When a component fails, the instinct of every good operations team is to fix it and restart. That instinct is exactly what destroys failure investigations. The evidence that explains a failure is fragile: it can be wire-brushed off, cut through, rusted away or simply thrown in a skip — usually within the first day.
It is also time-bound. Evidence degrades on its own, with nobody touching it. Oxidation forms over a fresh fracture surface and masks the very features that identify the mechanism. Residual corrodants keep reacting and alter the surface they sit on. Water, wind and airborne particulates erode fine detail — the beach marks, striations and initiation features that carry the answer. A part left in the open for a fortnight can lose evidence that no laboratory will recover, even though no one laid a hand on it.
What you preserve in those first hours largely decides whether the eventual answer is confident or bounded. MTIS general guidance to preserve evidence includes:
Keep the part — all of it
Retain the failed component and every piece of it, including fragments and debris. Both halves of a broken part matter: fracture surfaces are read as matching pairs, and the half that looks "less interesting" often carries the initiation site. Keep associated hardware too — the bolts from the failed flange, the bearing from the seized shaft — because failures are rarely confined to the part that finally let go.
Keep what it ran against
A failure is a relationship between two surfaces, and only one of them broke. The shaft the bearing turned on, the seat the valve closed against, the bore the piston ran in, the flange face the gasket sat on — the part that did not fail carries the record of how the two behaved together. Wear patterns, scoring, fretting, contact marks and misalignment evidence live on the counterface, and they are frequently the only remaining indication of whether the failed part was ever loaded the way its designer assumed.
Keep it, and keep it identifiable. If it has been left in service, photograph it in position and record that it is still installed — we may need to look at it later, and it will have kept running in the meantime.
Quarantine the sisters
Find the other components of the same type, the same batch, the same heat or cast number, and the same duty. They matter twice over. As evidence, they show what an unfailed example of the same part looks like, which is how we establish whether the failure came from something wrong with that one part or something wrong with all of them. As risk, they are the next thing to fail.
Take them out of the store and set them aside before they get fitted. If any are already in service, list where they are and how long they have run. A supplier facing one failed part argues bad luck; a supplier facing a batch does not.
Leave things where they are
Try to avoid removing or moving anything where it is possible not to. Position, relative distances, angles and orientation all carry information, and that information may support or dispute an eventual diagnosis. Which way a fragment travelled, how far it landed from its housing, which face it came to rest on, how two broken halves sat relative to one another — none of that survives a tidy-up, and none of it can be reconstructed afterwards from the parts alone. If something must be moved, photograph it first, then mark and record where it was and which way round it sat.
Do not clean anything
Deposits, corrosion products, oil and process residues on a fracture surface are evidence, not dirt. They date the crack, identify the environment and often distinguish one mechanism from another. Cleaning, wire-brushing, pressure-washing or acid-dipping a fracture surface can erase the very features the investigation needs. Why cleaning destroys evidence →
Photograph in place, before removal
Before anything is moved, photograph:
- The failed component in position, from multiple angles, with something for scale.
- The surrounding context — supports, connections, adjacent damage, leaked product.
- Any fracture faces, as found, before any handling.
- Nameplates, markings and identification numbers.
Phone photographs are fine; taken-before-disturbance beats perfect-but-late every time.
Protect the fracture surfaces
Do not touch fracture faces with fingers, do not fit the halves back together ("just to see"), and do not let them rub in transport. Fitting mating faces together bruises the fine features fractography reads. Wrap surfaces in clean, dry material; for steel at risk of rusting, a light coat of clean oil on the fracture only after photography is acceptable if transport will take time — note that it was applied.
Store it dry, and not in a sealed bag
This follows directly from the evidence being time-bound. A component sealed in a plastic bag while still damp creates its own humid microclimate, and it will corrode faster inside that bag than it would have done on a shelf. The fracture surface is the first thing to go.
Dry the part gently before packing. Store it somewhere dry, indoors, at stable temperature, away from washdown, welding fume and salt air. Silica gel or a vapour-phase corrosion inhibitor in the container is better than a tight seal. Support the part so nothing rubs against the fracture in transit. Do not put it in a workshop drawer with other steel, and do not leave it outside until someone has time to look at it.
Take the fluids, not just the solids
Debris and deposits are the obvious samples. The fluid the component was running in is just as much evidence, and it is the one nobody takes.
