Impact of #Reblocking on #Pit #Optimization

In our line of work what I think is key is how your block model compares with the actual (ie the Mill vs Block model recon), and what ever block model used in that exercise can be used for your optimization, because you know and understand how to bring it close to the actual. So in short I don't think Reblocking should affect the out come of an optimization exercise provided recon is fully understood.

I have had more than 25 years experience in pit optimization which has shown over and over again that appropriate reblocking has little impact on the selection of a robust pit design. So I was initially surprised by this post! The process described in above article as reblocking is actually the process which used to be called regularization in Datamine. Whereby all the individual blocks are averaged together to find a new average for this larger “superblock”. As the blocks get bigger it is inevitable sub-grade material will get included in the average. Since the bigger blocks just has one value when it comes to any full block by block evaluation (not just pit optimization) it will clearly show that the blocks become heavily diluted often to the point they can no longer be considered ore grade and the block size increases. The combined extra dilution (material that has to be processed but generates insufficient revenue to pay for its processing) and loss of resource because the large block is now below cut-off will quickly compound losses. So I do understand the outcome described here. The important piece of understanding missing is that during any successful pit optimization process you need to be considering the grade distribution (as well as you can model it) that can be expect when you come to mine the material. There are two aspects to understand this, the volume variance affect and SMU (selective mining unit). There is a fair bit of earlier geostatistical literature that clocks these concepts in magic, but they are pretty simply. Firstly the volume/variance affect, if you collect any smaller samples they will show more variability. Drill hole intersections measuring perhaps a metre of core length can be expected to show variability in grade and perhaps some very high grades. If however when you come to mine this and the mineralization is narrower than the excavator bucket some low grade will inevitably be included in a bucketful, by the time the truck is filled the grade will become even more averaged. In mathematical terms the variance decreases with the size of the unit being sampled and this relationship probably holds over a whole geological domain if not the whole mineral body. The selective mining unit is a concept, rather than physical entity, representing the smallest unit you are able to successfully mark out and the mine as a single defined unit. Unfortunately selective mining units can vary by mining technique, benching size, equipment size, mineralization orientation and continuity, just to mention a few factors. They can be difficult to precisely predict, however they are likely to be small, perhaps a few truck loads. Often this is smaller than the block model size on which the geostatistcal estimate can be safely based. So the pit optimization process needs a mechanism that can preserve the grade distribution at this SMU size, and a widely used approach is the idea of parcels (or partial blocks), These are smaller units within the bigger block, but preserving the grade distribution (they may not necessarily occupy a known position, it is just know that they are somewhere within that block). When the pit optimizer is evaluating the block, it still has to decide to mine or not mine the whole block, but you can read through the parcels and decide which parcels or above cut-off, therefore mined at a profit, and which are below cut-off, therefore discarded as waste. The safe step, now widely known as reblocking, is to maintain all the parcels as the block size is increased. So it doesn’t matter what is the super block size your estimate of the above cut-off material will be the same. I am not familiar of with ThreeDify’s flowpit software, however if it is not capable of handling parcels (partial blocks) then clearly it should not be used for reblocking. This however in no way should this be extrapolated to the many other pit optimisation packages, that do honour parcels, dominantly those based on the Lerchs-Grossmann algorithm and specifically the Geovia/Whittle programs. Most established mining packages (eg Surpac, Vulcan, Minemap, Micromine, Datamine) will do this safe reblocking (and create parcels) as the model is passed out to the pit optimiser.

In addition to the dilution issue, the efficiency of the solution algorithm (as I found it is a new implementation of the LG) should be taken into account. When you use fast solution techniques, you need to verify the capability of the solution techniques in obtaining optimum (or at least near optimum) solution. That can be done either by comparing FlowPit with re-blocking factor of 1, with those package which use re-blocking, or comparing FlowPit with standard solvers (e.g, CPLEX) for small to medium size instances.

I would be careful to say here that the project is making more using small blocks than it would using the larger blocks. This is more of a matter of dilution, and the right answer there is to dilute the model to a selective mining unit, or include either dilution and ore loss that will acceptably model this (I prefer the first method). When you do the re-blocking in a program like Whittle, then you don't actually dilute the model on the interior as it maintains the partials of the various materials inside of the model, you just lose the ability to state where exactly it is in the model. Thus, when re-blocking in this manner, you really don't lose value. What you lose is the resolution on the outer portions of the pit shells. Just a couple of thoughts.