ARTICLE pubs.acs.org/est
Determination of the Depth of Localized Radioactive Contamination by 137Cs and 60Co in Sand with Principal Component Analysis Jamie C. Adams,*,† Matthew Mellor,‡ and Malcolm J. Joyce† † ‡
Engineering Department, Lancaster University, Lancaster LA1 4YR, U.K. Createc Ltd., Unit 8 Derwent Mills Commercial Park, Cockermouth, Cumbria CA13 0HT, U.K. ABSTRACT: A method to determine the depth of buried localized radioactive contamination nonintrusively and nondestructively using principal component analysis is described. The γ-ray spectra from two radionuclides, cesium-137 and cobalt-60, have been analyzed to derive the two principal components that change most significantly as a result of varying the depth of the sources in a bespoke sand-filled phantom. The relationship between depth (d) and the angle (θ) between the first two principal component coefficients has been derived for both cases, viz. dðj Þ ¼ x þ y loge j where x and y are constants dependent on the shielding material and the γ-ray energy spectrum of the radioactivity in question, and j is a function of θ. The technique enables the depth of a localized radioactive source to be determined nonintrusively in the range 5 to 50 mm with an accuracy of (1 mm.
’ INTRODUCTION Profiling the depth of radiological contamination remotely would advance the remediation and decommissioning of many nuclear legacy facilities. Depth profiling is challenging as the land or building material that entrains the source of contamination also acts to attenuate the radiation that is emitted from it. The relative attenuation method is one method by which the depth of entrained radioactive contamination can be determined.1,2 This method exploits the relative attenuation of the X-ray and γ-ray photopeaks of the corresponding photon energy spectrum, and can be used to characterize the depth of radiological material to a maximum of ∼20 mm when shielded with sand or concrete. The technique is also constrained by the radioactive substance in question because of its reliance on the presence of a significant X-ray emission in addition to the γ-ray spectrum. Cobalt-60, for example, does not yield a sufficiently prominent X-ray, and thus, this technique is not effective for this radionuclide and others constrained in this way.3 This paper describes a technique to derive the depth of localized radioactive contamination in sand based on principal component analysis (PCA). Coefficients derived from γ-radiation spectra have been extracted for two separate radionuclides entrained at 1 mm depth increments between 5 and 50 mm in sand. These exhibit a trend enabling depth to be determined up to 50 mm that is independent of specific features of the γ-ray spectra. The upper limit of 50 mm was achieved using this experimental arrangement and does not preclude the technique from deeper assay. ’ PRINCIPAL COMPONENT ANALYSIS OF γ-RAY ENERGY SPECTRA PCA can be used to identify dependencies in multivariable data sets.4 It has numerous applications in such fields as r 2011 American Chemical Society
environmental science, climatology, facial recognition systems, and the social sciences, to name a few. In the case of a γ-ray spectrum, where one bin (or channel) usually contains a large number of counts, it is likely that the bins immediately adjacent will contain significant numbers of counts as well. When this is observed, it may be possible to reduce the dimension of a data set without losing significant amounts of information by capturing these correlations. PCA is perhaps the simplest method for data reduction; it is a quantitative, linear method for simplifying data by representing it as a sum of a smaller number of principal components (PCs). These PCs are uncorrelated versions of the original correlated variables. PCA finds the optimal set of PCs such that each PC is orthogonal in p-space to all others in the set, and such that a data reduction to dimension N is optimized in the sense that the sum-of-squares error between the dimensionally reduced data set and original data is minimized. It is usual for the first few PCs to account for the majority of the original data set; in this case the lost dependencies usually have a minimal influence on the information provided by the PCA. In this paper we make reference to one specific output of the PCA: PC coefficients. These are linear combinations of the spectral data which display correlations and which are used to generate the PCs. More information on the derivation of PC coefficients can be found in ref 5.
’ EXPERIMENTAL METHOD A phantom filled to a depth of 0.26 m with fine-grade silica sand was used for all the experiments in this research, shown in Figure 1.6 Sand has been used to shield both a cesium-137 Received: May 12, 2011 Accepted: August 9, 2011 Revised: August 8, 2011 Published: August 09, 2011 8262
dx.doi.org/10.1021/es201619r | Environ. Sci. Technol. 2011, 45, 8262–8267
Environmental Science & Technology
ARTICLE
in this research: 137Cs and 60Co. The sources used had activities of 382 kBq and 21.57 kBq, respectively. Real-time measurement periods of 300 ( 3 s and 5597 ( 1 s were used for 137Cs and 60 Co, respectively, due to the difference in activities of both sources. These counting times allow for comparable counting statistics and a well-defined spectrum. This allowed a 5 to 50 mm PCA depth model to be constructed for both radionuclides independent of one another.