Sample the process fluid, the lubricant, the coolant, the hydraulic oil, the feedwater — whatever was in contact with the failed surface — and sample it before the system is flushed, topped up or changed. Chemistry at the moment of failure explains corrosion mechanisms, tells us whether an inhibitor was actually present, and shows contamination that nobody knew about. Once the system has been drained and refilled, that information is gone permanently and cannot be reconstructed from anything else.
Use clean containers, fill them nearly full to limit trapped air, label each with the point it was drawn from and the date, and keep them out of sunlight.
Export the process data today
Of everything on this page, this is the item with the shortest deadline, and it is the one most often lost.
Control-system historians, condition-monitoring systems and vibration analysers almost all record to a rolling buffer. High-resolution data is kept for days or weeks and then thinned to averages or overwritten entirely. The one-second trend that shows what actually happened in the ninety seconds before the failure may cease to exist by the end of the month, while the broken part sits safely in a box waiting for someone to look at it.
Export the raw data now, at full resolution, for a window that starts well before the event — days before, not minutes. Include process conditions, alarms, trips, operator actions, vibration and temperature histories, and anything from a permanently installed monitoring system. Save it somewhere outside the historian, and note the tag names and units. This costs an hour of somebody’s time today and cannot be recovered at any price later.
Capture the history while people remember
- What happened, and exactly when — with the timeline of alarms, trips and observations.
- Operating conditions before the event: loads, temperatures, pressures, chemistry, recent excursions.
- Maintenance history: last inspection, recent work on or near the component.
- Any earlier failures of the same or similar parts.
- Samples of any debris, deposits, scale, sludge or other loose material found at or near the failure — collected separately, sealed in labelled bags, and recorded with where each one came from. Loose material is often the only surviving record of the process environment, and it is the first thing swept up.
Memories fade and shift within days; a rough note written today outweighs a polished recollection next month.
Find the paperwork
Documents go missing faster than parts, because nobody thinks of them as evidence. Recover the material certificate, the heat or cast number, the mill test report, the purchase order and the goods-inwards record. Add the drawing, the specification the part was bought to, the welding procedure and the qualification records, any heat-treatment certificate, and the installation and commissioning records.
This is what allows us to say whether the component that failed was actually the component that was specified. A material substitution, a missed heat treatment or an unqualified weld procedure is a common root cause, and it is provable only on paper. Where the supply chain runs through a distributor, ask for the traceability back to the mill before anyone has a reason to be unhelpful about it.
Record who has had it
From the moment the part is recovered, keep a simple written record: who took it, when, from where, who has held it since, where it has been stored, and who has had access to it. A page in a notebook is enough. Each transfer gets a date and a name.
If the failure never becomes contentious, you have lost ten minutes. If it does, that record is the difference between evidence that is relied on and evidence that is challenged — because the first question asked of any physical exhibit is whether anyone could have altered it. It is also a requirement of any investigation intended to support a claim, and it cannot be created retrospectively with any credibility.
Do not start testing yet
There is a strong temptation to send the part somewhere and get numbers back. Resist it until the sequence has been decided, because testing is not reversible and the order matters.
Everything non-destructive comes first: visual and macro examination, photography, dimensional checks, surface replication, non-destructive testing, and analysis of deposits while they are still in place. Sectioning, hardness traversing, metallography and mechanical testing all consume the evidence, and a sample cut in the wrong plane or from the wrong location cannot be uncut. A hardness indent in the wrong place destroys the surface a fractographer needed.
If a dispute is possible this matters more still: the other side is entitled to be told before destructive work is carried out, and testing done without that notice can be excluded no matter how good the result. See forensic investigation for how that is handled.
If you must cut
Sometimes a section must be cut out to release the line or free the equipment. If so: cut far from the failure, never through or near a fracture or crack; mark the orientation and position of every cut piece; and photograph before and after cutting. Tell the investigator what was cut and how — heat from flame cutting alters nearby microstructure, and knowing where that influence ends matters.
When to contact MTIS
Ideally: before anything is cleaned, cut or moved. A short call at that moment lets us tell you exactly what to keep for your specific failure — it costs nothing and often saves the investigation. If the part is already cleaned or partially lost, contact us anyway: bounded answers from imperfect evidence are our normal work, and we will tell you honestly what can still be established. Start a job request → or see how a failure investigation works →.