🧠 Stop Chasing Correlations—Validate the Function: What NDT Professionals Can Learn from Papadakis

In the world of nondestructive testing (NDT), we often hear phrases like:

But here’s the question: Are we chasing the right relationship? Or just validating a proxy for a proxy?

📊 What’s the Difference Between a Correlation and a Function?

In his landmark 1993 article in Materials Evaluation, physicist Emmanuel P. Papadakis laid out a logic chain that should be required reading for every NDT professional developing or validating a new test method.

He argued that many organizations fall into a trap:

• Instead of linking a new NDT method to the actual material property of interest (like strength),

• They correlate it to an older test (like Brinell hardness), which itself only loosely correlates to strength.

Papadakis showed that if:

• R₁₂ = correlation between property and old test

• R₂₃ = correlation between old test and NDT signal, Then:

• R₁₃, the correlation between property and NDT signal, is always less than R₁₂ × R₂₃.

Translation? If you link your new method to an old method instead of to the true property, you’ve mathematically limited how good your results can ever be.

🔎 Real-World Example: Eddy Current vs Brinell in Cast Iron

In one case, Ford engineers used eddy current to estimate yield strength of gray cast iron.

Instead of validating ECT directly against strength, they were required to correlate it to Brinell Hardness Number—because BHN was the “accepted” standard.

The result?

• Good parts were rejected

• Bad parts got through

• Over one-third of the components needed to be re-checked with BHN anyway.

If they had validated the eddy current method directly to strength, they could have built a more powerful, more accurate, and faster inspection solution.

✅ When It Works: Ultrasonic Velocity and Nodular Iron

On the flip side, Papadakis highlighted a success story:

Engineers used ultrasonic velocity to predict tensile strength in nodular iron crankshafts.

Because velocity has a direct physical relationship to modulus—and graphite morphology affects both—this test:

• Was repeatable

• Had a smooth nonlinear function

• Delivered low false positives and negatives.

The difference? It measured what mattered—not what used to stand in for it.

💡 Sidebar: Are You a Scientist or an Engineer?

Papadakis noted a culture clash in validation:

• A scientist wants controlled experiments, one variable at a time, clean functions, and error bars.

• An engineer wants a correlation that works across dirty shop-floor conditions, batch-to-batch variability, and noise.

Good NDT method development needs both.You need the rigor of the scientist and the practical instincts of the engineer.

But most of all, you need clarity:👉 What are you really trying to measure?

🔁 Final Thought: Validate the Destination, Not the Shortcut

If you’re developing a new eddy current procedure, probe, or acceptance criterion, ask yourself:

• Are you correlating your signal to something useful?

• Or are you just checking boxes by mimicking an older method?

🚫 Correlation ≠ validation.✅ Function = understanding.

To build the future of NDT, stop chasing someone else’s shortcut. Build your own map.