Figure 1. The sand phantom used in this research.
(137Cs) and a cobalt-60 (60Co) source due to its uniform grain size and ease of use. Further, it is a major constituent of most shielding materials at nuclear legacy sites where radionuclides have been released into the natural environment as a result of the nuclear fuel cycle. The phantom is made of acrylic glass (polymethylmethacrylate (PMM)) as this is easily machined and has a minimal influence on the radiation under study. A rectangular chamber traverses the phantom along its horizontal plane in which a plastic slider is inserted. This is designed to accommodate a sealed radioactive source (radius = 25 mm, thickness = 2 mm). This enables the source to be placed at specific depths ranging from 5 to 50 mm in the sand by adjusting the lateral position of the slider within the chamber. Depth increments of 1 ( 0.5 mm were used, calibrated using trigonometry with respect to the position of the slider at the front face of the phantom.
’ RADIATION DETECTION A single γ-ray detector has been used in this research: an ICS4000e radionuclide identifier used in conjunction with an XP submersible gamma ray probe (XRF Corporation, MA).7 The integrated unit comprises a 10 10 1 mm3 cadmium telluride (CdTe) detector. As the depth of the source was varied by moving it along the horizontal plane in the slider in the phantom, the source was also tracked by moving the CdTe detector parallel with the source along the external, front face of the phantom (pointing toward the source). The γ-ray detector was kept at a constant height from the floor, at the same height as the source, as it was moved laterally through the chamber of the phantom. This enabled a γray photon energy spectrum to be measured at each 1 mm-depth increment with a depth precision of (0.5 mm, measured with the integrated rule in the phantom slider. Tracking the detector with the source as it is moved through the phantom ensures that the point of maximum intensity was being measured at each source depth. No measurements were taken at the far extremes of the phantom where scattering is not equilibrated and might result in perturbative edge effects. ’ RADIONUCLIDES USED IN THIS STUDY Two commonly encountered anthropogenic sources of radioactive contamination in the nuclear industry have been investigated
’ SPECTRAL ANALYSIS AND EMPIRICAL DEVELOPMENT OF THE METHOD As the depth of shielding increases between a γ-ray emitting source and a detector, the full-energy γ-ray photopeak becomes more attenuated. This results in a lower total count rate for the complete energy spectrum when measured for a constant time period relative to the unshielded case. Therefore, in order to compare different depth spectra with one another, each spectrum was normalized to its total count. This was achieved by dividing the count rate for each 1 keV-wide energy bin by the total count rate across the entire energy range of the spectrum in question. PCA was carried out to derive PC coefficients for each spectrum, and then the angle (θ) between the two most significant PC coefficients was determined. The PC-derived angles were plotted against the depth (d) at which the radionuclide was placed for each spectrum. The relationship between d and θ was then derived empirically by extracting the best-fitting continuous function that was consistent with the limiting depth behavior of the data corresponding to the rate of change of depth with θ, and d f ∞ and d f 0 corresponding to full attenuation and zero attenuation of the γ-ray photopeak, respectively. Full attenuation is defined here as the complete loss of the γ-ray line to the shielding material (sand), resulting in no discernible photopeak above background levels. The normalized spectra for 137Cs and 60Co measured in this research are shown in Figures 2a,b, respectively, for 5 mm and 50 mm depths. This enables direct comparison to be made between spectra taken at different depths. Each model contained 46 separate spectra, corresponding to each 1 mm depth increment between 5 and 50 mm. Both plots in Figure 2 show a lowenergy scatter peak at ∼40180 keV that increases with depth, and a full-energy γ-ray photopeak that decreases with depth. The scatter region comprises photons which have scattered within the environment and imparted some of their energy to shielding and surrounding materials, before interacting with the radiation detector. The characteristic γ-ray photopeak of 137Cs is clearly observed at 662 keV, while the two γ-ray photopeaks of 60Co at 1173.2 and 1333.5 keV are less well-defined in comparison. For the 137Cs source a low-energy X-ray is also observed at 32 keV. Figures 3a,b was generated by plotting the first two PC coefficients for each depth variable for 137Cs and 60Co, respectively, as biplots. All 46 depth spectra (5 to 50 mm) for each radionuclide display the same depth trend. The first two PC coefficients relating to the 137Cs source, shown in Figure 3a, indicate a trend in which the value of PC coefficient 2 decreases with increased depth. The percentage of the data explained by the first two PCs for the 137Cs source has a combined value of 96.8%. Due to this high percentage, only the first two PCs have been used for subsequent analysis, constraining the technique to 2 dimensions in this research. A similar trend is observed for the 60Co source, shown in Figure 3b. The percentage of the 60Co data explained by the first 8263
dx.doi.org/10.1021/es201619r |Environ. Sci. Technol. 2011, 45, 8262–8267
Environmental Science & Technology
ARTICLE
Figure 2. Complete γ-ray photon energy spectra for (a) 137Cs and (b) 60Co. The spectra have been normalized against the total count in each case. For improved clarity only the spectra for 5 and 50 mm depths have been plotted.
