Application of Neutron Activation Analysis to the Determination of Alumina Grades in Bulk Bauxite Samples MIHAI CSIRO.
Division
of Mineral
BORSARU
and PETER
L. EISLER
Physics. P.O. Box 124. Port Melbourne. (Receiord
19 March
Victoria
3207. Australia
1980)
A thermal neutron activation method has been developed to determine alumina grades in both dried and undried bulk bauxite samples (- 16 kg). The particle sizes of the dried samples (0.2%5.0% free moisture) varied from powder to natural pisolites (~2.8 cm diameter). For the undried samples (IO- 14% free moisture), the particles were in pisolitic form only. The accuracy for bauxite containing 48-62% alumina was 0.45% A1,03 (1 SD) for both categories of samples. Data analysis showed that the essential oarameters for accurate alumina determinations were the count rates of both the 1.78 MeV ‘*Al y-rays and the thermal neutrons.
Introduction
drogen contents (as contained water) and concentrations of strong neutron absorbers. Under conditions that closely approximate this assumption, neutron activation analysis achieves good accuracy in relying only on the count rates of the emitted y-rays. However, with bulk samples (many kilograms each). significant hydrogen contents moderate the neutron flux considerably and significant concentrations of strong neutron absorbers distort the flux appreciably. As a result, when neutron activation methods are applied to bulk samples. they must be modified to include the monitoring of variations in the neutron flux.
THE NEED for faster and more reliable methods of quality control in the mining industry has created much interest in methods of both statically and dynamically analysing bulk samples. Conventional chemical analysis uses comparatively small samples, so there is always a trade-off between the representiveness of the samples and the time and effort saved in preliminary sampling and preparation. Bulk analysis, on the other hand, uses large samples that more easily meet the requirements of preliminary sampling and preparation. Moreover, the relatively short time required for a single static bulk analysis permits the number of measurements made during an industrial operation (e.g. ship loading) to be increased, thereby improving the precision with which large amounts of ore can be monitored. Static bulk analysis is also a necessary step in the development of dynamic, or “onstream”. analysis methods. This paper discusses an investigation of a method for determining alumina grades in bulk bauxite samples via the thermal neutron activation of aluminium. This technique has already been used successfully to determine aluminium impurities in bulk iron ore samples.’ ’ ) Neutron activation, which depends on the hard radiation of y-rays and neutrons, is preferable to other bulk sample methods that use comparatively soft X-rays because the effective volume of the sample is substantially enlarged by the greater penetration of the hard radiation. Hitherto, the accepted methods of using neutron activation to determine aluminium concentrations have been restricted to small samples (to.25 kg).‘2m5’ With such samples, it is implicitly assumed that the neutron flux causing activation is virtually constant over a significant range of both hy-
Experimental The bauxite samples used in the present work were taken from a gibbsite-boehmite deposit in the Weipa and Andoom areas of the Cape York Peninsula in Australia. The samples were typical of the mineralization in that area, with the alumina contents varying from 48 to 62% by weight and the silica contents varying from I to 11% by weight. Two categories of samples were investigated. The first group of 35 samples was dried to a free moisture content of 0.2-5.0x by weight in order to simulate the free moisture conditions in bauxites that require no washing. The samples in this group had different particle size distributions: 3 of the samples were powdered, 22 were crushed to a particle size of - 6 mm, and the remaining 10 were left as natural pisolites with diameters of up to 2.7 cm. A second group of 50 samples was not processed in any way after they had been washed. (In normal plant practice, washing occurs before the bauxite reaches the automatic samplers.) The sample particles were retained in their original pisolitic form and were not 43
44
M. Borsuru
and P. L. Eisler
