Automating Sieve Size Analysis
February 1, 2008
The use of automated image analysis can provide actual sieve size results on a large aliquot of powder without any prior correlation with actual sieve data.
Sieve size analysis is a well-known and important tool for the characterization of powders over a wide range of materials and applications. Everything from ground coffee to abrasive material is priced and sold based on measured sieve fractions, and many reference methods require the use of sieves.
Though sieve size analysis has been around for thousands of years, it is still in wide use for a number of reasons. First, it is a simple and inexpensive tool. No complicated sample preparations are needed beyond simply pouring the powder into the top sieve. Secondly, sieves can test large amounts of powder-literally hundreds of grams. In contrast, many of the more modern and high-tech particle sizing methods, like laser light scattering, might ascertain particle size from just micrograms or milligrams of material. The statistical accuracy of sieve data will therefore be much better. Finally, sieves directly provide mass-weighted particle size distributions, something that no other technique can do.
Unfortunately, the use of sieves is very labor-intensive and involves long measurement times (after the powder is added, vibratory energy must be applied and each sieve has to be reweighed, cleaned and dried). In addition, sieving can be very problematic for fragile powders because the vibratory energy required to separate the particles can also fracture them and cause the size distribution to change. Another disadvantage with sieve size analysis is that the sieving time is one of several factors that can significantly affect the final results and must be controlled. As a result, the accuracy of sieving results can be dependent on the skills of the operator. Furthermore, the long measurement times preclude the use of sieve size analysis as a process monitoring tool.
With the development of digital image capture technology and inexpensive and fast desktop computing, a new instrumental solution now exists that enables statistically accurate and fast sieve size measurements. Process engineers can monitor powder particle size on- or at-line and obtain much-improved control over the specifications of their final powder product.
It is important to know that automated image analysis tools have two common configurations.1 The first configuration, dynamic image analysis (DIA), uses a conveyor belt or vibratory feeder to drop powder in front of the camera. Because the particles are under the control of gravity, only a moderate amount of material can be imaged and the orientation of the particles cannot be controlled as they fall. Thus, it is possible for flake-like particles that might normally be captured in a smaller (larger-sized) mesh sieve to be imaged from a dimension that would suggest a smaller size. The second configuration is that of static image analysis (SIA), where particles are introduced to the camera after dispersal on an optical surface. While this method provides exact orientation control, only a minute amount of material is imaged.
The new solution combines these two configurations to capitalize on their individual advantages. The powder is introduced to the camera by a laminar horizontal air flow that passes between two closely spaced glass windows (see Figure 1). The air flow is produced by a simple vacuum pump. Due to the laminar flow conditions and significant acceleration, particles are oriented with their largest surface facing the camera at the instant of image capture. The use of a telecentric lens produces a large depth of field that aids in keeping all of the particles in focus, thereby maximizing detail. The particle containment design ensures that the electronics, optics and other components are not contaminated, and allows the operator to run hundreds of analyses without any cleaning. In addition, the particles achieve such a speed that up to 500 g of material can be imaged per minute.
The second issue was to define a size parameter that would produce results that match with sieve size analysis. It may come as a surprise to those unfamiliar with particle size analysis that particle size or particle diameter is a not a well-defined parameter (except for spherical particles); in reality, many definitions exist.2 Most particle size analyzers do not measure particle size directly. Instead, they measure a physical property of the particle that is related to size, like sedimentation in a liquid, the scattering of light, and so on. Particle size is then defined relative to the measured physical property, so a particle might sediment like a 60-micron sphere, scatter light like an 80-micron sphere, or fit through a 300-mesh sieve.
Unlike most other particle sizing technologies, image
analysis provides a direct measure of particle size. It is similar in this
regard to sieve size analysis, which also selects directly based on size. All
that is required to generate sieve-type data through image analysis is to
correctly define the size of the particle based on its image. As can be seen in
Figure 2, once one has a picture of the particle, the size of that particle can
be defined in a variety of ways. For example, size can be defined as the
greatest distance between two points on the particle perimeter, the so-called
maximum size or maximum distance. One can also define particle size as the
diameter of a sphere that has the same area as the imaged particle, the equivalent
disc diameter, and so on.