two components has a combined value of 84.6%. This is lower compared to the 137Cs model, with the remaining percentage lost as noise and features that are not correlated by the first two PCs. This lower percentage variance may also be due, in part, to the lack of a prominent X-ray in the case of 60Co. Despite this, a trend is exhibited similar to the 137Cs case. To quantify the relationship between the PC coefficients and depth, the angle (θ) between the first two PC coefficients, shown in Figure 3, has been defined in eq 1 1 PCC2 ð1Þ θ ¼ tan PCC1 where PCC1 is the first principal component coefficient and PCC2 is the second principal component coefficient. θ has units in radians. Figure 4 displays the dependence of the PCA angle (θ) when plotted against its corresponding depth. A function of depth d was then established to provide a best fit to PCA angle (θ) for both radionuclides. The best-fit function is derived from the form of a FermiDirac distribution and is given in eq 2 θðdÞ ¼
"
w #
dx
e
y
!z
ð2Þ
þ 1
where x relates to the depth (d) when θ = 0 radians and y and z are constant offsets. The values for all three constants are
x = 24.2 mm, y = 6.88, and z = 1.01, where w is radionuclide specific: w = 2.27 radians for 137Cs and 2.21 radians for 60Co. Rearranging eq 2 for both sets of PCA angles (θ) enables d(j) to be determined, where j = j(θ), and is given in eq 3 dðjÞ ¼ x þ y loge j where j is given in eq 4 vθ j¼ θ þ z
ð3Þ
ð4Þ
and v = 1.26 for 137Cs and 1.20 for 60Co. The derivation of v is shown in eq 5 v ¼ wz
ð5Þ
The parameter v appears dependent on the photon energy spectrum of the radionuclide in question; however, both radionuclides studied in this work yield similar values for parameters v and w. This may indicate the technique has potential for sources of mixed radionuclides not just singular isotopes. This may imply a universal fit parameter for different γ-ray emitting radionuclides, with the parameter only changing depending on the shielding material in question. As the only shielding material used in this study was sand, further research is required to determine the influence of the shielding material. Reduced chisquares of 161 for 137Cs and 315 for 60Co were observed for the fits in this work indicating reasonable experimental agreement between the calculated fit and the derived PCA angle for all depths. 8264
dx.doi.org/10.1021/es201619r |Environ. Sci. Technol. 2011, 45, 8262–8267
Environmental Science & Technology
ARTICLE
Figure 3. Biplots of the first two principal component coefficients for each depth, for (a) 137Cs and (b) 60Co. Not all depths have been plotted for improved visual clarity.
