The idea is fine and plausible in theory (and to an extent in practice), but because the XRay emitter/detector window is small, the rock texture can cause significant variability between the pXRF and a traditional laboratory XRF result/what would be considered the whole rock assay.

As an example, a porphyritic textured rock - different results depending on whether you are sited on a phenocrysts or the groundmass. Similarly the rock face being analyzed must be clean (and dry). You can find work around etc (increase the number of readings per rock, crush and pulverize the rock sample etc), but be aware of the issue.

A number of vendors sell grinder-based onsite XRF samplers to reduce XRF field sampling spot read errors. Some vendor representatives teach "sweep" measurement techniques (not good safety) or replicate averaging across freshly exposed faces. However, hand-held portability limits the power (penetration) and size of the XRF measurement beam. Detector resolution and deconvolution algorithms are limited in even the best hand-held XRF instruments. Even in the absence of undetected matrix bias, field XRF measurements should be treated as qualitative guidance and tempered with caution.

Unfortunately, what you are expecting is not possible. Regardless what XRF brand/ model you use, the X-rays can only analyze what they see. Therefore if your sample or mining face is not homogenous and/or coarse grained then the resulting analysis will be biased. In addition though most instruments show good precision, they do not always give accurate readings "out of the box" and so if ultimate accuracy is required then you will need to develop custom calibrations for your project.

THAT DEPENDS ... 😉

IF you want to avoid the critical XRF path length variability (which you must using a handheld XRF), somehow or other you must 'shoot' at a flat (or a reasonably flat) surface - this is non-negotiable! I am sure you understand. Producing a flat suitable surface is easy in the lab, but what about in the field or in the mine? Well there is hope and "YES, WE CAN".  

Sorry if I break a dream...

Most pXRF analyzers are able to provide laboratory-level analyses on samples which were prepared in the same way as for the lab. On the reverse, they will provide an accurate, but non-representative qualitative analysis of any unprepared surface. In between, you may expect a reasonable but still rough level of accuracy after simple preparation operations.

Expecting an accurate analysis without preparation is a hopeless dream, for simple geological, mineralogical or physical reasons - regardless of the instrument: any rock is made of mineral grains. You may expect useful information from direct shooting on rock faces, in the same way as by pointing a hand lens towards a mineral grain. But you cannot expect anything near statistical representativity.

A pXRF handheld will behave like a miniature SEM, in this case.

Thank you all for your kind guidance. It definitely helps me in understanding not only the limitations of Handheld technology but also see if there could be a possible work around.

Great paper (and Road show presentation) by Dr Michael Gazley in AusIMM's Monograph 30: Guide to good practice. Michael discusses handheld XRF and associated QAQC practices. I believe he also presents training courses in the subject. http://is.gd/Tz3Wbv And http://is.gd/ZyZYOz

Bottom line for impact on accuracy of this process is the texture and the homogeneity of the sample on the rock face.

The handheld instruments themselves will also be limited in their power, and so their ability to get sufficient counts, and also deconvolution of spectra to accurately determine % of individual elements will be limited. This can be a problem on even larger benchtop EDXRF instruments.

In my view the handheld instruments do have their place, but it is for general qualitative analysis. Good for finding the answer to general questions rather than providing results to be used for mining or process planning.

If you are working in remote areas and need quick accurate results, as a minimum you will need to perform some degree of sample preparation to homogenise the sample, and even better still provide a flat surface for analysis. The result will still be subject to the area on the rock face which the sample of ore was taken from. The remedy for this is then larger samples with splitting and larger number of analysis etc. There would be solutions which you might be able to fit in the back of a utility vehicle for accurate results (depending on the ore!), small grinder/press or fuser/EDXRF unit?

The quickest shortcut from rock face to sample preparation is using a battery-operated disc cutter or grinder, fitted with a dust collection device, and dig a small trench in the rock face, then analyse the collected dust. One such device was developed for ThermoNiton.

But beware of many sampling error issues in relation with dust size, weight and density, mechanic resistance heterogeneity, and cutting disc contamination. This is not yet routine technology.

I would like to comment on the reverse approach proposed above. This clever device aims at preparing a better shooting surface on rock faces, improving therefore the accuracy of pXRF measurement. It will not solve mineral grain heterogeneity or size issues, but I am confident it can be efficient on fine grained rocks - I would love to test it on sandstone or shale!

On coarse rocks or variable textures (banded, for instance), I would prefer the disc option. 

Good, interesting comments folks. Anyone care to comment on the following?