If we think about how sieves actually select for size, we can conceive of a particle diameter that is equal to the size of the largest sphere that fits within the measured perimeter of the particle (see C in Figure 2). We can call this the sieve diameter. It is easy to see that this size, directly measured from the image data, can provide a well-defined particle size that will produce data that should match sieve size analysis.
The final piece of the puzzle was to properly estimate the particle volume in order to infer a mass distribution. In sieve size analysis, this is done directly by weighing the amount of material in each sieve fraction. It presents a special problem for image analysis, however, as only two dimensions of each particle are measured. From stereological theory, we know that the volume of an object is proportional to its projected area. Though it cannot help us measure the absolute particle volume, we can use this information to determine the relative volume of the particles, which in turn allows us to produce a mass distribution (assuming uniform density). The volume estimator is based on the measured area of the particle and a specially defined mean diameter.
To determine whether the method would provide accurate and precise sieve data, a 100-g aliquot of BCR 68 (a quartz material distributed by the Community Bureau of Reference for use as a sieve size analysis standard) was tested. The mass distribution was certified by sieve size analysis at five separate laboratories and can be seen in Table 1.
Figure 3 contains the
results from each of the five sieve analyses, along with the IA results. The
agreement is excellent and within the certified standard deviation. Again, it
should be emphasized that the IA data was derived directly from the images of
the particles; the measured sizes were not re-calibrated. Furthermore, the
measurement was made over several minutes, less time than it would take just to
clean a comparable set of sieves.
The
analysis is repeatable as well. Two 150-g aliquots of a refined alumina powder
were tested next, and the results are graphed in Figure 4. The data demonstrate
that, if done properly, image analysis can provide both accurate and precise
sieve results-and much more quickly than actual sieve size analysis. With image
analysis, engineers can access real-time sieve results for use in process
monitoring.
For more information, contact Particle Sizing Systems, 8203 Krystel Circle, Port Richey, FL 34668; (727) 846-0866; fax (727) 846-0865; e-mail pohagan@pssnicomp.com; or visit www.pssnicomp.com.
Functioning automated image analyzer for measuring sieve sizes.
Sieve size analysis is a well-known and important tool for the characterization of powders over a wide range of materials and applications. Everything from ground coffee to abrasive material is priced and sold based on measured sieve fractions, and many reference methods require the use of sieves.
Though sieve size analysis has been around for thousands of years, it is still in wide use for a number of reasons. First, it is a simple and inexpensive tool. No complicated sample preparations are needed beyond simply pouring the powder into the top sieve. Secondly, sieves can test large amounts of powder-literally hundreds of grams. In contrast, many of the more modern and high-tech particle sizing methods, like laser light scattering, might ascertain particle size from just micrograms or milligrams of material. The statistical accuracy of sieve data will therefore be much better. Finally, sieves directly provide mass-weighted particle size distributions, something that no other technique can do.
Unfortunately, the use of sieves is very labor-intensive and involves long measurement times (after the powder is added, vibratory energy must be applied and each sieve has to be reweighed, cleaned and dried). In addition, sieving can be very problematic for fragile powders because the vibratory energy required to separate the particles can also fracture them and cause the size distribution to change. Another disadvantage with sieve size analysis is that the sieving time is one of several factors that can significantly affect the final results and must be controlled. As a result, the accuracy of sieving results can be dependent on the skills of the operator. Furthermore, the long measurement times preclude the use of sieve size analysis as a process monitoring tool.
With the development of digital image capture technology and inexpensive and fast desktop computing, a new instrumental solution now exists that enables statistically accurate and fast sieve size measurements. Process engineers can monitor powder particle size on- or at-line and obtain much-improved control over the specifications of their final powder product.
Figure 1. Air flow carries particles past the camera
in a controlled orientation.
