- Email: [email protected]
A reply to further comments by D. Bonen on the paper “quantitative backscattered electron analysis of cement paste”
CEMENT and CONCRETE RESEARCH. Vol. 24, pp. 186-188, 1994. Printed in the USA. 0008-8846/94. $6.00+00. Copyright © 1993 Pergamon Press Ltd.
A R E P L Y TO F U R T H E R COMMENTS BY D. BONEN ON THE PAPER " Q U A N T I T A T I V E B A C K S C A T T E R E D E L E C T R O N ANALYSIS OF C E M E N T PASTE"1
Hong Zhao2 and David Darwin Structural Engineering and Materials Laboratory 2006 Learned Hall, University of Kansas Lawrence, Kansas 66045
The authors appreciate Bonen's continued interest in the subject of quantitative backscattered electron imaging of cement paste. Bonen has doubts about the applicability and reproducibility of the process. However, since the original research (1, 2), thousands of images have been successfully acquired and analyzed using the techniques described in the paper. The procedures described are specifically designed for imaging large areas of cement paste to accurately obtain information on the morphology of the material. The statistical justification for imaging large areas was presented in the paper (2) and re-emphasized in our reply (3) to Bonen's earlier discussion (4). The need for the calibration procedure is based on the fact that, in a multi-user facility, scanning electron microscope settings change from day to day. In addition, within a single viewing session, the beam current can be expected to change over time due to changes that occur in the SEM filament. The calibration procedure (1, 2) allows the operator to correct the microscope settings and helps insure that the grey levels obtained from backscattered electron (BSE) imaging consistently represent the same backscattered electron coefficients by adjusting the output of the processed signal. While this process may be done by eye, a random error will be superimposed on the resulting data, even when the adjustment is made by an experienced operator. The procedures described in the paper (2) involve the identification of specific phases within hydrated cement paste based on grey level segmentation (referred by Bonen as binary segmentation). The grey levels selected to segment the image are initially operator selected. Once set, however, the calibration procedure allows the grey levels to be reproduced faithfully for multiple images of the same specimen and for similar specimens of the same material. Bonen feels that it is "good practice to redefine the grey levels of the phases for every new specimen mounted in the SEM" due to the dependance of interaction volume on specimen orientation and because "backscattered coefficients are defined only for an electron beam incident normal to the surface." The authors disagree. The reason is that, while the interaction volume does depend on the specimen orientation, the volume is relatively insensitive for small changes in angle, 1 CCR 22(4) 695-706 (1992) 2 Current address: RMS2 Engineers Design Group, Inc., 3785 NW 82nd Ave., Suite 209, Miami, Florida 33166 186
Vol. 24, No. 1
DISCUSSIONS
187
as are the backscattered electron coefficients (5). For the materials in hydrated cement paste, the effect of an angle change of up to 8 degrees would have an imperceptible effect on the contrast between the phases (6). Therefore, for reasonably well prepared and mounted flat specimens, the backscatter coefficients and relative grey levels produced by the backscattered electron signal are not sensitive to minor variations in specimen orientation, nor do they need to be redefined for every new specimen mounted in the SEM. Bonen does not agree with the authors belief that expanding the grey levels provides more information and allows easier identification of phases. He states that "if there are minima in the grey level histogram, discrimination between phases is straightforward" but that "where no minima exists, expanding the grey level would do very little good, if any, to resolve this problem." He then goes on to state that "lack of well-defined peaks does not prevent discrimination among phases, as, for example, by a binary segmentation technique." The authors believe that Bonen is missing an important point. When the contrast in the image is reduced, the ability of the operator to segment the image is decreased. The process of using a reduced image contrast throws away useful information. This is information that is needed to initially segment the grey levels or to "redefine the grey l e v e l s . . , for every new specimen," as advocated by