Failure Analysis: Learning from Broken Parts

Broken parts are the best teachers

The failure rung earlier in this ladder taught you the mechanisms — how a crack runs, how fatigue creeps forward a striation at a time, how metals cleave in the cold. This guide runs the film backwards. You are no longer designing a part that might one day break; you are standing over one that already has, holding the wreckage, and asking the hardest question in engineering: *what actually happened here, and how do I make sure it never happens again?* That reverse question is a discipline of its own, called failure analysis, and it is equal parts materials science and detective work.

Here is the reason it deserves its own guide in a rung about selection and design: almost every rule you now trust was written in the aftermath of a failure. Rounded window corners on aircraft, notch-toughness requirements for ship steel, mandatory bridge inspections, the very idea of a factor of safety — none of these were derived cleanly on a chalkboard. They were paid for by parts that broke and people who studied why. A design engineer who never learns to read broken parts is doomed to re-buy those lessons at full price.

The investigation: a detective's procedure

A good failure analysis is not a guess dressed up in jargon; it is a disciplined march from the outside in, cheapest and least destructive tests first, cutting metal only once the surface has given up its story. The single golden rule underlies all of it: preserve the evidence. Once a clue is destroyed it is gone forever, so the order of operations matters as much as the tests themselves.

1. Gather the history before you touch the part. What was it, how long had it served, what load and temperature and environment did it see, and did anything change just before it broke? Half of all root causes are hiding in this paperwork, not in the metal.
2. Photograph everything, then look with the naked eye and a hand lens. The overall shape already sorts the case: a necked, fibrous, grey break says ductile overload; a flat break at 90 degrees with chevron marks says brittle; concentric beach marks say fatigue. Follow the markings back like arrows to the origin.
3. Zoom in with the electron microscope — this is where fractography earns its keep. Dimples confirm ductile tearing, flat river-patterned facets confirm cleavage, and a ladder of fine parallel striations is the unmistakable fingerprint of fatigue, one line per load cycle.
4. Only now cut, mount, polish, and etch a cross-section to read the microstructure, and fire chemistry at it — an energy-dispersive detector on the microscope names the elements in a corrosion pit or a stray inclusion. Confirm the alloy is actually what the drawing called for; the wrong grade is a classic buried cause.
5. Assemble the timeline and name the root cause. Where did it start, why did it start there, what drove it forward, and what finally tipped it into fast fracture? A finished report does not stop at 'fatigue' — it says why the fatigue crack was allowed to begin, and what to change so the next part does not follow it.

The line-up of usual suspects

Most in-service failures fall into a handful of families, and each leaves a distinctive fingerprint. Learning the line-up turns a bewildering broken part into a short list of suspects you can test one by one. Notice how many trace back to a single geometric villain — a sharp corner, a thread root, a weld toe — because a stress concentration is where nearly every crack chooses to be born.

MODE            TELL-TALE FINGERPRINT ON THE SURFACE          WHERE IT USUALLY STARTS
-----------     ------------------------------------------     -----------------------
Ductile         necked, grey, fibrous; dimples up close;       gross overload / wrong
  overload      part visibly bent before it let go             load path / undersized

Brittle         flat ~90 deg break, bright facets, chevrons    a notch + low temp
  fracture      pointing back to one origin; NO warning        (below the DBTT)

Fatigue         smooth 'beach marks' (clamshell rings) +       a stress raiser under
  (most common) fine striations; small final overload zone     millions of cycles

Stress-corr.    branched, brittle-looking cracks in a          tensile stress + the
  cracking      normally ductile metal; needs a specific       wrong environment
                environment (e.g. chloride on stainless)       together

Creep           cavities strung along grain boundaries;        constant load at HIGH
                oxide scale; part sagged/bulged first          temperature, over time

Wear / erosion  scored, galled, or polished surfaces;          rubbing, particles, or
                material simply removed, not cracked           fluid over long service

Naming the mode is only half the job; you must also name the root-cause bucket, and there are five: design (a sharp corner, too little material, no redundancy), material (wrong grade, a bad heat, hidden porosity), manufacture (a weld defect, a quench crack, decarburized surface), service (overload, an unplanned chemical, an impact), and maintenance (a missed inspection, the wrong lubricant). Two honest cautions here. First, real failures are usually a chain, not a single villain — the Comet cracked because a stress-raiser met a fatigue load met an optimistic inspection interval, all at once. Second, resist the reflex to blame the metal: bad material is statistically the rarest of the five buckets, yet it is the first place blame is thrown. Guard against that bias, and note that fatigue and its cousin stress-corrosion cracking together account for the large majority of unexpected structural failures, dwarfing simple overload and slow creep.

Case files: the failures that rewrote the rules

The most powerful way to learn failure analysis is to sit with the great cases, because each one is a rule you now obey without thinking. Take the de Havilland Comet, the world's first jet airliner. In 1954 two of them broke apart in mid-air, and the investigation — including sinking a whole fuselage in a water tank and pressurising it thousands of times until it cracked — traced the origin to fatigue cracks starting at the sharp corners of near-square cutouts in the pressurised cabin. Every takeoff pumped the fuselage up like a balloon; the stress concentration at those corners quietly grew a crack cycle by cycle until it ran. The lessons are now aviation gospel: round every corner, design so a crack is caught and stopped by structure before it becomes catastrophic (fail-safe, damage-tolerant design), and prove it with full-scale fatigue testing rather than analysis alone.

Now the Silver Bridge over the Ohio River, which collapsed without warning in 1967 and killed 46 people. The whole span hung from a chain of flat steel eyebars, and one eyebar had a tiny crack in its eye, a fraction of an inch deep, grown by corrosion and stress-corrosion cracking over decades in a spot no inspector could see. When that single link failed there was no second load path, so the entire bridge came down like a dropped necklace. Two rules were born from those deaths: never build a critical structure with no redundancy, and inspect ageing infrastructure on a real schedule — the modern national bridge inspection programs trace directly back to this wreck. (The same reasoning, one rung back, explains the WWII Liberty ships that split in cold water: a welded hull gave a brittle crack one continuous path with nothing to arrest it.)

Closing the loop: from wreckage back to the drawing board

Failure analysis only earns its cost when the finding travels back into design and material choice — and that is why it belongs in this selection rung. A well-diagnosed failure hands the designer concrete moves: soften the geometry so no corner concentrates stress, drop the working stress by choosing a bigger section or a higher factor of safety, switch to a tougher grade even if it costs a little stiffness or money, add a redundant load path so one crack cannot end the story, or set an inspection interval short enough to catch a crack while it is still small. Notice these are exactly the levers of materials selection you met earlier in this rung — the failure just tells you which lever to pull, and how hard.

Stay honest about the limits, though. A failure investigation runs on hindsight, and hindsight makes the cause look obvious once you know it — the real skill is finding it before the next part breaks. Root cause is rarely a single line; it is usually a chain, and stopping at the first plausible suspect ("the steel was bad") is how investigators get it wrong and the failure repeats. Non-destructive testing can miss a crack smaller than its resolution, so "we inspected and found nothing" is never a guarantee. And remember the deeper lesson from brittle solids: a part fails at its worst flaw, not its average one, so a single strength number can flatter a material that will still fail early wherever the worst defect happens to sit. The point of learning from broken parts is humility with teeth — respect the small beginning, design so it cannot grow to critical, and let every wreck you study make the next design a little harder to break.