Chemical Engineering Science 60 (2005) 289 292 www.elsevier.

com/locate/ces

Shorter Communication

Measurement of the size, shape and orientation of convex bodies

Received 17 March 2004; received in revised form 2 July 2004; accepted 6 July 2004 Available online 15 September 2004

Abstract Different types of solid bodies (particles) with specic shape and size are needed for industrial processes. For spherical particles, sizer measurements are usually reported as sphere diameters. For non-spherical particles, particle shape and especially orientation must be taken into account. Particles of a specic shape will present different views when looked at from different directions. We have employed a Camsizer instrument to measure the distribution of projected area under different physical conditions for solid cylindrical particles: Under ideal conditions, corresponding to a uniform distribution of particle orientation, measurements agreed with predictions. The celebrated theorem of Cauchy applies only to this case. For two other situations, measured and predicted results differed. However, from the experimental data it was possible to infer the particle orientations and hence theoretically predict the projected area distribution. Excellent agreement between theory and practice was then restored. 2004 Elsevier Ltd. All rights reserved.
Keywords: Particle shape; Cylinder; Projected area; Orientation

1. Particle shape The shapes of particles used in practice have been variously described as spherical, sharp-edged irregular, thin and aky or brous (Davies, 1975). It is this shape which dominates the way in which particles rest on a xed surface, move in a uid and interact and react physically and chemically. Independently of the particle property actually measured, the particle size is usually reported as a linear dimension. This may be the diameter of the sphere having the same property, the sphere equivalent diameter or SED (International Standard, 1998). However, SED values for anisodiametric particles have been shown to lack agreement when different physical criteria are used to measure a given sample
Corresponding author. Tel.: +44-114-222-3718; fax: +44-114-2223789. E-mail address: [email protected] (G.T. Vickers).

(Jennings and Parslow, 1988). In addition, errors have followed the use of the equivalent sphere approach to the calculation of particle volumes from spheres of equivalent area and particle areas from spheres of equivalent volume (Vickers and Brown, 2001). In an evaluation of methods for predicting the drag coefcient on particles of different shapes, the best mean error of 16% was obtained by using equal volume sphere diameter and sphericity together (see Chhabra et al., 1999). The effect of particle shape and orientation on settling rates is not well understood: asymmetrical particles may exhibit sliding, oscillating, gliding or turbulent motion (Heiskanen, 1993).

2. Calculations of particle projected area When individual three-dimensional particles are viewed from a distance, they present a two-dimensional image on a

0009-2509/$ - see front matter doi:10.1016/j.ces.2004.07.056

2004 Elsevier Ltd. All rights reserved.

290

D.J. Brown et al. / Chemical Engineering Science 60 (2005) 289 292

projection plane. For a sphere, this image is always a circle but for other convex shapes it depends upon the orientation of the particle, giving a spectrum of images. It has long been known from the work of Cauchy (1908) that the mean area of these projected images is one quarter of the total surface area provided that the body is convex and that all orientations are equally likely (henceforth referred to as a uniform distribution). More recently, it has been explicitly recognised that this mean value is but one property of the probability density function (PDF) for the projected area (Walters, 1947; Vickers, 1996; Brown and Vickers, 1998). From the PDF, it is possible to calculate the probability that any given range of areas will actually be measured under the assumption of uniformity, which might well be appropriate under turbulent conditions. Any discrepancy between measured and calculated distributions will reveal a bias in the presentation of the particles to the measuring device. The theoretical PDF has been found analytically for the major shape classes of general ellipsoid, general cylinders (so the including cubes) and some cones (Vickers and Brown, 2001).

2.2

11.5

12.5

Length (mm)

1.8

Fig. 1. Scatter plot for the dimensions of the cylinders used.

