When a pump or compressor fails long before it should have, four explanations compete. The machine was the wrong choice for the duty. It was installed into a system that loaded it in ways the datasheet never contemplated. It was run outside the envelope it was sold for. Or it was not maintained. These are not merely technical alternatives; they map onto different parties, contracts and insurance towers. Each leaves a distinct documentary trace, and those traces decide the question more often than the hardware does.
The duty point is the first document
The datasheet states what the machine was bought to do: flow, head, fluid properties, temperature, suction conditions, materials. What matters is how that compares with what the machine actually saw. Rated, normal and best efficiency flows are three different numbers, and a machine sitting far from the third is under stress the purchase order never described.
Conservatism in specification is a recurring contributor. Margins stacked by successive reviewers produce a machine that must then be throttled back to a fraction of its best efficiency point, where recirculation, high radial load and shortened seal and bearing life follow predictably. API 610 and ISO 13709 for centrifugal pumps, API 674 and API 676 for positive-displacement machines and API 617 for compressors define what the specification was meant to capture.
Service drifts and machines do not
Plants change. Feedstock changes, capacity is debottlenecked, an impeller is trimmed, a variable-frequency drive is retrofitted. Each alters the hydraulics the machine sees, and each is supposed to leave a management-of-change record showing whether anyone rechecked them. Where that record is absent, its absence is itself a finding, and easier to establish than an argument about the machine's inherent suitability.
The piping the machine did not choose
Suction piping is an installation decision with hydraulic consequences: available straight run, eccentric reducers oriented to avoid trapping vapour, elbows close-coupled to the nozzle, vortexing at the source vessel, machines sharing a header. ANSI/HI 9.6.6 addresses pump piping specifically, and departures from it show on drawings long afterwards.
Nozzle loads are the other half. API 610 sets allowable forces and moments at the pump nozzles precisely because piping that pulls on a casing distorts it, closes internal clearances and redistributes bearing load. Grouting, foundation stiffness and structural natural frequencies belong to the same category.
Seal support systems are part of the installation
A mechanical seal is only as good as the environment supplied to it. API 682 defines the piping plans that flush, quench, cool or pressurise a seal, and a seal that lost its flush or sat behind a barrier fluid system at the wrong pressure has failed for an installation or maintenance reason rather than a manufacturing one. The evidence is unusually accessible: plan drawings, reservoir level and pressure trends, and cooler cleaning records.
Operation, and what the envelope actually was
The operating envelope is a set of specific limits, not a general sense of care. Minimum continuous stable flow and whether the recycle valve did its job. Suction margin at the temperature that actually applied. Start-stop cycling. Parallel operation in which one machine backs off against the other.
For compressors the anti-surge system is the centre of the question: where the surge control line sat relative to the surge limit, how the controller was tuned, whether it was in manual, how quickly the recycle valve could stroke. API 617 frames the envelope and API 670 the protection system meant to keep the machine inside it.
What the historian records and what it does not
Process historians are configured for trending, not forensics. Scan rates, deadbands and compression mean a smooth line may be an artefact of storage rather than a description of the process, and reconstructing a surge needs sub-second resolution a thirty-second scan will never contain.
Higher-resolution material usually exists elsewhere: sequence-of-events and alarm logs, trip records, and the protection system's own transient captures. Establishing which system holds what, at what resolution and for how long determines what any later opinion can be built on.
The vibration record and the standard of care
ISO 20816 provides the framework for measuring and evaluating machine vibration and the zones that alarm and trip settings are usually derived from. A rising trend over months does more than identify a developing fault: it establishes how long the condition existed and what opportunities to intervene existed.
That converts a mechanical question into a decision question. Was the alarm acknowledged, was the setpoint raised to stop it recurring, was the channel bypassed, was route-based collection completed? Those answers sit in the monitoring system's configuration history and the maintenance management system, and are frequently the most consequential documents in the file.
Repairs, parts and what was actually fitted
A machine overhauled several times is partly a different machine. Repair-shop records show as-found readings, what was replaced, whether components were original or aftermarket, what clearances were set, and to what grade the rotor was balanced, for which ISO 21940-11 supplies the framework. This is where warranty and product-liability arguments live, and a defence built on non-original internals has to be met with records.
Weighting the candidates rather than picking one
Premature failures usually have more than one contributor: a marginal selection that would have survived good suction conditions, or an installation defect that only mattered once the machine was pushed off its curve. Treating the four as mutually exclusive overstates whichever the available evidence happens to favour.
The more durable approach states what each candidate would have to look like in the record, what the record shows, and which contributors were necessary as against sufficient. That reasoning survives cross-examination in a way a single cause named early does not.
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.