dried. Consequently, their free moisture contents ranged from IO to 14% by weight. The results of activation measurements with these samples were therefore more relevant to the ultimate practical scope of the method than were the measurements with the first group of samples. Chemical assays indicated that the samples contained various amounts of chemically bound water ranging from 19 to 28% by weight. Thus, the use of both dried and undried samples served to evaluate the effect of a systematic and appreciable variation in total moisture content, a parameter that influences the average neutron flux in samples. The analysis method used was based on obtaining a measure of the count rates of the neutrons that produce 28A1 (2.3 min half-life) in the bulk sample via the 27Al(~~,y)28Al reaction, and of the 1.78 MeV y-rays emitted by that radioisotope. The bulk bauxite samples (- 16 kg each), which simulated the mass and volume of samples on a plant in a brass box conveyor belt, were contained (40 x 33 x 8 cm deep) and were irradiated from underneath by thermalized neutrons. These neutrons were produced by a 36 ng 252Cf source located at the bottom of a hole (10 cm dia. x 9 cm deep) drilled into a polyethylene block that was surrounded by paraffin bricks. The neutron detector (4 atm 3He, 15 cm long) was attached to this assembly below the sample box. A sheet of silicone rubber (Dow-Corning RT 3110). impregnated with “B and suitably shaped, shielded the detector from those thermal neutrons emanating directly from the thermalizing source assembly. A paraffin block, 18 cm thick and located above the sample, reflected neutrons transmitted out of the sample. In this configuration, the neutron detector monitored the thermal neutron flux of the bulk sample more reliably than if it had been placed above the sample box because of the better backscattering geometry. After irradiation (6min), the samples were transferred within 15 s for counting (5 min) by a 127 x 127 mm NaI(T1) y-ray detector that was some 7 m away from the source. The separation between the source and the detector added considerable distance-shielding to the already appreciable shielding provided by concrete bricks placed around the source and detector assemblies. The background radiation from the source was further reduced by a 3 cm thick lead shield around the body of the detector that left only the upper surface of the detector exposed to y-rays. Spectrum stabilization was achieved with a Canberra Industries 1520 analogue stabilizer, using 662 keV y-rays from a r3’Cs source for the reference peak.
the technique of stepwise multiple linear regression.‘“,‘) In this technique, the variables enter regression analysis one at a time on the basis of statistical significance, which is calculated according to partial F test criteria. Evaluation of statistical significance occurs at all stages of the regression analysis so that intercorrelations between variables are accounted for in the final selection or rejection of variables. The linear regression model, which used the constant coefficients A, B. C, and D, had the form Y=A+By+Cw+Dn+c
(1)
where y represents the number of 1.78 MeV y-rays recorded during the 5 min counting period (in thousands), w is the weight (in kilograms) of the bulk bauxite and sample box together, II is the number of neutrons recorded during the same counting period (in thousands), and e is the difference between the predicted grade (y) and the chemically assayed grade (Y). The regression equation for the dried samples was y = 100.42 + 0.1191; - 1.698~ - 0.0078n
(2)
with an estimated standard deviation of s = 0.5% A120,. A plot of assayed (on a dry basis) alumina grades versus predicted grades for the dry samples is shown in Fig. 1. The importance of monitoring variations in the neutron flux is illustrated by the fact that when n was excluded from the analysis, the resulting estimated standard deviation (s’ = 1.0% A1,03) was considerably greater than the above value. It is also worth noting that the standard deviations cited in this work include errors introduced by sampling and chemical analysis. It was found independently that these errors amounted to 0.2% A1203, and if this was taken into account. the error in the regression analysis was about 0.45% A1203. Also. it appeared from the
Results and Discussion Linear
regression
analysis
The grades of alumina were predicted from the neutron and y-ray counts and from the sample weights by
% alumina
FIG. I. Comparison assays
(chemical
analysis)
of neutron activation and for AI,O, in dried samples.
chemical
45
Neutrons activation analysis of hulk samples
+
’ 4%
50
I
”
1”
52
% alumina
54 (chemical
”
I 56
50
60
analysis)
FIG. 2. Comparison of neutron activation and chemical assays for Al,O, in undried samples.