Would a light jaw crusher / RSD combination - capable of reducing rock (65 mm lump max ) to 95% minus 1 mm be a viable option for this type of work ? The unit would be gasoline (small industrial Honda powered), possibly mounted on a small trailer or pickup. I guess the concept is to provide a full sized, finely crushed exploration sample.

The unit would simultaneously produces one small sub sample for field XRF and one larger sub sample (250/500/1000 grams) which is retained for possible further analyses at a laboratory. The Crusher/RSD unit would be based on an existing laboratory crusher with a throughput of approximately 1 kg per minute at 95% minus 1 mm.

When I posted this question last week, my aim was to see whether we could, at an operational mine which already has an onsite laboratory reduce the current 24h turnaround time by reading off the values underground (directly from a freshly blasted face) without having done any sample preparation work. After applying careful thought to all these ideas and augmenting this with one or two papers, it seems to me that the most important aspect is sample preparation. Furthermore it would seem that if I am working on an exploration project and want an indication of whether my mineral/metal of interest is high or low I could use a handheld device to get an indicative result. I could further improve the quality of this result by breaking/pulverizing the rock using field sample preparation equipment. The other favoured approach would be to take a thousand data points and average them out.

The possible sources of error that would result from shooting beams directly onto a mining face include sample heterogeneity, limited depth of X-Ray penetration, moisture content (moisture can lead to a reduction in the actual value by up to 60%). Since mining faces are washed, ironically to assist with collecting better samples, moisture is always present. The ideas of constantly drying a face, generating a flat surface and as some have suggested collecting a lot of data points to build a calibration database over time doesn’t sound appealing.

Given that the mine in question already has a laboratory, it would seem, all things considered, that replacing it (the lab) with a pXRF would be unwise. Provided that the samples are collected in a representative manner the laboratory provides a superior environment for sample preparation and analysis. It is possible that I could be missing some important aspects of this discussion in which case I would once again appreciate a guiding hand.

I've thoroughly enjoyed reviewing all your comments. Thank you to all.

Have a little faith friends 😉 OF COURSE, it is only AFTER the surface issue has been dealt with properly that the avid hXRF sampler can address the rock mineralogy/texture/grainsize heterogeneity issues. This is, of course, handled by performing a 'composite sensor sampling' i.e. by using a set of individual analyses (each with a much too small beam width footprint for comfort, but look what they can do in too). If and when there is time, we carry out a pilot experiment (background method preparation), or employ as high a number of individual field 'shots' as can be accommodated by surface area prepared by the FRAT (typically for igneous and metamorphic rocks 8-12 shots; sediments and shale’s are EASY). All is explained in Martin Holding's B.Sc. thesis, referred to above, in which we set ourselves the goal: WHAT does it take to improve today's "blind hXRF shooting"? Using TOS and a hefty dosage of geological common sense. It did not take all that much 😉

I appreciate very much your efforts in contributing to the improvement of operating conditions for better analyses. I wish to know more about the capabilities of the device, and its development plan (I am also a R&D project leader with BRGM). However, there will be always intrinsic limitations due to the relative size of the X beam and mineral grains - even if they can be addressed by statistical measurement.

There will be always intrinsic limitations due to the depth of penetration of the X beam. Keep in mind that they affect variably the analysed elements - the lighter elements being more affected. This may require geochemist's skills to understand properly the measurement results. 

The comments have all centered on XRF. The alternative technology is Laser ablation (LIBS) which employs emission spectroscopy. The tool is capable of 60 samples per second and could shoot at thousands of spots on the face from a distance of some meters. Again, the penetration is small (an mm or two), but the sample is better and fast.

What we really need is a fast channel sampling tool that can be used U/G with full sample collection. Such a tool would provide a sample of adequate mass that could be returned to the lab for a quick XRF automated fusion, followed by good accuracy XRF. I am looking for folks with whom to collaborate on such a sampling tool development, as I think this is the correct way to go. If the channel samples are 'mechanically correct' (constant cross-section with full material recovery), they can feed back results into the mine plan as well.

Agree 100% as to these principal limitations. The hXRH is not, NOT, although much hoped for, THE final "shot-and-forget" field weapon. But with insight many of its intrinsic limitations can be reduced - is all ,-)

LIBS are less represented here because it is still by far more experimental than pXRF. Its promising points are a better aptitude to deal with light elements and the possibility to use it at distance from sample.

Its current drawbacks are unpredictable results with matrix changes, and the necessity of custom calibrations for each specific sample composition. In a mining context, it could be helpful if the matrix is homogeneous enough to allow the development of a site-specific calibration.