Why Master Defects Mislead: A Closer Look at Preventive Crack Testing
We continue our conversation with Dr. Eddy on crack inspection in production, this time digging into why good-part-teaching outperforms traditional defect-based calibration, and how ibg's iPMFT method makes it possible.
What is the big advantage of using good parts for crack detection?
Dr. Eddy: “Good parts” and “good surfaces” can be specified in the strict sense: they can be described precisely in terms of surface roughness, dimensional tolerances, geometric tolerances, and other measurable parameters. “Defects”, by contrast, cannot be specified sufficiently because they are in many cases unpredictable. That is why the requirement “crack-free” is often stated. This requirement is also verbalized as “technically crack-free”. However, since this specification is not quantifiable, a “minimum defect” is defined (“master defect or flaw”), simply to have a benchmark for approving production and test processes. Yet, that this is only a workaround, not a genuine solution to the underlying problem.
Is it really that “unprofessional” to define a master defect? After all, NDT experts and operators around the world refer to a so-called “standard defect” or “crack standard”?
Dr. Eddy: What is “unprofessional” is to “calibrate” or set up the crack testing based solely on such a defect. Since these defects are specified in terms of size and shape, they must be reproducible. That is why they are created artificially. This makes them different from naturally occurring defects in many relevant aspects. And as a result, the eddy current signal of an artificial defect can differ unpredictably from that of a natural defect.
Here is an example: On the left (Figure 1), a natural hardening crack, thin and zigzagging. On the right side (Figure 2), a well-rounded, typical artificial defect. The two will produce very different eddy current response modulations That's what makes it risky to set up or "calibrate" eddy current testing using only a master defect. It is a good way to ensure minimum detection, but not more than that.
That really makes sense. Therefore, an artificial defect doesn't help at all in establishing a reliable crack inspection process in production?
Dr. Eddy: You can’t really make a general statement like that. It is advisable to have a defined master defect or flaw, but not for “calibrating” the eddy current testing. Rather, it should be used for monitoring during operation. Therefore, master defects should be used for verification, and not for “calibration” and set up. For that, you need good-part teaching with iPMFT (see Dr. Eddy No. 1 and No. 2), the way ibg and others do it.
But then why doesn't everyone do it that way? I've heard some people are still use eddy current devices that must be manually adjusted on defects?
Dr. Eddy: That is a very good question. In the past, the technical and computational capabilities simply didn’t allow for the multi-filter analysis of the eddy current signal required for good-part calibration and testing at the speed modern production demands. But things have improved. By digitizing the signal at the very beginning of the chain harnessing massive computing power, we can now break the signal down into many narrow filter bands, the “fingerprints”. Each of these filter bands is then evaluated individually. In what is known as the polar field, a boundary line (tolerance zone) is drawn around each of the signals picked up by the good parts (30 tolerance zones in total). This is exactly what is known as deterministic machine learning, as practiced by ibg NDT.
Could you explain in more detail exactly how ibg performs the good-part “calibration”?
Dr. Eddy: Basically, this works exactly like ibg’s well-respected microstructure inspection using iPMFT. The term stands for “Preventive Multi-Filter Testing” by ibg.
Setting up iPMFT testing requires only good parts or surfaces. A master defect or flaw is optional and used only for verification, as mentioned above. These good parts are scanned with the crack sensor, just as in the actual inspection, and this can even be done in the testing machine's automatic mode. During this process, the signals generated by the surface properties and material texture of the parts are recorded and broken down into 30 bandpass filters. For each filter band, the polar field (the xy signal) is captured, and "envelopes," or tolerance zones, are drawn around these signal clouds. ibg calls these the "fingerprints" of the good parts.
To represent natural variation in production, up to a hundred parts or surfaces can be taught. However, since tolerance zones can be expanded using an adjustable factor, 8 to 13 good parts are usually sufficient. A master defect or flaw can be used to verify the teaching, and outliers should be inspected anyway to help evaluate production variance. After this fast, easy teaching process, test parts are then compared against the tolerance zones preventively, even against unexpected defects and flaws.
And how does it look on the display?
Dr. Eddy: In Figure 3 you see a typical screen for crack and grinding burn testing on an ibg controller. On the left is the bar graph withthe 30 band-pass filters. Green bars indicate that the filter signal is below the threshold established from the good parts; if a signal exceeds the threshold, the bar turns red. On the right side, the signal of the selected 100Hz filter is displayed in detail. The xy display shows the “butterfly”, the typical shape of the tolerance zone applying the expansion factor. Here you can see the orientation of the actual signal in orange against the tolerance zone in green. In this case, everything is within range.
And what exactly are the advantages of this “preventive” iPMFT method?
Dr. Eddy: Let me highlight the most important ones:
- Comprehensive detection from a single teaching process. One teaching session on good parts or surfaces can detect all types of defects, including grinding burn.
- No need to predefine every defect type. Defects can be detected even if you haven't anticipated or previously encountered them.
- No eddy current expertise required. Teaching and operation of an ibg instrument or controller are fully automated, since deterministic machine learning calculates the tolerance zones. Thanks to intuitive visualization, outliers are easy to spot if a bad part or surface is mixed in.
- Deeper analysis for experts. NDT and eddy current specialists can view signals from 30 perspectives without manually adjusting filter settings, phase angles, and gains. Manual adjustment against known master defects is still easily possible if desired.
That sounds too good to be true for eddy current testing on crack and grinding burn. What's the catch?
Dr. Eddy: The challenge is that new technology always breaks with traditions and pushes beyond established norms and standards. Many people are wary of new technology because they're set in their routines or feel bound by them. It’s much like the transition to electric vehicles. But progress can't be stopped. As we say in Germany: “The better is the enemy of the good” (Das Bessere ist der Feind des Guten).