Bonen. Since the ability of the human eye to distinguish differences in grey level is quite limited, any reduction in the available contrast between phases of interest reduces the reproducibility of any results based on that grey level range. Therefore, the process of initially segmenting the grey levels should be done with the highest possible contrast for the phases of interest. The authors agree that phase analysis based on grey level segmentation may involve error. However, the desire is to keep that error to a minimum. The authors agree with Bonen that when backscattered electron coefficients overlap, x-ray analysis is needed to identify individual phases. Bonen disagrees with the authors' statement that chemical image analysis provides poor spacial resolution, on the order of 5 to 10 micrometers. He states that "the resolution of chemical image analysis is directly related to the resolution attained by the binary segmentation process, and is about the same order as that of the backscattered imaging." Bonen is partially correct, and then only if the images are acquired at low enough magnification and with low SEM accelerating voltage that the lateral dimensions of the interaction volume are no larger than twice the pixel size. Under these circumstances, the x-ray sampling volume (nearly equal to the interaction volume) will be optimally sized with respect to the pixel (7). At an accelerating voltage of 10 kV, the depth of the interaction volume (or electron range) for CSH is approximately 1.6 ~tm; the lateral dimensions of the interaction volume are about one-half of the electron range. At 25 kV, the electron range is about 7.6 Bm. For backscattered electrons, the sampling depth is about one-third of the electron range (7). For a low density material like CSH, the lateral dimensions of the BSE sampling are about one-tenth of the electron range. As a result of the differences in the dimensions of the respective sampling volumes, x-ray spectra are generated from a region that is significantly larger than that from which backscattered electrons are generated and can represent materials that are significantly different from those in the phases at the specimen surface. The lower the magnification and the larger the pixel size, the less significant are differences in the lateral dimensions of the sampling volumes since resolution is governed by the larger of the lateral dimension of the sampiing volume or the pixel size (7). Under any circumstances, x-ray spectra will be generated at a greater depth than backscattered electrons. Another point to keep in mind is that quantitative energy dispersive spectrometry is based on the assumption that the material being analyzed is homogeneous - an assumption with questionable grounding for hydrated cement paste. Therefore, chemical image analysis will produce an image in which individual phases can be identified only approximately and from which the relative strength of the signal can be used to identify approximate phase boundaries. Backscattered electron imaging can be used in conjunction with chemical image analysis to sharpen the phase boundaries. As stated earlier (3), the authors believe that chemical image analysis is a powerful technique. However, the technique has its own drawbacks, since it takes much more time and provides poorer spatial resolution than backscattered electron imaging. For work carded out by the
188
H. Zhao and D. Darwin
Vol. 24, No. 1
authors, the time required for imaging is important because of the large number of images needed to obtain a statistically valid sample. Due to long acquisition time requirements, such a statistically valid sample is virtually impossible to obtain with chemical image analysis, which is best used for studying specific details in cement paste morphology. The authors agree that, by using a segmentation process, phases can be identified reasonably well; however, the uncertainty of the position of the phase boundaries, based solely on chemical image analysis, is, as a general rule, greater than that obtained from backscattered electron imaging. For that reason, the two processes are usually used together (8). The authors appreciate Bonen's persistence in trying to establish the terminology used to describe the phases observed in polished specimens of hydrated cement paste. Diamond and Bonen (9) have developed new terminology that was not available to the authors at the time of the initial reply (3). At the highest level, Diamond and Bonen categorize the morphology of hydrated cement paste as consisting of