The Applied Mineralogy User CommunityAugust 8, 2017
Evaluation of SEM-EDS Particle Count StatisticsAugust 29, 2017
One of the key questions about SEM-EDS data is whether or not you’ve measured enough particle sections to produce a representative result. The critical part to remember is that SEM-EDS data are collected on particle sections exposed in a sample block, which is unlike other bulk analysis techniques such as XRF or XRD. It follows that the data from any one particle section is inherently biased and not a true representation of the character of the sample. Only a population of particle sections are able to provide accurate information. Adequate particle statistics are critical for applications such as operational mineralogy where liberation, grain size and mineral associations play a key role in mineral processing behaviour. The question is, how many particle sections are enough?
If you’re measuring large areas, such as those in thin-sections, clearly the number of particles are irrelevant. Instead the question becomes, have you measured enough of the sample surface? For a rock section the preference will almost always be to measure the whole section, except for particularly homogenous rocks such as a clean sandstone, in which case measuring just a small portion of the sample surface ought to be adequate to achieve a representative result. For solid surfaces the application may also dictate the size of the measured area.
For particles the problem is slightly more complex. If we would like to arrive at a representative result for the original sample we first have to assume that the sample preparation process has not introduced any bias and that the section of particles exposed for measurement is indeed representative of the original sample. Once we are happy to make this assumption we have to consider the number of particles to measure. SEM-EDS measurements produce a range of outputs, including bulk mineralogy, mean grain and particle size, mineral associations, mineral liberation, etc. The number of particles required to achieve a representative result may vary for each of these outputs.
One way of checking for representative outputs from a SEM-EDS measurement is to analyse the output variance as a function on the number of particles measured. As the number of measured particles increases, the output variance decreases as the mean for that output converges to some constant value. If the output variance is considered appropriately low, and we are confident in sample preparation, then we can conclude that enough particles have been measured and that the output result is representative.
Some SEM-EDS systems include built-in functionality to terminate measurements based on particle statistics, however for data generated by QEMSCAN one has to look at this information retrospectively.