In a preliminary effort to validate the technique for true unknown environments, a series of blind tests have been conducted using just the 137Cs source. A colleague was asked to choose three different sand depths at random, and take a detector measurement at the point of maximum intensity for each, using the same method as before employing the use of the phantom. The three blind spectra were then added into the complete 137Cs model and had PCA conducted on this new 49 spectral data set. The depth of shielding for the 3 blind depth spectra were then established using the three new derived PCA angles and eq 3. These blind tests were conducted using prior knowledge of the point of maximum intensity. Table 1 shows both the true and inferred depths for these blind tests. To remove any bias in the results the true depths were only confirmed after the depths had been inferred using the technique. It is envisaged that in the future an iterative robotic sampling method can be employed to take detector measurements in unknown environments. This would provide a more efficient sampling method for locating the point of maximum intensity in unknown environments. A number of existing techniques could be utilized for this autonomous optimization of the measurement point, the most useful being N-Visage.8
’ DISCUSSION A technique has been described by which γ-ray spectra obtained with a relatively simple experimental setup are processed
to determine the depth at which localized radioactivity is entrained in a substance. The technique has been tested with two pure γ-ray emitting radioactive sources, 137Cs and 60Co, entrained in fine-grade silica sand. A mathematical dependence linking the depth of the radioactivity and the angle between the first two PC coefficients arising from PCA was identified from these spectra. This research indicates that depth can be measured to at least a limit of 50 mm and for more than one radionuclide. The novelty of the technique described in this paper is its ability to infer the depth of entrained radioactivity, nonintrusively and nondestructively. This has been achieved by constructing a PCA model for a dry silica-sand shielding environment using both a 137Cs and 60Co source, independent of one another. These models then allow blind spectra to be assigned an accurate depth parameter by combining them with the appropriate radionuclide model and performing PCA on this new data set; from here the angle (θ) can be extracted for the blind depth spectra. This has been demonstrated by conducting blind tests and testing the robustness of the empirical method, in particular eq 3. Such a technique provides an assay of depth in terms of range and isotopic composition, and removes the need for direct interference with the contamination. The latter would often result in the generation of waste and potentially exacerbate the migration of the contamination. This can occur for example as a result of core logging in which bore-hole samples are extracted, which is common industry practice. Core logging is intrusive, 8265
dx.doi.org/10.1021/es201619r |Environ. Sci. Technol. 2011, 45, 8262–8267
Environmental Science & Technology
ARTICLE
Figure 4. PCA angle θ (radians) plotted against the shielding depth (mm) for (a) 137Cs and (b) 60Co. A function (eq 2) was derived on the basis of least-squares weighted fit to the angle data calculated using eq 1 and shown by the dotted lines.
Table 1. Comparison of the True and Unbiased Experimentally Derived Depths Using Equation 3 for Three Blind 137Cs Depth Spectraa
Nuclear Power Development Establishment began in 1955 and is located in the far north of Scotland, U.K. The site was operated by the United Kingdom Atomic Energy Authority and was the site of three reactors: two fast breeder reactors and a thermal research reactor. These reactors have ceased operations with the final reactor being taken offline in 1994. The decommissioning of the site is still ongoing with the site now owned by the Nuclear Decommissioning Authority. The Dounreay “particles” are grain-sized fragments of irradiated fuel that have been located in the natural environment.14 Radiation from these particles is dominated by that from 137Cs. The discovery of each individual particle described above is often separated by a period of several years. Also, the terrestrial environment at Dounreay is well understood and largely homogeneous which corresponds well with the dry silica-sand used in this study and can be easily replicated in the same way. For these reasons, the technique is of particular relevance for use on-site. The burial depth of these particles is currently estimated from variation in the characteristics of the measured spectrum, and the PCA technique offers the potential to significantly enhance this. There are some potential limitations to the described PCA technique that have not yet been explored. For example it is yet to be tested for distributed sources of radioactivity and multiple point sources, which may prove challenging. The following shielding environments also remain untested: environments containing voids and large pore spaces, and also mixtures containing high z materials. These might for example include land-fills and building sites. The technique is also limited by the depth with which a radiological source can be detected above background levels. For example, a low activity source, buried at significant depth within a material, may not yield sufficient radiation.
’ AUTHOR INFORMATION
true depth (mm)
inferred depth (mm)
Blind spectrum 1 (BS1)
45 ( 0.5
42 ( 1
Corresponding Author
Blind spectrum 2 (BS2)
49 ( 0.5
45 ( 1
Blind spectrum 3 (BS3)
14 ( 0.5
14 ( 1
*Phone: +44 (0)1524 593326; fax: +44 (0)1524 381707; e-mail:
[email protected].
a
Good experimental agreement is observed for BS3. For the two deeper depth spectra (BS1 and BS2), reduced experimental agreement is observed. This is due to the reduced accuracy of the 137Cs fit at deeper depths, shown in Figure 4a.