phenograins (meaning distinct grains) and groundmass (everything else). Bonen states that "CH has been found to form intimate mixtures with the amorphic CSH particles confined to the groundmass, and not with the outer hydration CSH shell of phenograin or fully hydrated phenograins... " The "outer hydration shell of the phenograin" was referred to in the paper (2) and in the authors' reply (3) as "inner product." The authors statement that "a significant portion of the inner product has the same signal intensity as massive regions of calcium hydroxide" is based on calibration settings selected to optimize the identification of the hydrated phases. The statement means to convey a description of the signal within the region normally referred to as inner product. With the settings used, the grey level range of some regions within the "inner product"/"outer hydration shell of the phenograin" overlap those of clearly identifiable regions of CH surrounded by "outer product"/"groundmass." Calcium hydroxide has been observed within the boundaries of the inner product region by Rayment et al. (10, 11). The CH in this region is likely in a microcrystaUine form, as observed by Groves (12), at a scale below 100 nm. At this scale, the differences in the sampling volumes of backscattered electrons and x-rays will make differences in mean atomic number easier to identify than differences in chemistry. References
1. Zhao, H. and Darwin, D., SM Report No. 24, Univ. of Kansas Center for Research, Lawrence, Kansas (1990). 2. Zhao, H. and Darwin, D., Cem. Concr. Res. 22, 695 (1992). 3. Zhao, H. and Darwin, D., Cem. Concr. Res. 23,754 (1993). 4. Bonen, D., Cem. Concr. Res. 23, 749 (1993). 5. Newbury, D. E., Yakowitz, H. and Myklebust, R. L., Appl. Phys. Lett., 23, 488 (1973). 6. Arnal, F., Verdier, P. and Vincinsini, P.-D., C. R. Acad. Sci. Paris, 268, 1526 (1969). 7. Goldstein, J. I., Newberry, D. E., Echlin, P., Joy, D. C., Romig, A. D., Lyman, C. E., Fiori, C. and Lifshin, E., Scanning Electron Microscopy and X-Ray Microanalysis, 2nd Ed., Plenum Press, New York and London. 8. Bonen, D. and Diamond, S., Mat. Res. Soc. Symp. Proc. 245,291 (1992). 9. Diamond, S. and Bonen, S., J. Am. Ceramic Soc. 76 (1993) (in press). 10. Rayment, D. L. and Majumdar, A. J., Cem. Concr. Res. 12, 753 (1982). 11. Rayment, D. L. and Lachowski, E. E., Cem. Concr. Res. 14, 43 (1984). 12. Groves, G. W., Cem. Concr. Res. 11,713 (1981).
A R E P L Y TO F U R T H E R COMMENTS BY D. BONEN ON THE PAPER " Q U A N T I T A T I V E B A C K S C A T T E R E D E L E C T R O N ANALYSIS OF C E M E N T PASTE"1
Hong Zhao2 and David Darwin Structural Engineering and Materials Laboratory 2006 Learned Hall, University of Kansas Lawrence, Kansas 66045
The authors appreciate Bonen's continued interest in the subject of quantitative backscattered electron imaging of cement paste. Bonen has doubts about the applicability and reproducibility of the process. However, since the original research (1, 2), thousands of images have been successfully acquired and analyzed using the techniques described in the paper. The procedures described are specifically designed for imaging large areas of cement paste to accurately obtain information on the morphology of the material. The statistical justification for imaging large areas was presented in the paper (2) and re-emphasized in our reply (3) to Bonen's earlier discussion (4). The need for the calibration procedure is based on the fact that, in a multi-user facility, scanning electron microscope settings change from day to day. In addition, within a single viewing session, the beam current can be expected to change over time due to changes that occur in the SEM filament. The calibration procedure (1, 2) allows the operator to correct the microscope settings and helps insure that the grey levels obtained from backscattered electron (BSE) imaging consistently represent the same backscattered electron coefficients by adjusting the output of the processed signal. While this process may be done by eye, a random error will be superimposed on the resulting data, even when the adjustment is made by an experienced operator. The procedures described in the paper (2) involve the identification of specific phases within hydrated cement paste based on grey level segmentation (referred by Bonen as binary segmentation). The grey levels selected to segment the image are initially operator selected. Once set, however, the calibration procedure allows the grey levels to be reproduced faithfully for multiple images of the same specimen and for similar specimens of the same material. Bonen feels that it is "good practice to redefine the grey levels of the phases for every new specimen mounted in the SEM" due to the dependance of interaction volume on specimen orientation and because "backscattered coefficients are defined only for an electron beam incident normal to the surface." The authors disagree. The reason is that, while the interaction volume does depend on the specimen orientation, the volume is relatively insensitive for small changes in angle, 1 CCR 22(4) 695-706 (1992) 2 Current address: RMS2 Engineers Design Group, Inc., 3785 NW 82nd Ave., Suite 209, Miami, Florida 33166 186
