Averages have the ability to express an overview—but, used incorrectly, they can hide a lot of variation that can have a serious impact on performance. Let's dig in.
This is the second article on concepts to consider when looking at optimising fleet performance. The previous article focused on Truck Factor accounting. Enjoy.
Averages may well be the best way to hide inconsistent performance—but we use them every day. How many of us can honestly say we have not used averages to blotch out instances of poor performance? Well, if you can do it, so can those who provide you with feedback.
We mine in real time, load by load.
If we lose track of the fact that what we deliver at the end of a measuring period is, ultimately, the sum of a myriad of small actions… The collective outcome of all those small actions is totalled up into what is delivered at the end of a measurement period. We do not drill at average penetration rates, load at average loading rates, or haul at average hauling rates. We drill at instantaneous rates, load at instantaneous rates, and haul at instantaneous rates. And unless we get our heads around managing these small, incremental, instantaneous actions, our industry performance will remain… well, average.
In the past, it was simply impossible to record (never mind manage) all of these small actions in the average mining operation. Tens, if not hundreds of thousands of drill holes, shovel and truck loads, and a myriad of supporting activities. But things have changed, and we need to probe how technology can assist us in dipping down to a level where we can make sense and use the incremental, instantaneous data to improve our performance.
In soccer, if the striker misses one shot to the left and then another to the right, what was his average performance? The average of the two shots was spot on, right? Reality is… he wasn't. He missed both shots.
Two wrongs, on average, don't make a right.
Let's use the example of truck payloads. Picture the following scenario: truck payload has been a key discussion point for some time on an operation, and the feedback at today's meeting is an easy one—we have an average payload of 98% of rated payload. How do we get to that number? Easy: we take the tonnes moved and divide by the number of truckloads. Cheers all round, job well done, right? Well, let's consider.
As we said previously, trucks do not load average payloads—they carry instantaneous loads. When we understand this, we can analyse our data accordingly. If you put two histograms side by side — both showing the occurrence of specific payloads over a measured period, both with the same mathematical average — the "average" payload reads as spot on in either case. But Scenario 1 might be hiding 26% of loads under-loaded (wasting resources) and 14% of loads above the 10:10:20 guideline. That has serious safety implications: the truck braking system is not designed to stop a truck overloaded by 30%, frame life is compromised, tyres overloaded… the list goes on.
But on average, we did okay, right? Well, maybe not so after all. If we want to get real about performance, we need to dive deeper into the individual actions and apply simple statistical tools to assist us in understanding how things really look. We could then create measurement systems that give the feedback to operators in real time, and assist them in making decisions on the fly. It is being done elsewhere, and it can be done in mining.
The decisions a planner needs to make live in the tails, not the mean.