Developing an Automated Solution
When properly implemented, the use of automated image analysis (IA) can provide actual sieve size results on a large aliquot of powder without any prior correlation with actual sieve data. Three issues had to be addressed when determining how to implement this solution. The first challenge was introducing the powder particles to the camera in a way that was fast and produced images of the particles in controlled orientations.It is important to know that automated image analysis tools have two common configurations.1 The first configuration, dynamic image analysis (DIA), uses a conveyor belt or vibratory feeder to drop powder in front of the camera. Because the particles are under the control of gravity, only a moderate amount of material can be imaged and the orientation of the particles cannot be controlled as they fall. Thus, it is possible for flake-like particles that might normally be captured in a smaller (larger-sized) mesh sieve to be imaged from a dimension that would suggest a smaller size. The second configuration is that of static image analysis (SIA), where particles are introduced to the camera after dispersal on an optical surface. While this method provides exact orientation control, only a minute amount of material is imaged.
The new solution combines these two configurations to capitalize on their individual advantages. The powder is introduced to the camera by a laminar horizontal air flow that passes between two closely spaced glass windows (see Figure 1). The air flow is produced by a simple vacuum pump. Due to the laminar flow conditions and significant acceleration, particles are oriented with their largest surface facing the camera at the instant of image capture. The use of a telecentric lens produces a large depth of field that aids in keeping all of the particles in focus, thereby maximizing detail. The particle containment design ensures that the electronics, optics and other components are not contaminated, and allows the operator to run hundreds of analyses without any cleaning. In addition, the particles achieve such a speed that up to 500 g of material can be imaged per minute.
The second issue was to define a size parameter that would produce results that match with sieve size analysis. It may come as a surprise to those unfamiliar with particle size analysis that particle size or particle diameter is a not a well-defined parameter (except for spherical particles); in reality, many definitions exist.2 Most particle size analyzers do not measure particle size directly. Instead, they measure a physical property of the particle that is related to size, like sedimentation in a liquid, the scattering of light, and so on. Particle size is then defined relative to the measured physical property, so a particle might sediment like a 60-micron sphere, scatter light like an 80-micron sphere, or fit through a 300-mesh sieve.
Figure
2. Particle size can be defined in different ways, including A, maximum size;
B, equivalent disc diameter; and C, sieve diameter.
If we think about how sieves actually select for size, we can conceive of a particle diameter that is equal to the size of the largest sphere that fits within the measured perimeter of the particle (see C in Figure 2). We can call this the sieve diameter. It is easy to see that this size, directly measured from the image data, can provide a well-defined particle size that will produce data that should match sieve size analysis.
The final piece of the puzzle was to properly estimate the particle volume in order to infer a mass distribution. In sieve size analysis, this is done directly by weighing the amount of material in each sieve fraction. It presents a special problem for image analysis, however, as only two dimensions of each particle are measured. From stereological theory, we know that the volume of an object is proportional to its projected area. Though it cannot help us measure the absolute particle volume, we can use this information to determine the relative volume of the particles, which in turn allows us to produce a mass distribution (assuming uniform density). The volume estimator is based on the measured area of the particle and a specially defined mean diameter.
Table 1. Mass distribution certified by sieve size analysis for BCR-sand 68 (100 grams).
Achieving Accurate Data
How does this combination of fast and oriented particle introduction to the digital camera, carefully selected definition of particle size, and volume fraction estimator work in terms of producing accurate sieve size data? A working version of the device is shown in the lead image. Powder is loaded into the vibratory feeder on the left. The powder drops into the air stream, which moves through the rectangular-shaped pipe. The light source (monochromatic collimated blue light for best resolution) is mounted underneath the pipe and protected by a removable glass window. The digital camera and telecentric lens, positioned above the light source, are also protected by a removable glass window.To determine whether the method would provide accurate and precise sieve data, a 100-g aliquot of BCR 68 (a quartz material distributed by the Community Bureau of Reference for use as a sieve size analysis standard) was tested. The mass distribution was certified by sieve size analysis at five separate laboratories and can be seen in Table 1.
Figure 3. Comparison of image analysis and sieve size
analysis on BCR 68.
Figure
4. Comparison of multiple runs of an alumina powder.
For more information, contact Particle Sizing Systems, 8203 Krystel Circle, Port Richey, FL 34668; (727) 846-0866; fax (727) 846-0865; e-mail pohagan@pssnicomp.com; or visit www.pssnicomp.com.