Vol. 24, No. 1
DISCUSSIONS
187
as are the backscattered electron coefficients (5). For the materials in hydrated cement paste, the effect of an angle change of up to 8 degrees would have an imperceptible effect on the contrast between the phases (6). Therefore, for reasonably well prepared and mounted flat specimens, the backscatter coefficients and relative grey levels produced by the backscattered electron signal are not sensitive to minor variations in specimen orientation, nor do they need to be redefined for every new specimen mounted in the SEM. Bonen does not agree with the authors belief that expanding the grey levels provides more information and allows easier identification of phases. He states that "if there are minima in the grey level histogram, discrimination between phases is straightforward" but that "where no minima exists, expanding the grey level would do very little good, if any, to resolve this problem." He then goes on to state that "lack of well-defined peaks does not prevent discrimination among phases, as, for example, by a binary segmentation technique." The authors believe that Bonen is missing an important point. When the contrast in the image is reduced, the ability of the operator to segment the image is decreased. The process of using a reduced image contrast throws away useful information. This is information that is needed to initially segment the grey levels or to "redefine the grey l e v e l s . . , for every new specimen," as advocated by Bonen. Since the ability of the human eye to distinguish differences in grey level is quite limited, any reduction in the available contrast between phases of interest reduces the reproducibility of any results based on that grey level range. Therefore, the process of initially segmenting the grey levels should be done with the highest possible contrast for the phases of interest. The authors agree that phase analysis based on grey level segmentation may involve error. However, the desire is to keep that error to a minimum. The authors agree with Bonen that when backscattered electron coefficients overlap, x-ray analysis is needed to identify individual phases. Bonen disagrees with the authors' statement that chemical image analysis provides poor spacial resolution, on the order of 5 to 10 micrometers. He states that "the resolution of chemical image analysis is directly related to the resolution attained by the binary segmentation process, and is about the same order as that of the backscattered imaging." Bonen is partially correct, and then only if the images are acquired at low enough magnification and with low SEM accelerating voltage that the lateral dimensions of the interaction volume are no larger than twice the pixel size. Under these circumstances, the x-ray sampling volume (nearly equal to the interaction volume) will be optimally sized with respect to the pixel (7). At an accelerating voltage of 10 kV, the depth of the interaction volume (or electron range) for CSH is approximately 1.6 ~tm; the lateral dimensions of the interaction volume are about one-half of the electron range. At 25 kV, the electron range is about 7.6 Bm. For backscattered electrons, the sampling depth is about one-third of the electron range (7). For a low density material like CSH, the lateral dimensions of the BSE sampling are about one-tenth of the electron range. As a result of the differences in the dimensions of the respective sampling volumes, x-ray spectra are generated from a region that is significantly larger than that from which backscattered electrons are generated and can represent materials that are significantly different from those in the phases at the specimen surface. The lower the magnification and the larger the pixel size, the less significant are differences in the lateral dimensions of the sampling volumes since resolution is governed by the larger of the lateral dimension of the