3. Measurements of projected areas The projected area distributions for a collection of glass beads were measured using a Camsizer . The beads were fed from a hopper onto a horizontal vibrating channel and fell over the end of the channel between a light source and two digital full-frame CCD cameras. A basic camera registered the shadow area distribution for large areas and a zoom camera did the same for small areas. The digital images were processed in real time, with claimed linear measurement limits from 15 m to 90 mm and high precision obtained between 30 m and 30 mm. The movement of individual beads could be followed. The 152 test objects were cylindrical glass beads, each of which had a cylindrical central hole (but this was unimportant in the process). The arithmetic mean dimensions of the cylinders (measured by a Zeiss Axiotech microscope) were length : 12.09 0.19 mm, diameter : 2.07 0.11 mm. Fig. 1 is a scatter plot of the dimensions. In a typical experiment, the beads were passed through the system 18 times to give 2727 projected area values (9 short of the possible maximum). For sampling with replacement, the standard error of the 18 mean values was 1.9%. f1 (x) =

a2
2

(0 < x < 2la), + a2

2lax
2 2

x2

(2la < x < ), (0 < x < a 2 ), ( a 2 < x < ),

2lax
2 2

x2

2 a 4 + 4a 2 l 2 , the maximum value of the where = projected area. Cauchys theorem applies here and so the mean value of the projected area is a(a + l)/2. 2. Cylinders falling end-over-end (tumbling): In this case the axis of the cylinders is always at right angles to the camera and the appropriate PDF is now f2 , where 2
2

f2 (x) =

4
2

x2 x2

( a 2 < x < 2al), (2al < x < ).

4. Formulae used in the calculation of PDF for cylinders

1. Cylinders with a uniform distribution of orientations: The PDF for a cylinder of radius a and length l is f1 , where

provided that a 2 < 2al. The celebrated result of Cauchy referred to earlier does not apply to the tumbling case. Indeed, the mean value of the projected area is 2a( a + 2l)/ . Hence the ratio of the total surface area to the mean value of the projected area lies between and 2 /2 (depending upon the shape of the cylinders) rather than being the Cauchy value of 4. 3. Cylinders which are always seen lengthways: There is now only one value for the projected area, namely 2al. Trivially, this is also the mean value and so the ratio of the total surface area to the mean projected area is (1+a/ l).

D.J. Brown et al. / Chemical Engineering Science 60 (2005) 289 292

291

0.1

0.1

0.0 0 10

20

2 30 (mm)

f
Fig. 2. Strong vibration. PDFs ( f ) for the projected area (x) of cylinders. The solid histogram shows the experimental results and the dashed histogram the theoretical PDF for an isotropic distribution of orientations. The sketches on the left of this gure (and subsequent ones) show typical orientations of the cylinders.
0.0 0 10

20

(mm) 2

30

Fig. 3. Weak vibration. PDFs ( f ) for the projected area (x) of cylinders. The solid histogram shows the experimental results and the dashed histogram is the theoretical PDF for tumbling.

To produce the theoretical distributions referred to in the next section, the formulae given above were averaged over all the different dimensions for each bead.

0.3

5. Experimental results under different conditions

1. Strong vibration of the feeder channel: The cylinders were bouncing as they left the feeder channel. Fig. 2 shows plots of the PDFs for the experimental (solid line) and the uniform distribution. The agreement between the two is good. 2. Weak vibration of the feeder channel: Fig. 3 is a plot of the PDF (f) for the projected area (x) of the cylinders. The experimental results are shown by the solid line. The theoretical PDF is for tumbling. Clearly, the results are consistent with the cylinders falling end-over-end past the cameras. Further investigation revealed that the weak vibration had caused the long axes of the cylinders to be aligned on the feeder channel. 3. Dropping single beads vertically using a funnel: Fig. 4 shows the experimental and theoretical results for this case. Again the agreement is satisfactory although a few of the cylinders showed signs of rotating.

f
0.1 0.0 0 10 20 30
(mm)

x
Fig. 4. Vertical dropping. PDFs ( f ) for the projected area (x) of cylinders. The solid histogram shows the experimental results and the dashed histogram the theoretical PDF.

