One of the things we were taught most thoroughly at our Air Force technical school was how to look for cracks.
Not the visible kind. The ones too small to see, that grow a little every flight until a part fails. Many, many tests, and all of them with the same constraint: you cannot cut the part open to look, because then it is no longer a part. This is called non-destructive testing.
The fluid that glows
The one everyone remembers is the fluid. You pour a special liquid over the surface. Then you turn the lights off, and the cracks glow.
It works because the liquid is thin enough to seep into a crack far too narrow to see, and it carries a dye that glows under ultraviolet light. The excess is cleaned off the surface, and what stays behind is only the liquid that got into a gap. In the dark, under the lamp, a crack you could not see a minute ago draws itself as a bright line. It is called fluorescent penetrant inspection.
Looking inside
A penetrant only finds what reaches the surface. For flaws inside the material there is X-ray, the same idea as at the doctor: a crack or a void lets more radiation through than solid metal, and it shows up on the image.
Ultrasound does the same job with sound. A probe sends short pulses into the part and listens for the echoes. Solid metal sends the pulse back from the far wall. A crack sends back an echo too early, and you see it as a spike on the screen.
Eddy current testing uses a coil instead. Its changing magnetic field drives small electric currents in the metal, and a crack near the surface gets in their way. The instrument picks up the disturbance, without paint, fluid or radiation.
There were many more methods, each suited to a different material, shape or kind of flaw. The lesson was that there is no single test. You choose it by what you are afraid of.
The tap test
The simplest one is a tap test. You take a particular kind of small hammer, tap the surface gently, and listen.
A sound, solid area answers with a clear note. Where layers have come apart or something is wrong underneath, the sound changes and goes dull. As I remember it, there was a threshold in decibels: past it, you knew something was up.
It looks almost too simple to be engineering. It is measurement all the same: a defined stimulus, a response you can compare, and a limit that turns a sound into a decision.
Why so many ways to look
Aircraft parts are inspected like this because of how they fail. Rarely all at once. A crack starts small, under vibration and changing loads, and grows a little every time. Catch it early and it is a part to replace. Miss it and it is the reason something breaks.
What stayed with me is the attitude more than any one method: assume the flaw is there, and choose the test that would show it to you.