sampiing volume or the pixel size (7). Under any circumstances, x-ray spectra will be generated at a greater depth than backscattered electrons. Another point to keep in mind is that quantitative energy dispersive spectrometry is based on the assumption that the material being analyzed is homogeneous - an assumption with questionable grounding for hydrated cement paste. Therefore, chemical image analysis will produce an image in which individual phases can be identified only approximately and from which the relative strength of the signal can be used to identify approximate phase boundaries. Backscattered electron imaging can be used in conjunction with chemical image analysis to sharpen the phase boundaries. As stated earlier (3), the authors believe that chemical image analysis is a powerful technique. However, the technique has its own drawbacks, since it takes much more time and provides poorer spatial resolution than backscattered electron imaging. For work carded out by the
188
H. Zhao and D. Darwin
Vol. 24, No. 1
authors, the time required for imaging is important because of the large number of images needed to obtain a statistically valid sample. Due to long acquisition time requirements, such a statistically valid sample is virtually impossible to obtain with chemical image analysis, which is best used for studying specific details in cement paste morphology. The authors agree that, by using a segmentation process, phases can be identified reasonably well; however, the uncertainty of the position of the phase boundaries, based solely on chemical image analysis, is, as a general rule, greater than that obtained from backscattered electron imaging. For that reason, the two processes are usually used together (8). The authors appreciate Bonen's persistence in trying to establish the terminology used to describe the phases observed in polished specimens of hydrated cement paste. Diamond and Bonen (9) have developed new terminology that was not available to the authors at the time of the initial reply (3). At the highest level, Diamond and Bonen categorize the morphology of hydrated cement paste as consisting of phenograins (meaning distinct grains) and groundmass (everything else). Bonen states that "CH has been found to form intimate mixtures with the amorphic CSH particles confined to the groundmass, and not with the outer hydration CSH shell of phenograin or fully hydrated phenograins... " The "outer hydration shell of the phenograin" was referred to in the paper (2) and in the authors' reply (3) as "inner product." The authors statement that "a significant portion of the inner product has the same signal intensity as massive regions of calcium hydroxide" is based on calibration settings selected to optimize the identification of the hydrated phases. The statement means to convey a description of the signal within the region normally referred to as inner product. With the settings used, the grey level range of some regions within the "inner product"/"outer hydration shell of the phenograin" overlap those of clearly identifiable regions of CH surrounded by "outer product"/"groundmass." Calcium hydroxide has been observed within the boundaries of the inner product region by Rayment et al. (10, 11). The CH in this region is likely in a microcrystaUine form, as observed by Groves (12), at a scale below 100 nm. At this scale, the differences in the sampling volumes of backscattered electrons and x-rays will make differences in mean atomic number easier to identify than differences in chemistry. References
1. Zhao, H. and Darwin, D., SM Report No. 24, Univ. of Kansas Center for Research, Lawrence, Kansas (1990). 2. Zhao, H. and Darwin, D., Cem. Concr. Res. 22, 695 (1992). 3. Zhao, H. and Darwin, D., Cem. Concr. Res. 23,754 (1993). 4. Bonen, D., Cem. Concr. Res. 23, 749 (1993). 5. Newbury, D. E., Yakowitz, H. and Myklebust, R. L., Appl. Phys. Lett., 23, 488 (1973). 6. Arnal, F., Verdier, P. and Vincinsini, P.-D., C. R. Acad. Sci. Paris, 268, 1526 (1969). 7. Goldstein, J. I., Newberry, D. E., Echlin, P., Joy, D. C., Romig, A. D., Lyman, C. E., Fiori, C. and Lifshin, E., Scanning Electron Microscopy and X-Ray Microanalysis, 2nd Ed., Plenum Press, New York and London. 8. Bonen, D. and Diamond, S., Mat. Res. Soc. Symp. Proc. 245,291 (1992). 9. Diamond, S. and Bonen, S., J. Am. Ceramic Soc. 76 (1993) (in press). 10. Rayment, D. L. and Majumdar, A. J., Cem. Concr. Res. 12, 753 (1982). 11. Rayment, D. L. and Lachowski, E. E., Cem. Concr. Res. 14, 43 (1984). 12. Groves, G. W., Cem. Concr. Res. 11,713 (1981).