6. Equivalent spheres revisited When the orientations are uniformly distributed, the mean value of the projected area a(a + l)/2 and so the radius is of the equivalent sphere is a(a + l)/2. This allows the calculation of the volume of the equivalent sphere and hence

the construction of the solid curve of Fig. 5. When the presentation of the cylinders to the camera is not uniform (i.e. the orientations are not uniformly distributed over the surface of a sphere) then the mean value of the projected area is likely to be different and hence the equivalent sphere has a different radius. In general, if r is the radius of the equivalent sphere then the mean projected area is r 2 and so RV = actual volume = volume of SED a2l
4 3

r3

3a 2 l . 4r 3

292
2.0

D.J. Brown et al. / Chemical Engineering Science 60 (2005) 289 292

RV

1.0

log10 r
-1.0 0.0 1.0

Fig. 5. The ratio (RV ) of the actual volume to the SED volume against the log of the aspect ratio, r = l/(2a). The solid curve is for uniform orientations, the dotdash line for tumbling and the dashed line is for dropping.

Table 1 The dependence of RV upon the method of presentation Condition of cylinders Uniform distribution Mean projected area a(a + l) 2 2a( a + 2l) RV 3l/a 2(1 + l/a)3/2 3 3 l/a

shape measured, using a micromanipulator or similar device as needed. Theory should then be used to predict the projected area distribution under the assumption of a uniform distribution of orientations. If the measurements are not in agreement with the predictions, then attention should be focused on the measurement environment. Are there features which suggest the manner in which the orientations may not be uniformly distributed, due to the measuring device itself or the application? The cause of the discrepancy may be found by feeding the device with test particles. The procedure just outlined will sometimes be impractical. In particular, it may not be possible to insert test particles into the (industrial) environment. In this case, it will still be necessary to determine (or at least have some quantitative information on) the orientation distribution of the particles. If this is not uniform, then some correction will have to be made to avoid the use of biased data. Another common problem is that more than one particle may be in the eld of view of the camera, but we do not follow this further. For simplicity, we have only considered circular cylinders. In fact the PDF (under the assumption of a uniform distribution of orientations) has been found for various shapes (Vickers and Brown, 2001), particularly ellipsoids (with the three axes having any values). Indeed, by approximating the shape of any convex body by a collection of plane areas, its PDF may be found. Acknowledgement We thank Professor Michael Hounslow for his advice and encouragement. References
Brown, D.J., Vickers, G.T., 1998. The use of projected area distributions in particle shape measurement. Powder Technology 98, 250257. Cauchy, A., 1908. Oevres Compltes dAugustin Cauchy, 1er Sr, Tome II. Gauthier-Villars, Paris, pp. 167177. Chhabra, R.P., Agarwal, L., Sinha, N.K., 1999. Drag on non-spherical particles: an evaluation of available methods. Powder Technology 101, 288295. Davies, R., 1975. A simple feature-space representation of particle shape. Powder Technology 12, 111124. Heiskanen, K., 1993. Particle Classication, Chapman & Hall, London. (Chapter 2), ISBN 0412493004. International Standard, ISO 9276H: 1998. (E)Presentation of results of particle size analysis, Part 1: graphical representation. Jennings, A.R., Parslow, K., 1988. Particle size measurement. Proceedings of the Royal Society A 419, 137149. Vickers, G.T., 1996. The projected areas of ellipsoids and cylinders. Powder Technology 86, 195200. Vickers, G.T., Brown, D.J., 2001. The distribution of projected area and perimeter of convex solid particles. Proceedings of the Royal Society A 457, 283306. Walters, A.G., 1947. The distribution of projected areas of fragments. Proceedings of the Cambridge Philosophical Society 43, 343347.

Tumbling

8 2( + 2l/a)3/2 2al 3 8 a 2l

Vertical dropping

Table 1 shows the values of RV and how it has been calculated for the three orientation distributions (corresponding to the situations of random, tumbling and dropping). From this information, the other two curves in Fig. 5 were drawn. When the aspect ratio, r (the ratio of length to diameter), is unity, the values of the three functions in Fig. 5 are 0.82, 0.86 and 1.04. For the particles used in our experiments, r is about 6 and the values are now 0.54, 0.70 and 0.43. The use of equivalent-sphere values alone leads to ambiguity of interpretation which may be removed when information about orientation is added. 7. Conclusions Which is the recommended way to measure the distribution of size and shape for a collection of convex particles? A rst stage is to sample the material and sieve it to provide sub-samples of different size ranges. These should be examined by eye under the microscope and their size and