regression analysis that the values of the residuals were independent of particle size. For the undried samples, the regression equation was y = 109.56 + 0.177~ - 2.07~ - 0.0095n
(3)
with
an estimated standard deviation of s = 0.5% which is identical to that for the dried Al,O,. samples. A plot of assayed alumina grades (on a dry basis) versus predicted grades for the undried samples is shown in Fig. 2. In this case, the accuracy of the alumina determination was not greatly improved by the inclusion of II in the regression equation because the exclusion of II gave an estimated standard deviation of s’ = 0.6% Al,O,. The reason why the neutron count rate had a relatively greater significance in the analysis of the dried samples is probably connected with the fact that n has a different range of variation in the two groups of samples. In the dried samples the coefficient of variation of 11 is about 2.5%, whereas in the undried samples the coefficient of variation is abou,t 0.8% Thus, the ratio of variation between the two cases is about 3: 1. The different variations in n for each group of samples probably reflect the different levels of neutron flux saturation in relation to the different water contents. Possible sources of interference The radioisotope 28Al is also produced by reactions other than 27Al(n,~)28Al and can therefore interfere with measurements based on that reaction. For example, the 28Si(r?, P)~‘AI reaction could lead to interference. This is a fast neutron reaction with an energy threshold of 3.8 MeV. The magnitude of the interference depends inversely on the degree of thermalization in the neutron flux incident on the sample. A 20 kg sample of pure silica was irradiated to estimate how much the fast neutron reaction contributed
to the area of the 1.78 MeV y-ray peak. The resulting calculated contribution was equivalent to less than 0.1% of the I .78 MeV spectral peak for bauxites containing up to 10% silica by weight and not less than 48% alumina by weight. Another fast neutron reaction that can lead to interference is 31P(n,a)28AI. However, because of the low concentration of phosphorus in the bauxite deposit sampled (<0.07’?), and because of the low cross section of this reaction (1.5 mb at 5 MeV), the interference from phosphorus was considered to be insignificant. Yet another possible source of interference is the formation of 56Mn via the reactions 55Mn(n, y)56Mn and 56Fe(n, p)56Mn. This radioisotope (2.58 h half-life) emits y-rays with energies of 0.847, 1.811 and 2.113 MeV. The most important source of interference is the 1.811 MeV y-radiation, which cannot be resolved by NaI(TI) detectors from I .78 MeV y-radiation. However, the concentration of manganese in Cape York Peninsula bauxites is very low (- 20 ppm). It was previously found (I’ that even larger concentrations of manganese produce no significant contribution to the observed 1.78 MeV y-ray count. The amount of 56Mn formed via the 56Fe(n,p)56Mn reaction is also insignificant according to the same reference. In fact, the anticipated contribution to the y-ray counts from iron in the present work is much less in absolute terms than the 0.4% contribution cited in the above reference. The alumina concentration in the present work is about 1 order of magnitude greater and the iron concentration is much less (about 4) than it was in the earlier work. Conclusions Activation analysis using thermal neutrons is applicable to the determination of alumina grades in 16 kg bulk samples of bauxite. The accuracy of the method, obtained with samples containing 48-62x alumina, is about 0.45% Al,03. The advantage of the method is that no sample preparation or drying is required. The requirements for the analysis of bulk samples differ from those for small samples in that variations in the neutron flux caused by variations in the bulk material must be taken into account. In this way, it is possible to achieve an accuracy such as that stated above, which is acceptable to the mining industry. On the basis of Poissonian counting statistics, using the same configuration and the same source as described here, the total analysis time in the present work (1 I min for both irradiation and counting) can in fact be reduced to 5 min without loss of accuracy. In the light of considerations described in earlier work”) it is evident, again on the basis of Poissonian counting statistics, that the present method can be extended to on-stream analysis, using a side belt. In order to obtain count rates comparable to those in the present work, and hence a similar precision, the
46
M. Borsur11 uud P. L. Eislcr
speed of the side belt would need to be about 3 m/min. In comparison with other nuclear techniques that use small (-. I g) samples, bulk sample analysis in the application described here gives a competitive accuracy, and it has the added advantage of using the smaller, safer. and more easily handled neutron sources.
Ack,iOM’/Cdy(‘)tlt’)ll.s~The authors thank J. Aylmer and C. Ceravolo of the CSIRO Division of Mineral Physics for their valuable help and advice. and Comalco Limited for their helpful interaction with the authors and for providing the assayed samples.
References 1. BORSAHU M. and HOLMES R. J. A,Iu/u. Chr,n. 48. I699 (1976). 2. GIJBELS R. H. and HERTOG~N J. Pure uppl. Clw~. 1555 (1977).
49.
4. ALAER~S L.. OP IIE BEECK J. P. and HOSE J. -l~~~r/~~trc~[r chim. Actu 70, 253 (1974). 5. ALAERTS L., OP DE BEECK J. P. and HOSTE J. 4nul~rtu cl~inl. Acru 78, 329 (1975). 6. DRAPER N. R. and SMITH H. Applied Rrywwmz ,41x+ sis, pp. 171-172. Wiley, New York (1968). 7. Ibid.. pp. 178-195. 8. HOLMES R. J.. BORSARU M. and WYLIL A. W. In Pro<, North Qurrnsla~~d Conf, pp. 235-244. Australasian Institute of Mining and Metallurgy. Melbourne (197X).