destructive to the shielding environment, and fraught with intrinsic difficulties such as potential hole collapse during excavation. Also, and perhaps most importantly, the uncertainty on this current intrusive technique can be (100 mm or greater. The technique in its current form, tested using single radionuclides, provides a response to the need to better characterize particulate anthropogenic radiation known to exist in the terrestrial and marine environments.9,10 For the case of nuclear site decommissioning in the U.K. the objective is to reach a point where “no danger” can be demonstrated.11 The “no danger” criterion is defined both in terms of dose and an additional requirement that it is “as-low-as-reasonablypracticable” (ALARP).12,13 Knowledge of the depth of a radiation source within the ground could inform the decision as to whether it is ALARP to recover the contamination, or for it to remain, subject to demonstrating that the dose criterion is met. An example of point-source contamination for which this PCA technique could be applicable is the Dounreay “particles”. These particles have been located in the vicinity of the nuclear legacy facilities at Dounreay (Caithness, Scotland). The Dounreay
’ ACKNOWLEDGMENT The authors thank Kelum Gamage, Mark Salisbury, and Dr Alan Shippen of Lancaster University; Dr Michael Aspinall of Hybrid Instruments Ltd, (Lancaster); and Dr John Heathcote and Bill Thomson of Dounreay Site Restoration Ltd. for useful discussions and experimental assistance. We also thank Bethany Colling of Lancaster University for assistance with the blind tests. Thanks also to Sue Thompson at Dounreay Communications. This research is sponsored by REACT Engineering Ltd., the Engineering and Physical Sciences Research Council (EPSRC), and the NorthWest Regional Development Agency (NWDA). ’ REFERENCES (1) Shippen, A.; Joyce, M. J. Profiling the depth of caesium-137 contamination in concrete via a relative linear attenuation model. Appl. Radiat. Isot. 2010, 68 (45), 631-634; DOI 10.1016/j.apradiso.2009.09.046. (2) Alpatov, V. G.; Bizina, G. E.; Davydov, A. V.; Kartashov, G. R.; Sadovskij, A. A. Nonlinear finding depth of radioactive atoms diffusion —using difference in absorption of gamma quanta and X-ray photons with depth of diffusion of atoms. Soviet Union Patent SU1589227, 1989. (3) Shippen, B. A.; Joyce, M. J. Extension of the linear depth attenuation method for the radioactivity depth analysis tool (RADPAT). IEEE Trans. Nucl. Sci. 2011, 58 (3), 1145-1150; DOI 10.1109/ TNS.2011.2115253 8266
dx.doi.org/10.1021/es201619r |Environ. Sci. Technol. 2011, 45, 8262–8267
Environmental Science & Technology
ARTICLE
(4) Chatfield, C.; Collins, A. J. Introduction to Multivariate Analysis; Chapman and Hall Ltd: London, 1980. (5) Jolliffe, I. T. Principal Component Analysis; Springer: New York., 2002. (6) Shippen, B. A.; Joyce, M. J. The design and calibration of a phantom for depth profiling measurements of entrained radioactivity in silica-based media. Nucl. Eng. Des. 241 (2), 2011, 526532; DOI 10.1016/j.nucengdes.2010.12.001. (7) ICS 4000; http://www.laurussystems.com/products/products_ pdf/XRF_Catalog.pdf. (8) Adams, J. C.; Mellor, M.; Joyce, M. J. Depth determination of buried caesium-137 and cobalt-60 sources using scatter peak data. IEEE Trans. Nucl. Sci. 2010, 57 (5), 27522757; DOI 10.1109/ TNS.2009.2038480. (9) Santschi, P. H.; Li, Y.; Adler, D. M.; Amdurer, M.; Bell, J.; Nyffeler, U. P. The relative mobility of natural (Th, Pb and Po) and fallout (Pu, Am, Cs) radionuclides in the coastal marine environment: results from model ecosystems (MERL) and Narragansett Bay. Geochim. Cosmochim. Acta 1983, 47, 201–210. lvarez-Iglesias, P.; Quintana, B.; Rubio, B.; Perez-Arlucea, M. (10) A Sedimentation rates and trace metal input history in intertidal sediments from San Simon Bay (Ría de Vigo, NW Spain) derived from 210Pb and 137Cs chronology. J. Environ. Radioact. 2007, 98, 229-250; DOI 10.1016/j.jenvrad.2007.05.001 (11) Nuclear Installations Act, OPSI, c.57, 1965; http://www. legislation.gov.uk/ukpga/1965/57. (12) The tolerability of risk from nuclear power stations, OPSI, 1992; http://www.hse.gov.uk/nuclear/tolerability.pdf (13) HSE criterion for delicensing nuclear sites. HSE, 2005; http:// www.hse.gov.uk/nuclear/delicensing.pdf. (14) Wilkins, B. T.; Harrison, J. D.; Smith, K. R.; Phipps, A. W.; Bedwell, P.; Etherington, G.; Youngman, M.; Fell, T. P.; Charles, M. W.; Darley, P. J.; Aydarous, A. Sh. Health Implications of Fragments of Irradiated Fuel at the Beach at Sandside Bay, Module 6: Overall Results; RPD-EA-032006; Scottish Environment Protection Agency: Stirling, U.K., 2006; http://www.sepa.org.uk/radioactive_substances/ decommissioning/dounreay/dounreay_particles_research.aspx.
8267
dx.doi.org/10.1021/es201619r |Environ. Sci. Technol. 2011, 45, 8262–8267