.; Henderson, E.; Jakobsen, H. Characterization of calcium aluminate phases in cements by 27Al MAS NMR spectroscopy. Inorg. Chem. 1993, 32, 1013. (16) Haga, K.; Shibata, M.; Hironaga, M.; Tanaka, S.; Nagasaki, S. Silicate anion structural change in calcium silicate hydrate gel on dissolution of hydrated cement. J. Nucl. Sci. Technol. 2002, 39, 540. (17) Johansson, K.; Larsson, C.; Antzutkin, O. N.; Forsling, W.; Kota, H. R.; Ronin, V. Kinetics of the hydration reactions in the cement paste with mechanochemically modified cement 29Si magic-angle-spinning NMR study. Cem. Concr. Res. 1999, 29, 1575. (18) Skibsted, J.; Jakobsen, H.; Hall, C. Qunatification of calcium silicate phases in Portland cements by 29Si MAS NMR spectroscopy. J. Chem. Soc., Faraday Trans. 1995, 91, 4423. (19) Bishop, M.; Bott, S. G.; Barron, A. R. A new mechanism for cement hydration inhibition: Solid-state chemistry of calcium nitrilotris(methylene)triphosphonate. Chem. Mater. 2003, 15, 3074. (20) Bishop, M.; Barron, A. R. Cement hydration inhibition with sucrose, tartaric acid, and lignosulfonate: Analytical and spectroscopic study. Ind. Eng. Chem. Res. 2006, 45, 7042. (21) Furman, G. B.; Goren, S. D. Nuclear magnetic resonance detection of the nuclear spins near paramagnetic impurities in solids. J. Phys.: Condens. Matter 2002, 14, 873. (22) Fujiu, T.; Ogino, M. 29Si NMR study on the structure of leadsilicate glasses. J. Non-Cryst. Solids 1984, 64, 287. (23) Chudek, J. A.; Hunter, G.; McQuire, G. W.; Rochester, C. H.; Smith, T. F. S. 29Si Magic-angle spinning NMR study of Phillips catalysts: Multiexponential spin-lattice relaxation resulting from adsorption of paramagnetic chromium species onto a silica surface. J. Chem. Soc., Faraday Trans. 1996, 92, 453.
ReceiVed for reView February 7, 2007 ReVised manuscript receiVed May 11, 2007 Accepted May 19, 2007 IE070220M
