Data Integrity & The MSA

Is Your Data Lying to You? Why You Must Verify the Measurement Before You Diagnose the Mine
We live in the era of "Big Data." Mining dashboards are filled with real-time tonnages, recovery percentages, and reagent consumption rates. We make multimillion dollar decisions based on these numbers. But here is the uncomfortable truth: Most mining data is "noisy," and some of it is outright wrong.
As a Lean Six Sigma Expert, I never trust a metric or KPI until I’ve seen the Measurement Systems Analysis (MSA). If your flotation cell sensor says one thing, but the manual assay says another, you aren't managing a process, you are chasing a ghost.
The Cost of the Ghost Signal
I recall a recovery improvement project where the mill team was convinced they were losing gold in the tails. They had spent months adjusting the circuit, chasing "noise" in their sensors. When we performed an MSA, we discovered the sensors weren't the problem, the sampling method was. The samples were being taken from a location that didn't represent the true flow.
The team was optimizing based on a lie.

Example of how manual froth depth to calibrate the sensors may be erroneous based on the height of operator taking samples
In mining, data integrity fails in three ways:
Calibration: The sensor is physically off.
Precision (Repeatability): The same sensor gives different readings for the same material.
Accuracy (Bias): The sensor is consistently higher or lower than the "true" value.
If you feed this "ghost data" into an AI or an optimizer, the machine will make the "mathematically correct" decision for the "physically wrong" reality. This is how you end up with mill "crashes" or unexplained tonnage drops.

Missing process to validate the Metrics in Most Mines; A Calibration of Sensors or an ISO Audit is not enough
The MSA: Your Data Truth-Serum
At Analyze and Improve, the MSA is a core component of any of our improvement initiatives. Before we analyze your bottlenecks, we analyze your data integrity. We use a Gage R&R (Repeatability and Reproducibility) study to ask:
Can the equipment repeat the measurement?
Can different operators get the same result?
Is the measurement system capable of detecting the actual process variance?
If your measurement system’s "noise" is larger than your process’s "signal," you cannot use statistics to improve. You must fix the measurement first.

Typical Gage Reproducibility and Repeatability Study showing unreliable Measurement System
The Foundation for Digital Transformation
This is the non-negotiable step for Operational AI. AI models are hungry for data, but they are incredibly sensitive to quality. If you have "drift" in your sensors, your AI model will drift too.
In our "Prep for AI" audits, we often find that a $10,000 recalibration and sampling standardization effort yields a higher ROI than a $1,000,000 software install. Why? Because clean data makes every other decision easier.
The Challenge for You This Week
Pick your most critical KPI (e.g., Mill Recovery or Drill Accuracy).
Ask your team: "When was the last time we performed a Gage R&R not just a calibration audit on the sensors driving this number?"
If the answer is "at commissioning" or "we don't know," you are managing with a blindfold on.
Let's stop guessing and start measuring.
Asim Khan, PhD
Helping Mines Bridge the Gap Between the Dirt and the Data.


Comments