The Plant Doesn't Run on the Average Ore

When I was Plant Manager at a large acid-leach operation, I became a willing student of its history. The plant had been commissioned in the 1970s, and several of the long-serving staff had lived through the start-up and the difficult years that followed. They enjoyed telling me every detail. The story was still referred to on site almost daily. What follows is their account, as it was told to me. I wasn't there.

“A large acid-leach plant commissioned in the 1970s ran into trouble because the samples and testwork behind its design had not captured how widely the ore varied, in hardness and grindability, carbonate content (which drives acid consumption) and grade. A plant never processes the average ore: hard ore costs throughput, carbonate-rich ore costs acid, low-grade ore costs revenue, and easier ore often doesn’t give it all back. Blending, the main remedy, is limited by the mining sequence, so it needs large stockpiles and handling facilities or leaves the plant off the design average. Testwork can’t cover every possible variation, but it should cover a representative set of samples and include some of the extremes.”

A start-up that went wrong in familiar ways

As I was told it, the plant ran into trouble almost immediately. The samples and the testwork the design rested on had not captured how widely the ore varied across the orebody. The variability covered hardness and grindability, carbonate content (which determines how much acid is consumed for each unit of product recovered) and the grade of the ore itself.

Each of these varied on its own, and they varied together. The effect on the economics was large and, I was told, usually in the unfavourable direction. Throughput, grind size, acid consumption and overall recovery have been managed since mostly by blending, with some success over the years.

Why the average hides the problem

A composite sample, or a small number of them, describes the average ore. A plant never processes the average. It processes a sequence of different ores, one after another, and the economics don't respond to that variation evenly. Hard ore costs throughput. Carbonate-rich ore costs acid. Low-grade ore costs revenue. Ore that is easier than average often doesn't give all of it back, because some other part of the plant becomes the limit. That asymmetry is why "it could be softer as well as harder" is an argument that deserves care when it is used to justify a smaller or cheaper design.

Blending, the main remedy here, works by making the plant see something closer to the average ore it was designed for. It is a way of correcting, in operation, for variability that was not tested in the design.

The real problem with relying on blending to fix ore variability is that mining has to proceed in a logical order, constrained by space, depth, time and throughput. In other words, the mine plan needs to be followed, and that means blending either needs massive stockpiles and handling facilities, or the blend will still be off by some margin from the average the plant was designed for.

Testwork is not a fixed list

Testing variability means actually testing it: samples from across the orebody, in the zones the mine plan will draw from, and not only a blended composite. It also means a programme that can change. In my experience, results don't always say what the programme expected them to say, and a good programme has to be adapted as unexpected results come in. That takes experienced engineers designing and running it, not just specifying it. The alternative is making the facts conform to the theory, and that happens more often than we like to think.

What this does and doesn't mean for a board

None of this is an argument for skimping on testwork or on representative sampling. It points the other way. The cost of not knowing didn't disappear here. It arrived later, in higher operating cost and reduced revenue, and it kept on arriving year after year. Nor is it an argument that every project can afford testwork covering every possible variation in an orebody. None can.

That said, testwork should cover a sufficiently representative set of samples to identify the variability in the orebody, and should include some of the extremes, to show how the plant is likely to respond to them.

What a board can reasonably ask is two plain questions. How much of the orebody's variability has the testwork actually covered? And what does the design assume about the part that hasn't been covered, and what happens to the economics if that part turns out unfavourable? Those won't find every surprise, but they put the known unknowns in front of the board in plain words, rather than leaving them to be assumed away because the report looks thorough.

The testwork told the truth about the samples it was run on. The operators discovered what the actual orebody had to say about that testwork.