The instinct after a rotating machine fails is to get it open, find the broken part and fit a spare. That instinct is correct for production and costly for evidence. Much of what a pump or compressor records about its own failure resides in no single component. It resides in relationships — where the rotor sat axially, how much float remained, which clearance closed first, which direction a rub ran, what was lying in the casing. Those relationships exist only while the machine is assembled, and teardown, cleaning and rebuild erase them in that order.
A failure is a sequence, and sequence lives in position
The component that broke most conspicuously is frequently not the one that failed first. Establishing order requires reading the machine as an assembly: the rotor's axial and radial position as found, the phase relationship between rub marks on rotating and stationary parts, which labyrinth closed and in what direction.
None of that survives a rotor pull unless captured first. Match-marking halves, indexing the rotor before anything is loosened, and photographing each interface before it separates costs an hour and preserves the sequence.
As-found alignment ends when the coupling is broken
Shaft alignment can only be measured with the machine as found and the coupling intact. Once the coupling is split, any later figure describes a new condition, not the one that failed. Misalignment is among the most commonly alleged contributors in these matters and the most easily rendered unprovable.
Pipe strain is the same. Loosening the suction and discharge flanges and measuring how far the machine moves is a one-time test that separates a piping problem from a machine problem. Soft foot, the shim stack and the state of the grout are equally perishable. API RP 686 sets out the installation practices these readings are compared against.
Cleaning removes the record
Deposits, coke, polymer and product residue are not dirt in this context; they are evidence of what the machine handled and how hot it became. Heat tint records the temperature a rub reached. Solvent-washing a seal face removes the film identifying the fluid it saw, and bead-blasting an impeller destroys pit morphology outright.
Strainer contents, the debris field in the casing and the oil in the reservoir are samples rather than waste. Oil analysis for babbitt, water and wear metals is often the only surviving record of when a bearing began shedding material.
Seals are consumables and evidence at once
A mechanical seal carries a legible history. Heat checking and blistering point at loss of the fluid film; an uneven wear band around the face circumference points at misalignment or distortion; coking, chipped outside diameters and elastomer swelling each narrow the field. API 682 defines the arrangements and support systems these observations are read against.
The trap here is commercial rather than technical. Failed cartridge seals routinely go back to the supplier on a core exchange for credit, removing the most informative component within days, usually before anyone has decided there is a case.
Bearings, thrust and the direction of the load
Within the machine's failure chain, the thrust bearing is the most articulate part. Which pads are damaged, active face or inactive, records the direction the rotor was pushed, and a thrust excursion is the signature of a transient rather than gradual wear. That depends entirely on knowing which pad came from which position: a bag of loose pads is a fraction as useful as the same pads tagged by position and orientation.
Clearances can be measured once
Wear-ring, throat bush, interstage and labyrinth clearances, balance-drum condition and rotor float are as-found values. Once the rotor is pulled and components are machined or scrapped, the numbers cannot be recovered from anything. Recorded against OEM tolerance they show how far the machine had drifted; unrecorded they become a matter of recollection.
The same applies to the rest of the train. Gearbox backlash and contact pattern, coupling condition and the state of any spacer belong to the story even when the failure is inside the pump.
The data expires sooner than the hardware
Process historians compress and eventually roll off. Machinery protection systems built to API 670 hold transient and startup captures in buffers of finite depth, and those buffers overwrite on the next event. Vendor archives, alarm logs and operator shift records have their own retention windows, most shorter than the interval between a failure and the decision to investigate it.
A written preservation request covering historian tags at native resolution, vibration waveform and orbit data, event logs and maintenance records is worth issuing before the machine is opened.
Who is present when it comes apart
Where more than one party may be responsible — an OEM, a repair shop, an installer, an operator, an insurer — the teardown is best conducted under an agreed protocol: notice to interested parties, an agreed sequence and photography, no destructive step without consent, and chain of custody for removed components.
The repair shop is frequently one of those parties. Its strip report is useful but is no substitute for independent observation, because the shop is describing work it may later have to defend.
When the machine has to go back in anyway
Production pressure is real and rarely negotiable. The useful point is that the steps that matter most are cheap and fit inside a single shift: photograph as found, measure alignment and pipe strain before uncoupling, bag and tag by position, take oil and debris samples, and freeze the data.
Returning the plant to service and preserving the answer are not in genuine conflict. They only appear to be when preservation is attempted afterwards, by which point most of it is no longer available at any price.
This article is general technical orientation, not a failure analysis, an engineering opinion, or advice on any specific matter. Determining the cause of a particular incident requires hands-on examination by a credentialed expert.